DBENGINE: pgc tuning, replication tuning (#19237)
* evict, a page at a time * 4 replication ahead requests per replication thread * added per job average timings for workers and dbengine query router * debug statement to find what is slow * yield the processor to avoid monopolizing the cache * test more page sizes in aral * more polite journal v2 indexing * pulse macros for atomics * added profile so that users can control the defaults of the agent * fix windows warnings; journal v2 generation yields the processor for every page * replication threads are 1/3 of the cores and they are synchronous * removed the default from the list of profiles * turn pgc locks into macros to have tracing on the functions that called them * log the size of madvise() when failing * more work on profiles * restore batch cache evictions, but lower the batch size significantly * do not spin while searching for pages in the cache - handle currently being deleted pages within the search logic itself * remove bottleneck in epdl processing while merging extents * allocate outside the lock * rw spinlock implemented without spinlocks; both spinlocks and r/w spinlocks now support exponential backoff while waiting * apply max sleep to spinlocks * tune replication * r/w spinlock prefers writers again, but this time recursive readers are bypassing the writer wait * tuning of rw spinlock * more tuning of the rw spinlock * configure glibc arenas based on profile * moving global variables into nd_profile * do not accept sockets that have not received any data; once sockets with data have been accepted, check they are not closed already before processing them * poll_events is now using nd_poll(), resulting in vast simplification of the code; static web files are now served inline resulting in another simplification of the web server logic (was required because epoll does not support normal files) * startup fixes * added debug info to poll_fd_close() * closed sockets are automatically removed from epoll(), by the kernel * fix for mrg acquired and referenced going negative * fixed bug in mrg cleanup, not deleting metrics that do not have retention * monitor strings index size * strings memory chart is now stacked * replication: do not lock data collection when running in batches * explicitly set socket flags for sender and receiver * normalize the event loop for sending data (receiver and sender) * normalize the event loop for receiving data (receiver and sender) * check all sender nodes every half a second * fix bug on sender, not enabling sending * first cleanup then destroy * normalize nd_poll() to handle all possible events * cleanup * normalize socket helper functions * fixed warnings on alpine * fix for POLLRDHUP missing * fix cleanup on shutdown * added detailed replication summary * moved logs to INFO * prevent crash when sender is not there * madvise _dontfork() should not be used with aral, madvise_dontdump() is only used for file backed maps * fix wording * fix log wording * split replication receiver and sender; add logs to find missing replication requests * fix compilation * fixed bug in backfilling, having garbage for counters - malloc instead of calloc * backfilling logs if it misses callbacks * log replication rcv and replication snd in node info * remove contention from aral_page_free_lock(), but having 2 free lists per page, one for incoming and another for available items and moving incoming to available when the available is empty - this allows aral_mallocz() and aral_freez() to operate concurrently on the same page * fix internal checks * log errors for all replication receiver exceptions * removed wrong error log * prevent health crashing * cleanup logs that are irrelevant with the missing replication events * replication tracking: added replication tracking to figure out how replication missed requests * fix compilation and fix bug on spawn server cleanup calling uv_shutdown at exit * merged receiver initialization * prevent compilation warnings * fix race condition in nd_poll() returning events for deleted fds * for user queries, prepare as many queries as half the processors * fix log * add option dont_dump to netdata_mmap and aral_create * add logging missing receiver and sender charts * reviewed judy memory accounting; adbstracted flags handling to ensure they all work the same way; introduced atomic_flags_set_and_clear() to set and clear atomic flags with a single atomic operation * improvement(go.d/nats): add server_id label (#19280) * Regenerate integrations docs (#19281) Co-authored-by: ilyam8 <22274335+ilyam8@users.noreply.github.com> * [ci skip] Update changelog and version for nightly build: v2.1.0-30-nightly. * docs: improve on-prem troubleshooting readability (#19279) * docs: improve on-prem troubleshooting readability * Apply suggestions from code review --------- Co-authored-by: Fotis Voutsas <fotis@netdata.cloud> * improvement(go.d/nats): add leafz metrics (#19282) * Regenerate integrations docs (#19283) Co-authored-by: ilyam8 <22274335+ilyam8@users.noreply.github.com> * [ci skip] Update changelog and version for nightly build: v2.1.0-34-nightly. * fix go.d/nats tests (#19284) * improvement(go.d/nats): add basic jetstream metrics (#19285) * Regenerate integrations docs (#19286) Co-authored-by: ilyam8 <22274335+ilyam8@users.noreply.github.com> * [ci skip] Update changelog and version for nightly build: v2.1.0-38-nightly. * bump dag req jinja version (#19287) * more strict control on replication counters * do not flush the log files - to cope with the rate * [ci skip] Update changelog and version for nightly build: v2.1.0-40-nightly. * fix aral on windows * add waiting queue to sender commit, to allow the streaming thread go fast and put replication threads in order * use the receiver tid * fix(netdata-updater.sh): remove commit_check_file directory (#19288) * receiver now has periodic checks too (like the senders have) * fixed logs * replication periodic checks: resending of chart definitions * strict checking on rrdhost state id * replication periodic checks: added for receivers * shorter replication status messages * do not log about ieee754 * receiver logs replication traffic without RSET * object state: rrdhost_state_id has become object_state in libnetdata so that it can be reused * fixed metadata; added journal message id for netdata fatal messages * replication: undo bypassing the pipeline * receiver cleanup: free all structures at the end, to ensure there are not crashes while cleaning up * replication periodic checks: do not run it on receivers, when there is replication in progress * nd_log: prevent fatal statements from recursing * replication tracking: disabled (compile time) * fix priority and log * disconnect on stale replication - detected on both sender and receiver * update our tagline * when sending data from within opcode handling do not remove the receiver/sender * improve interactivity of streaming sockets * log the replication cmd counters on disconnect and reset them on reconnect * rrdhost object state activate/deactivate should happen in set/clear receiver * remove writer preference from rw spinlocks * show the value in health logs * move counter to the right place to avoid double counting replication commands * do not run opcodes when running inline * fix replication log messages * make IoT harmless for the moment --------- Co-authored-by: Ilya Mashchenko <ilya@netdata.cloud> Co-authored-by: Netdata bot <43409846+netdatabot@users.noreply.github.com> Co-authored-by: ilyam8 <22274335+ilyam8@users.noreply.github.com> Co-authored-by: netdatabot <bot@netdata.cloud> Co-authored-by: Fotis Voutsas <fotis@netdata.cloud>
Costa Tsaousis committed
Dec 29, 2024 at 18:22 UTC
bc840a79943aeb7f570f1a21d9e0792e902ffc5b
170 files changed
+5201
-3381
CMakeLists.txt
+14
-4
@@ -946,6 +946,12 @@ set(LIBNETDATA_FILES
946
src/libnetdata/locks/spinlock.h
947
src/libnetdata/locks/rw-spinlock.c
948
src/libnetdata/locks/rw-spinlock.h
949
+ src/libnetdata/atomics/atomic_flags.h
950
+ src/libnetdata/atomics/atomics.h
951
+ src/libnetdata/waiting-queue/waiting-queue.c
952
+ src/libnetdata/waiting-queue/waiting-queue.h
953
+ src/libnetdata/object-state/object-state.c
954
+ src/libnetdata/object-state/object-state.h
955
)
956
957
set(LIBH2O_FILES
@@ -1122,6 +1128,8 @@ set(DAEMON_FILES
1128
src/daemon/pulse/pulse-db-rrd.h
1129
src/daemon/config/netdata-conf-cloud.c
1130
src/daemon/config/netdata-conf-cloud.h
1131
+ src/daemon/config/netdata-conf-profile.c
1132
+ src/daemon/config/netdata-conf-profile.h
1133
)
1134
1135
set(H2O_FILES
@@ -1439,8 +1447,6 @@ set(RRD_PLUGIN_FILES
1447
src/database/rrdcollector-internals.h
1448
src/database/rrd-database-mode.h
1449
src/database/rrd-database-mode.c
1442
- src/database/rrdhost-state-id.c
1443
- src/database/rrdhost-state-id.h
1450
src/database/rrdhost-system-info.c
1451
src/database/rrdhost-system-info.h
1452
)
@@ -1523,8 +1529,8 @@ set(STREAMING_PLUGIN_FILES
1529
src/streaming/stream-compression/zstd.h
1530
src/streaming/stream-receiver.c
1531
src/streaming/stream-sender.c
1526
- src/streaming/replication.c
1527
- src/streaming/replication.h
1532
+ src/streaming/stream-replication-sender.c
1533
+ src/streaming/stream-replication-sender.h
1534
src/streaming/h2o-common.h
1535
src/streaming/protocol/command-nodeid.c
1536
src/streaming/protocol/commands.c
@@ -1565,6 +1571,10 @@ set(STREAMING_PLUGIN_FILES
1571
src/streaming/stream-control.h
1572
src/streaming/stream-waiting-list.c
1573
src/streaming/stream-waiting-list.h
1574
+ src/streaming/stream-replication-receiver.c
1575
+ src/streaming/stream-replication-receiver.h
1576
+ src/streaming/stream-replication-tracking.c
1577
+ src/streaming/stream-replication-tracking.h
1578
)
1579
1580
set(WEB_PLUGIN_FILES
docs/diagrams/data_structures/web.svg
+1
-1
@@ -1,2 +1,2 @@
1
<!DOCTYPE svg PUBLIC "-//W3C//DTD SVG 1.1//EN" "http://www.w3.org/Graphics/SVG/1.1/DTD/svg11.dtd">
2
-<svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" version="1.1" width="1805px" height="765px" viewBox="-0.5 -0.5 1805 765" style="background-color: rgb(255, 255, 255);"><defs><clipPath id="mx-clip-7-61-322-26-0"><rect x="7" y="61" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-87-322-26-0"><rect x="7" y="87" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-113-322-26-0"><rect x="7" y="113" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-139-322-26-0"><rect x="7" y="139" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-165-322-26-0"><rect x="7" y="165" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-191-322-26-0"><rect x="7" y="191" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-217-322-26-0"><rect x="7" y="217" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-243-322-26-0"><rect x="7" y="243" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-269-322-26-0"><rect x="7" y="269" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-295-322-26-0"><rect x="7" y="295" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-321-322-26-0"><rect x="7" y="321" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-347-322-26-0"><rect x="7" y="347" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-373-322-26-0"><rect x="7" y="373" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-399-322-26-0"><rect x="7" y="399" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-425-322-26-0"><rect x="7" y="425" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-451-322-26-0"><rect x="7" y="451" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-477-322-26-0"><rect x="7" y="477" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-503-322-26-0"><rect x="7" y="503" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-529-322-26-0"><rect x="7" y="529" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-555-322-26-0"><rect x="7" y="555" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-581-322-26-0"><rect x="7" y="581" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-607-322-26-0"><rect x="7" y="607" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-633-322-26-0"><rect x="7" y="633" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-659-322-26-0"><rect x="7" y="659" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-685-322-26-0"><rect x="7" y="685" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-711-322-26-0"><rect x="7" y="711" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-737-322-26-0"><rect x="7" y="737" width="322" height="26"/></clipPath><clipPath id="mx-clip-477-165-252-26-0"><rect x="477" y="165" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-191-252-26-0"><rect x="477" y="191" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-217-252-26-0"><rect x="477" y="217" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-243-252-26-0"><rect x="477" y="243" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-269-252-26-0"><rect x="477" y="269" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-295-252-26-0"><rect x="477" y="295" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-321-252-26-0"><rect x="477" y="321" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-347-252-26-0"><rect x="477" y="347" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-373-252-26-0"><rect x="477" y="373" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-399-252-26-0"><rect x="477" y="399" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-425-252-26-0"><rect x="477" y="425" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-451-252-26-0"><rect x="477" y="451" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-577-252-26-0"><rect x="477" y="577" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-603-252-26-0"><rect x="477" y="603" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-629-252-26-0"><rect x="477" y="629" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-655-252-26-0"><rect x="477" y="655" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-681-252-26-0"><rect x="477" y="681" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-707-252-26-0"><rect x="477" y="707" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-733-252-26-0"><rect x="477" y="733" width="252" height="26"/></clipPath><clipPath id="mx-clip-837-113-232-26-0"><rect x="837" y="113" width="232" height="26"/></clipPath><clipPath id="mx-clip-837-139-232-26-0"><rect x="837" y="139" width="232" height="26"/></clipPath><clipPath id="mx-clip-837-165-232-26-0"><rect x="837" y="165" width="232" height="26"/></clipPath><clipPath id="mx-clip-837-191-232-26-0"><rect x="837" y="191" width="232" height="26"/></clipPath><clipPath id="mx-clip-837-217-232-26-0"><rect x="837" y="217" width="232" height="26"/></clipPath><clipPath id="mx-clip-837-243-232-26-0"><rect x="837" y="243" width="232" height="26"/></clipPath><clipPath id="mx-clip-837-269-232-26-0"><rect x="837" y="269" width="232" height="26"/></clipPath><clipPath id="mx-clip-837-295-232-26-0"><rect x="837" y="295" width="232" height="26"/></clipPath><clipPath id="mx-clip-837-321-232-26-0"><rect x="837" y="321" width="232" height="26"/></clipPath><clipPath id="mx-clip-837-347-232-26-0"><rect x="837" y="347" width="232" height="26"/></clipPath><clipPath id="mx-clip-837-373-232-26-0"><rect x="837" y="373" width="232" height="26"/></clipPath><clipPath id="mx-clip-1177-61-222-26-0"><rect x="1177" y="61" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-87-222-26-0"><rect x="1177" y="87" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-113-222-26-0"><rect x="1177" y="113" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-139-222-26-0"><rect x="1177" y="139" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-165-222-26-0"><rect x="1177" y="165" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-191-222-26-0"><rect x="1177" y="191" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-217-222-26-0"><rect x="1177" y="217" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-243-222-26-0"><rect x="1177" y="243" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-269-222-26-0"><rect x="1177" y="269" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-295-222-26-0"><rect x="1177" y="295" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-321-222-26-0"><rect x="1177" y="321" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-347-222-26-0"><rect x="1177" y="347" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-373-222-26-0"><rect x="1177" y="373" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-399-222-26-0"><rect x="1177" y="399" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-425-222-26-0"><rect x="1177" y="425" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-451-222-26-0"><rect x="1177" y="451" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-477-222-26-0"><rect x="1177" y="477" width="222" height="26"/></clipPath><clipPath id="mx-clip-1527-126-242-26-0"><rect x="1527" y="126" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-152-242-26-0"><rect x="1527" y="152" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-178-242-26-0"><rect x="1527" y="178" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-204-242-26-0"><rect x="1527" y="204" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-230-242-26-0"><rect x="1527" y="230" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-256-242-26-0"><rect x="1527" y="256" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-282-242-26-0"><rect x="1527" y="282" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-308-242-26-0"><rect x="1527" y="308" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-334-242-26-0"><rect x="1527" y="334" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-360-242-26-0"><rect x="1527" y="360" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-386-242-26-0"><rect x="1527" y="386" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-412-242-26-0"><rect x="1527" y="412" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-438-242-26-0"><rect x="1527" y="438" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-464-242-26-0"><rect x="1527" y="464" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-490-242-26-0"><rect x="1527" y="490" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-516-242-26-0"><rect x="1527" y="516" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-542-242-26-0"><rect x="1527" y="542" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-568-242-26-0"><rect x="1527" y="568" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-594-242-26-0"><rect x="1527" y="594" width="242" height="26"/></clipPath></defs><path d="M 3 56 L 3 30 L 333 30 L 333 56" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 3 56 L 3 758 L 333 758 L 333 56" fill="#ffffff" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 3 56 L 333 56" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><g fill="#000000" font-family="Helvetica" text-anchor="middle" font-size="12px"><text x="167.5" y="47.5">web_client</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-61-322-26-0)" font-size="12px"><text x="8.5" y="73.5">unsigned long long id</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-87-322-26-0)" font-size="12px"><text x="8.5" y="99.5">WEB_CLIENT_FLAGS flags</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-113-322-26-0)" font-size="12px"><text x="8.5" y="125.5">WEB_CLIENT_MODE mode</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-139-322-26-0)" font-size="12px"><text x="8.5" y="151.5">WEB_CLIENT_ACL acl</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-165-322-26-0)" font-size="12px"><text x="8.5" y="177.5">size_t header_parse_tries</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-191-322-26-0)" font-size="12px"><text x="8.5" y="203.5">size_t header_parse_last_size</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-217-322-26-0)" font-size="12px"><text x="8.5" y="229.5">int tcp_cork</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-243-322-26-0)" font-size="12px"><text x="8.5" y="255.5">int ifd</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-269-322-26-0)" font-size="12px"><text x="8.5" y="281.5">int ofd</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-295-322-26-0)" font-size="12px"><text x="8.5" y="307.5">char client_ip[NI_MAXHOST+1}</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-321-322-26-0)" font-size="12px"><text x="8.5" y="333.5">char client_port[NI_MAXSERV+1]</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-347-322-26-0)" font-size="12px"><text x="8.5" y="359.5">char decoded_url[NETDATA_WEB_REQUEST_URL_SIZE+1</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-373-322-26-0)" font-size="12px"><text x="8.5" y="385.5">char last_url[NETDATA_WEB_REQUEST_URL_SIZE+1]</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-399-322-26-0)" font-size="12px"><text x="8.5" y="411.5">struct timeval tv_in, tv_ready</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-425-322-26-0)" font-size="12px"><text x="8.5" y="437.5">char cookie1[NETDATA_WEB_REQUEST_COOKIE_SIZE+1]</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-451-322-26-0)" font-size="12px"><text x="8.5" y="463.5">char cookie2[NETDATA_WEB_REQUEST_COOKIE_SIZE+1]</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-477-322-26-0)" font-size="12px"><text x="8.5" y="489.5">char origin[NETDATA_WEB_REQUEST_ORIGIN_HEADER_SIZE+1]</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-503-322-26-0)" font-size="12px"><text x="8.5" y="515.5">char *user_agent</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-529-322-26-0)" font-size="12px"><text x="8.5" y="541.5">struct response response</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-555-322-26-0)" font-size="12px"><text x="8.5" y="567.5">size_t stats_received_bytes</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-581-322-26-0)" font-size="12px"><text x="8.5" y="593.5">size_t stats_sent_bytes</text></g><path d="M 333 615 L 353 615 L 353 10 L 168 10 L 168 23.63" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 168 28.88 L 164.5 21.88 L 168 23.63 L 171.5 21.88 Z" fill="#000000" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-607-322-26-0)" font-size="12px"><text x="8.5" y="619.5">struct web_client *prev</text></g><path d="M 333 641 L 353 641 L 353 10 L 168 10 L 168 23.63" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 168 28.88 L 164.5 21.88 L 168 23.63 L 171.5 21.88 Z" fill="#000000" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><g transform="translate(304.5,299.5)"><switch><foreignObject style="overflow:visible;" pointer-events="all" width="102" height="12" requiredFeatures="http://www.w3.org/TR/SVG11/feature#Extensibility"><div xmlns="http://www.w3.org/1999/xhtml" style="display: inline-block; font-size: 12px; font-family: Helvetica; color: rgb(0, 0, 0); line-height: 1.2; vertical-align: top; white-space: nowrap; text-align: center;"><div xmlns="http://www.w3.org/1999/xhtml" style="display:inline-block;text-align:inherit;text-decoration:inherit;background-color:#ffffff;">double linked list of</div></div></foreignObject><text x="51" y="12" fill="#000000" text-anchor="middle" font-size="12px" font-family="Helvetica">double linked list of</text></switch></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-633-322-26-0)" font-size="12px"><text x="8.5" y="645.5">struct web_client *next</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-659-322-26-0)" font-size="12px"><text x="8.5" y="671.5">netdata_thread_t thread</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-685-322-26-0)" font-size="12px"><text x="8.5" y="697.5">volatile int running</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-711-322-26-0)" font-size="12px"><text x="8.5" y="723.5">size_t pollinfo_slot</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-737-322-26-0)" font-size="12px"><text x="8.5" y="749.5">size_t pollinfo_filecopy_slot</text></g><path d="M 473 160 L 473 134 L 733 134 L 733 160" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 473 160 L 473 472 L 733 472 L 733 160" fill="#ffffff" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 473 160 L 733 160" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><g fill="#000000" font-family="Helvetica" text-anchor="middle" font-size="12px"><text x="602.5" y="151.5">response</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-165-252-26-0)" font-size="12px"><text x="478.5" y="177.5">BUFFER *header</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-191-252-26-0)" font-size="12px"><text x="478.5" y="203.5">BUFFER *header_output</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-217-252-26-0)" font-size="12px"><text x="478.5" y="229.5">BUFFER *data</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-243-252-26-0)" font-size="12px"><text x="478.5" y="255.5">int code</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-269-252-26-0)" font-size="12px"><text x="478.5" y="281.5">size_t rlen</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-295-252-26-0)" font-size="12px"><text x="478.5" y="307.5">size_t sent</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-321-252-26-0)" font-size="12px"><text x="478.5" y="333.5">int zoutput</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-347-252-26-0)" font-size="12px"><text x="478.5" y="359.5">z_stream zstream</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-373-252-26-0)" font-size="12px"><text x="478.5" y="385.5">Bytef zbuffer[NETDATA_WEB_RESPONSE_ZLIB_CHUNK_SIZE]</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-399-252-26-0)" font-size="12px"><text x="478.5" y="411.5">size_t zsent</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-425-252-26-0)" font-size="12px"><text x="478.5" y="437.5">size_t zhave</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-451-252-26-0)" font-size="12px"><text x="478.5" y="463.5">unsigned int zinitialized</text></g><path d="M 333 537 L 403 537 L 403 114 L 603 114 L 603 127.63" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 603 132.88 L 599.5 125.88 L 603 127.63 L 606.5 125.88 Z" fill="#000000" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 473 572 L 473 546 L 733 546 L 733 572" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 473 572 L 473 754 L 733 754 L 733 572" fill="#ffffff" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 473 572 L 733 572" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><g fill="#000000" font-family="Helvetica" text-anchor="middle" font-size="12px"><text x="602.5" y="563.5">clients_cache</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-577-252-26-0)" font-size="12px"><text x="478.5" y="589.5">pid_t pid</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-603-252-26-0)" font-size="12px"><text x="478.5" y="615.5">struct web_client *used</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-629-252-26-0)" font-size="12px"><text x="478.5" y="641.5">size_t used_count</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-655-252-26-0)" font-size="12px"><text x="478.5" y="667.5">struct web_client *avail</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-681-252-26-0)" font-size="12px"><text x="478.5" y="693.5">size_t avail_count</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-707-252-26-0)" font-size="12px"><text x="478.5" y="719.5">size_t reused</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-733-252-26-0)" font-size="12px"><text x="478.5" y="745.5">size_t allocated</text></g><path d="M 733 611 L 753 611 L 753 10 L 168 10 L 168 23.63" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 168 28.88 L 164.5 21.88 L 168 23.63 L 171.5 21.88 Z" fill="#000000" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><g transform="translate(724.5,12.5)"><switch><foreignObject style="overflow:visible;" pointer-events="all" width="57" height="11" requiredFeatures="http://www.w3.org/TR/SVG11/feature#Extensibility"><div xmlns="http://www.w3.org/1999/xhtml" style="display: inline-block; font-size: 11px; font-family: Helvetica; color: rgb(0, 0, 0); line-height: 1.2; vertical-align: top; white-space: nowrap; text-align: center;"><div xmlns="http://www.w3.org/1999/xhtml" style="display:inline-block;text-align:inherit;text-decoration:inherit;background-color:#ffffff;">linked list of</div></div></foreignObject><text x="29" y="11" fill="#000000" text-anchor="middle" font-size="11px" font-family="Helvetica">linked list of</text></switch></g><path d="M 733 663 L 753 663 L 753 10 L 168 10 L 168 23.63" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 168 28.88 L 164.5 21.88 L 168 23.63 L 171.5 21.88 Z" fill="#000000" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><g transform="translate(724.5,38.5)"><switch><foreignObject style="overflow:visible;" pointer-events="all" width="57" height="11" requiredFeatures="http://www.w3.org/TR/SVG11/feature#Extensibility"><div xmlns="http://www.w3.org/1999/xhtml" style="display: inline-block; font-size: 11px; font-family: Helvetica; color: rgb(0, 0, 0); line-height: 1.2; vertical-align: top; white-space: nowrap; text-align: center;"><div xmlns="http://www.w3.org/1999/xhtml" style="display:inline-block;text-align:inherit;text-decoration:inherit;background-color:#ffffff;">linked list of</div></div></foreignObject><text x="29" y="11" fill="#000000" text-anchor="middle" font-size="11px" font-family="Helvetica">linked list of</text></switch></g><path d="M 833 108 L 833 82 L 1073 82 L 1073 108" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 833 108 L 833 394 L 1073 394 L 1073 108" fill="#ffffff" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 833 108 L 1073 108" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><g fill="#000000" font-family="Helvetica" text-anchor="middle" font-size="12px"><text x="952.5" y="99.5">listen_sockets</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-837-113-232-26-0)" font-size="12px"><text x="838.5" y="125.5">struct config *config</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-837-139-232-26-0)" font-size="12px"><text x="838.5" y="151.5">const char *config_section</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-837-165-232-26-0)" font-size="12px"><text x="838.5" y="177.5">const char *default_bind_to</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-837-191-232-26-0)" font-size="12px"><text x="838.5" y="203.5">uint16_t default_port</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-837-217-232-26-0)" font-size="12px"><text x="838.5" y="229.5">int backlog</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-837-243-232-26-0)" font-size="12px"><text x="838.5" y="255.5">size_t opened</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-837-269-232-26-0)" font-size="12px"><text x="838.5" y="281.5">size_t failed</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-837-295-232-26-0)" font-size="12px"><text x="838.5" y="307.5">int fds[MAX_LISTEN_FDS]</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-837-321-232-26-0)" font-size="12px"><text x="838.5" y="333.5">int *fds_names[MAX_LISTEN_FDS]</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-837-347-232-26-0)" font-size="12px"><text x="838.5" y="359.5">int fds_types[MAX_LISTEN_FDS]</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-837-373-232-26-0)" font-size="12px"><text x="838.5" y="385.5">int fds_families[MAX_LISTEN_FDS]</text></g><path d="M 1173 56 L 1173 30 L 1403 30 L 1403 56" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 1173 56 L 1173 498 L 1403 498 L 1403 56" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 1173 56 L 1403 56" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><g fill="#000000" font-family="Helvetica" text-anchor="middle" font-size="12px"><text x="1287.5" y="47.5">POLLINFO</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-61-222-26-0)" font-size="12px"><text x="1178.5" y="73.5">POLLJOB *p</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-87-222-26-0)" font-size="12px"><text x="1178.5" y="99.5">size_t slot</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-113-222-26-0)" font-size="12px"><text x="1178.5" y="125.5">int fd</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-139-222-26-0)" font-size="12px"><text x="1178.5" y="151.5">int socktype</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-165-222-26-0)" font-size="12px"><text x="1178.5" y="177.5">char *client_ip</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-191-222-26-0)" font-size="12px"><text x="1178.5" y="203.5">char *client_port</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-217-222-26-0)" font-size="12px"><text x="1178.5" y="229.5">time_t connected_t</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-243-222-26-0)" font-size="12px"><text x="1178.5" y="255.5">time_t last_received_t</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-269-222-26-0)" font-size="12px"><text x="1178.5" y="281.5">time_t last_sent_t</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-295-222-26-0)" font-size="12px"><text x="1178.5" y="307.5">size_t recv_count</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-321-222-26-0)" font-size="12px"><text x="1178.5" y="333.5">size_t send_count</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-347-222-26-0)" font-size="12px"><text x="1178.5" y="359.5">uint32_t flags</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-373-222-26-0)" font-size="12px"><text x="1178.5" y="385.5">void (*del_callback)</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-399-222-26-0)" font-size="12px"><text x="1178.5" y="411.5">int (*rcv_callback)</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-425-222-26-0)" font-size="12px"><text x="1178.5" y="437.5">int (*snd_callback)</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-451-222-26-0)" font-size="12px"><text x="1178.5" y="463.5">void *data</text></g><path d="M 1403 485 L 1423 485 L 1423 10 L 1279 10 L 1279 22.63" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 1279 27.88 L 1275.5 20.88 L 1279 22.63 L 1282.5 20.88 Z" fill="#000000" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-477-222-26-0)" font-size="12px"><text x="1178.5" y="489.5">struct pollinfo *next</text></g><path d="M 1523 121 L 1523 95 L 1773 95 L 1773 121" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 1523 121 L 1523 615 L 1773 615 L 1773 121" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 1523 121 L 1773 121" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><g fill="#000000" font-family="Helvetica" text-anchor="middle" font-size="12px"><text x="1647.5" y="112.5">POLLJOB</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-126-242-26-0)" font-size="12px"><text x="1528.5" y="138.5">size_t slots</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-152-242-26-0)" font-size="12px"><text x="1528.5" y="164.5">size_t used</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-178-242-26-0)" font-size="12px"><text x="1528.5" y="190.5">size_t min</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-204-242-26-0)" font-size="12px"><text x="1528.5" y="216.5">size_t max</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-230-242-26-0)" font-size="12px"><text x="1528.5" y="242.5">size_t limit</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-256-242-26-0)" font-size="12px"><text x="1528.5" y="268.5">time_t complete_request_timeout</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-282-242-26-0)" font-size="12px"><text x="1528.5" y="294.5">time_t idle_timeout</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-308-242-26-0)" font-size="12px"><text x="1528.5" y="320.5">time_t check_every</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-334-242-26-0)" font-size="12px"><text x="1528.5" y="346.5">time_t timer_milliseconds</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-360-242-26-0)" font-size="12px"><text x="1528.5" y="372.5">void *timer_data</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-386-242-26-0)" font-size="12px"><text x="1528.5" y="398.5">struct run *fds</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-412-242-26-0)" font-size="12px"><text x="1528.5" y="424.5">struct pollinfo *inf</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-438-242-26-0)" font-size="12px"><text x="1528.5" y="450.5">struct pollinfo *first_free</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-464-242-26-0)" font-size="12px"><text x="1528.5" y="476.5">SIMPLE_PATTERN *access_list</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-490-242-26-0)" font-size="12px"><text x="1528.5" y="502.5">void *(*add_callback)</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-516-242-26-0)" font-size="12px"><text x="1528.5" y="528.5">void (*dell_callback)</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-542-242-26-0)" font-size="12px"><text x="1528.5" y="554.5">int (*rcv_callback)</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-568-242-26-0)" font-size="12px"><text x="1528.5" y="580.5">int (*snd_callback)</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-594-242-26-0)" font-size="12px"><text x="1528.5" y="606.5">void (*tmr_callback)</text></g><path d="M 1403 69 L 1648 69 L 1648 88.63" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 1648 93.88 L 1644.5 86.88 L 1648 88.63 L 1651.5 86.88 Z" fill="#000000" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 1773 420 L 1793 420 L 1793 10 L 1279 10 L 1279 22.63" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 1279 27.88 L 1275.5 20.88 L 1279 22.63 L 1282.5 20.88 Z" fill="#000000" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 1773 446 L 1793 446 L 1793 10 L 1278 10 L 1278 22.63" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 1278 27.88 L 1274.5 20.88 L 1278 22.63 L 1281.5 20.88 Z" fill="#000000" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/></svg>
\ No newline at end of file
2
+<svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" version="1.1" width="1805px" height="765px" viewBox="-0.5 -0.5 1805 765" style="background-color: rgb(255, 255, 255);"><defs><clipPath id="mx-clip-7-61-322-26-0"><rect x="7" y="61" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-87-322-26-0"><rect x="7" y="87" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-113-322-26-0"><rect x="7" y="113" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-139-322-26-0"><rect x="7" y="139" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-165-322-26-0"><rect x="7" y="165" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-191-322-26-0"><rect x="7" y="191" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-217-322-26-0"><rect x="7" y="217" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-243-322-26-0"><rect x="7" y="243" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-269-322-26-0"><rect x="7" y="269" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-295-322-26-0"><rect x="7" y="295" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-321-322-26-0"><rect x="7" y="321" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-347-322-26-0"><rect x="7" y="347" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-373-322-26-0"><rect x="7" y="373" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-399-322-26-0"><rect x="7" y="399" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-425-322-26-0"><rect x="7" y="425" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-451-322-26-0"><rect x="7" y="451" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-477-322-26-0"><rect x="7" y="477" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-503-322-26-0"><rect x="7" y="503" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-529-322-26-0"><rect x="7" y="529" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-555-322-26-0"><rect x="7" y="555" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-581-322-26-0"><rect x="7" y="581" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-607-322-26-0"><rect x="7" y="607" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-633-322-26-0"><rect x="7" y="633" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-659-322-26-0"><rect x="7" y="659" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-685-322-26-0"><rect x="7" y="685" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-711-322-26-0"><rect x="7" y="711" width="322" height="26"/></clipPath><clipPath id="mx-clip-7-737-322-26-0"><rect x="7" y="737" width="322" height="26"/></clipPath><clipPath id="mx-clip-477-165-252-26-0"><rect x="477" y="165" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-191-252-26-0"><rect x="477" y="191" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-217-252-26-0"><rect x="477" y="217" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-243-252-26-0"><rect x="477" y="243" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-269-252-26-0"><rect x="477" y="269" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-295-252-26-0"><rect x="477" y="295" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-321-252-26-0"><rect x="477" y="321" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-347-252-26-0"><rect x="477" y="347" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-373-252-26-0"><rect x="477" y="373" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-399-252-26-0"><rect x="477" y="399" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-425-252-26-0"><rect x="477" y="425" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-451-252-26-0"><rect x="477" y="451" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-577-252-26-0"><rect x="477" y="577" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-603-252-26-0"><rect x="477" y="603" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-629-252-26-0"><rect x="477" y="629" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-655-252-26-0"><rect x="477" y="655" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-681-252-26-0"><rect x="477" y="681" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-707-252-26-0"><rect x="477" y="707" width="252" height="26"/></clipPath><clipPath id="mx-clip-477-733-252-26-0"><rect x="477" y="733" width="252" height="26"/></clipPath><clipPath id="mx-clip-837-113-232-26-0"><rect x="837" y="113" width="232" height="26"/></clipPath><clipPath id="mx-clip-837-139-232-26-0"><rect x="837" y="139" width="232" height="26"/></clipPath><clipPath id="mx-clip-837-165-232-26-0"><rect x="837" y="165" width="232" height="26"/></clipPath><clipPath id="mx-clip-837-191-232-26-0"><rect x="837" y="191" width="232" height="26"/></clipPath><clipPath id="mx-clip-837-217-232-26-0"><rect x="837" y="217" width="232" height="26"/></clipPath><clipPath id="mx-clip-837-243-232-26-0"><rect x="837" y="243" width="232" height="26"/></clipPath><clipPath id="mx-clip-837-269-232-26-0"><rect x="837" y="269" width="232" height="26"/></clipPath><clipPath id="mx-clip-837-295-232-26-0"><rect x="837" y="295" width="232" height="26"/></clipPath><clipPath id="mx-clip-837-321-232-26-0"><rect x="837" y="321" width="232" height="26"/></clipPath><clipPath id="mx-clip-837-347-232-26-0"><rect x="837" y="347" width="232" height="26"/></clipPath><clipPath id="mx-clip-837-373-232-26-0"><rect x="837" y="373" width="232" height="26"/></clipPath><clipPath id="mx-clip-1177-61-222-26-0"><rect x="1177" y="61" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-87-222-26-0"><rect x="1177" y="87" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-113-222-26-0"><rect x="1177" y="113" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-139-222-26-0"><rect x="1177" y="139" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-165-222-26-0"><rect x="1177" y="165" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-191-222-26-0"><rect x="1177" y="191" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-217-222-26-0"><rect x="1177" y="217" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-243-222-26-0"><rect x="1177" y="243" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-269-222-26-0"><rect x="1177" y="269" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-295-222-26-0"><rect x="1177" y="295" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-321-222-26-0"><rect x="1177" y="321" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-347-222-26-0"><rect x="1177" y="347" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-373-222-26-0"><rect x="1177" y="373" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-399-222-26-0"><rect x="1177" y="399" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-425-222-26-0"><rect x="1177" y="425" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-451-222-26-0"><rect x="1177" y="451" width="222" height="26"/></clipPath><clipPath id="mx-clip-1177-477-222-26-0"><rect x="1177" y="477" width="222" height="26"/></clipPath><clipPath id="mx-clip-1527-126-242-26-0"><rect x="1527" y="126" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-152-242-26-0"><rect x="1527" y="152" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-178-242-26-0"><rect x="1527" y="178" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-204-242-26-0"><rect x="1527" y="204" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-230-242-26-0"><rect x="1527" y="230" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-256-242-26-0"><rect x="1527" y="256" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-282-242-26-0"><rect x="1527" y="282" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-308-242-26-0"><rect x="1527" y="308" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-334-242-26-0"><rect x="1527" y="334" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-360-242-26-0"><rect x="1527" y="360" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-386-242-26-0"><rect x="1527" y="386" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-412-242-26-0"><rect x="1527" y="412" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-438-242-26-0"><rect x="1527" y="438" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-464-242-26-0"><rect x="1527" y="464" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-490-242-26-0"><rect x="1527" y="490" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-516-242-26-0"><rect x="1527" y="516" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-542-242-26-0"><rect x="1527" y="542" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-568-242-26-0"><rect x="1527" y="568" width="242" height="26"/></clipPath><clipPath id="mx-clip-1527-594-242-26-0"><rect x="1527" y="594" width="242" height="26"/></clipPath></defs><path d="M 3 56 L 3 30 L 333 30 L 333 56" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 3 56 L 3 758 L 333 758 L 333 56" fill="#ffffff" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 3 56 L 333 56" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><g fill="#000000" font-family="Helvetica" text-anchor="middle" font-size="12px"><text x="167.5" y="47.5">web_client</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-61-322-26-0)" font-size="12px"><text x="8.5" y="73.5">unsigned long long id</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-87-322-26-0)" font-size="12px"><text x="8.5" y="99.5">WEB_CLIENT_FLAGS flags</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-113-322-26-0)" font-size="12px"><text x="8.5" y="125.5">WEB_CLIENT_MODE mode</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-139-322-26-0)" font-size="12px"><text x="8.5" y="151.5">WEB_CLIENT_ACL acl</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-165-322-26-0)" font-size="12px"><text x="8.5" y="177.5">size_t header_parse_tries</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-191-322-26-0)" font-size="12px"><text x="8.5" y="203.5">size_t header_parse_last_size</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-217-322-26-0)" font-size="12px"><text x="8.5" y="229.5">int tcp_cork</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-243-322-26-0)" font-size="12px"><text x="8.5" y="255.5">int fd</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-269-322-26-0)" font-size="12px"><text x="8.5" y="281.5">int fd</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-295-322-26-0)" font-size="12px"><text x="8.5" y="307.5">char remote_ip[NI_MAXHOST+1}</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-321-322-26-0)" font-size="12px"><text x="8.5" y="333.5">char remote_port[NI_MAXSERV+1]</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-347-322-26-0)" font-size="12px"><text x="8.5" y="359.5">char decoded_url[NETDATA_WEB_REQUEST_URL_SIZE+1</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-373-322-26-0)" font-size="12px"><text x="8.5" y="385.5">char last_url[NETDATA_WEB_REQUEST_URL_SIZE+1]</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-399-322-26-0)" font-size="12px"><text x="8.5" y="411.5">struct timeval tv_in, tv_ready</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-425-322-26-0)" font-size="12px"><text x="8.5" y="437.5">char cookie1[NETDATA_WEB_REQUEST_COOKIE_SIZE+1]</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-451-322-26-0)" font-size="12px"><text x="8.5" y="463.5">char cookie2[NETDATA_WEB_REQUEST_COOKIE_SIZE+1]</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-477-322-26-0)" font-size="12px"><text x="8.5" y="489.5">char origin[NETDATA_WEB_REQUEST_ORIGIN_HEADER_SIZE+1]</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-503-322-26-0)" font-size="12px"><text x="8.5" y="515.5">char *user_agent</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-529-322-26-0)" font-size="12px"><text x="8.5" y="541.5">struct response response</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-555-322-26-0)" font-size="12px"><text x="8.5" y="567.5">size_t stats_received_bytes</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-581-322-26-0)" font-size="12px"><text x="8.5" y="593.5">size_t stats_sent_bytes</text></g><path d="M 333 615 L 353 615 L 353 10 L 168 10 L 168 23.63" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 168 28.88 L 164.5 21.88 L 168 23.63 L 171.5 21.88 Z" fill="#000000" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-607-322-26-0)" font-size="12px"><text x="8.5" y="619.5">struct web_client *prev</text></g><path d="M 333 641 L 353 641 L 353 10 L 168 10 L 168 23.63" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 168 28.88 L 164.5 21.88 L 168 23.63 L 171.5 21.88 Z" fill="#000000" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><g transform="translate(304.5,299.5)"><switch><foreignObject style="overflow:visible;" pointer-events="all" width="102" height="12" requiredFeatures="http://www.w3.org/TR/SVG11/feature#Extensibility"><div xmlns="http://www.w3.org/1999/xhtml" style="display: inline-block; font-size: 12px; font-family: Helvetica; color: rgb(0, 0, 0); line-height: 1.2; vertical-align: top; white-space: nowrap; text-align: center;"><div xmlns="http://www.w3.org/1999/xhtml" style="display:inline-block;text-align:inherit;text-decoration:inherit;background-color:#ffffff;">double linked list of</div></div></foreignObject><text x="51" y="12" fill="#000000" text-anchor="middle" font-size="12px" font-family="Helvetica">double linked list of</text></switch></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-633-322-26-0)" font-size="12px"><text x="8.5" y="645.5">struct web_client *next</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-659-322-26-0)" font-size="12px"><text x="8.5" y="671.5">netdata_thread_t thread</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-685-322-26-0)" font-size="12px"><text x="8.5" y="697.5">volatile int running</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-711-322-26-0)" font-size="12px"><text x="8.5" y="723.5">size_t pollinfo_slot</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-7-737-322-26-0)" font-size="12px"><text x="8.5" y="749.5">size_t pollinfo_filecopy_slot</text></g><path d="M 473 160 L 473 134 L 733 134 L 733 160" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 473 160 L 473 472 L 733 472 L 733 160" fill="#ffffff" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 473 160 L 733 160" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><g fill="#000000" font-family="Helvetica" text-anchor="middle" font-size="12px"><text x="602.5" y="151.5">response</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-165-252-26-0)" font-size="12px"><text x="478.5" y="177.5">BUFFER *header</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-191-252-26-0)" font-size="12px"><text x="478.5" y="203.5">BUFFER *header_output</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-217-252-26-0)" font-size="12px"><text x="478.5" y="229.5">BUFFER *data</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-243-252-26-0)" font-size="12px"><text x="478.5" y="255.5">int code</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-269-252-26-0)" font-size="12px"><text x="478.5" y="281.5">size_t rlen</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-295-252-26-0)" font-size="12px"><text x="478.5" y="307.5">size_t sent</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-321-252-26-0)" font-size="12px"><text x="478.5" y="333.5">int zoutput</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-347-252-26-0)" font-size="12px"><text x="478.5" y="359.5">z_stream zstream</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-373-252-26-0)" font-size="12px"><text x="478.5" y="385.5">Bytef zbuffer[NETDATA_WEB_RESPONSE_ZLIB_CHUNK_SIZE]</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-399-252-26-0)" font-size="12px"><text x="478.5" y="411.5">size_t zsent</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-425-252-26-0)" font-size="12px"><text x="478.5" y="437.5">size_t zhave</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-451-252-26-0)" font-size="12px"><text x="478.5" y="463.5">unsigned int zinitialized</text></g><path d="M 333 537 L 403 537 L 403 114 L 603 114 L 603 127.63" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 603 132.88 L 599.5 125.88 L 603 127.63 L 606.5 125.88 Z" fill="#000000" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 473 572 L 473 546 L 733 546 L 733 572" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 473 572 L 473 754 L 733 754 L 733 572" fill="#ffffff" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 473 572 L 733 572" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><g fill="#000000" font-family="Helvetica" text-anchor="middle" font-size="12px"><text x="602.5" y="563.5">clients_cache</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-577-252-26-0)" font-size="12px"><text x="478.5" y="589.5">pid_t pid</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-603-252-26-0)" font-size="12px"><text x="478.5" y="615.5">struct web_client *used</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-629-252-26-0)" font-size="12px"><text x="478.5" y="641.5">size_t used_count</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-655-252-26-0)" font-size="12px"><text x="478.5" y="667.5">struct web_client *avail</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-681-252-26-0)" font-size="12px"><text x="478.5" y="693.5">size_t avail_count</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-707-252-26-0)" font-size="12px"><text x="478.5" y="719.5">size_t reused</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-477-733-252-26-0)" font-size="12px"><text x="478.5" y="745.5">size_t allocated</text></g><path d="M 733 611 L 753 611 L 753 10 L 168 10 L 168 23.63" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 168 28.88 L 164.5 21.88 L 168 23.63 L 171.5 21.88 Z" fill="#000000" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><g transform="translate(724.5,12.5)"><switch><foreignObject style="overflow:visible;" pointer-events="all" width="57" height="11" requiredFeatures="http://www.w3.org/TR/SVG11/feature#Extensibility"><div xmlns="http://www.w3.org/1999/xhtml" style="display: inline-block; font-size: 11px; font-family: Helvetica; color: rgb(0, 0, 0); line-height: 1.2; vertical-align: top; white-space: nowrap; text-align: center;"><div xmlns="http://www.w3.org/1999/xhtml" style="display:inline-block;text-align:inherit;text-decoration:inherit;background-color:#ffffff;">linked list of</div></div></foreignObject><text x="29" y="11" fill="#000000" text-anchor="middle" font-size="11px" font-family="Helvetica">linked list of</text></switch></g><path d="M 733 663 L 753 663 L 753 10 L 168 10 L 168 23.63" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 168 28.88 L 164.5 21.88 L 168 23.63 L 171.5 21.88 Z" fill="#000000" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><g transform="translate(724.5,38.5)"><switch><foreignObject style="overflow:visible;" pointer-events="all" width="57" height="11" requiredFeatures="http://www.w3.org/TR/SVG11/feature#Extensibility"><div xmlns="http://www.w3.org/1999/xhtml" style="display: inline-block; font-size: 11px; font-family: Helvetica; color: rgb(0, 0, 0); line-height: 1.2; vertical-align: top; white-space: nowrap; text-align: center;"><div xmlns="http://www.w3.org/1999/xhtml" style="display:inline-block;text-align:inherit;text-decoration:inherit;background-color:#ffffff;">linked list of</div></div></foreignObject><text x="29" y="11" fill="#000000" text-anchor="middle" font-size="11px" font-family="Helvetica">linked list of</text></switch></g><path d="M 833 108 L 833 82 L 1073 82 L 1073 108" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 833 108 L 833 394 L 1073 394 L 1073 108" fill="#ffffff" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 833 108 L 1073 108" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><g fill="#000000" font-family="Helvetica" text-anchor="middle" font-size="12px"><text x="952.5" y="99.5">listen_sockets</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-837-113-232-26-0)" font-size="12px"><text x="838.5" y="125.5">struct config *config</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-837-139-232-26-0)" font-size="12px"><text x="838.5" y="151.5">const char *config_section</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-837-165-232-26-0)" font-size="12px"><text x="838.5" y="177.5">const char *default_bind_to</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-837-191-232-26-0)" font-size="12px"><text x="838.5" y="203.5">uint16_t default_port</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-837-217-232-26-0)" font-size="12px"><text x="838.5" y="229.5">int backlog</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-837-243-232-26-0)" font-size="12px"><text x="838.5" y="255.5">size_t opened</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-837-269-232-26-0)" font-size="12px"><text x="838.5" y="281.5">size_t failed</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-837-295-232-26-0)" font-size="12px"><text x="838.5" y="307.5">int fds[MAX_LISTEN_FDS]</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-837-321-232-26-0)" font-size="12px"><text x="838.5" y="333.5">int *fds_names[MAX_LISTEN_FDS]</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-837-347-232-26-0)" font-size="12px"><text x="838.5" y="359.5">int fds_types[MAX_LISTEN_FDS]</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-837-373-232-26-0)" font-size="12px"><text x="838.5" y="385.5">int fds_families[MAX_LISTEN_FDS]</text></g><path d="M 1173 56 L 1173 30 L 1403 30 L 1403 56" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 1173 56 L 1173 498 L 1403 498 L 1403 56" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 1173 56 L 1403 56" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><g fill="#000000" font-family="Helvetica" text-anchor="middle" font-size="12px"><text x="1287.5" y="47.5">POLLINFO</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-61-222-26-0)" font-size="12px"><text x="1178.5" y="73.5">POLLJOB *p</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-87-222-26-0)" font-size="12px"><text x="1178.5" y="99.5">size_t slot</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-113-222-26-0)" font-size="12px"><text x="1178.5" y="125.5">int fd</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-139-222-26-0)" font-size="12px"><text x="1178.5" y="151.5">int socktype</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-165-222-26-0)" font-size="12px"><text x="1178.5" y="177.5">char *remote_ip</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-191-222-26-0)" font-size="12px"><text x="1178.5" y="203.5">char *remote_port</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-217-222-26-0)" font-size="12px"><text x="1178.5" y="229.5">time_t connected_t</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-243-222-26-0)" font-size="12px"><text x="1178.5" y="255.5">time_t last_received_t</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-269-222-26-0)" font-size="12px"><text x="1178.5" y="281.5">time_t last_sent_t</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-295-222-26-0)" font-size="12px"><text x="1178.5" y="307.5">size_t recv_count</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-321-222-26-0)" font-size="12px"><text x="1178.5" y="333.5">size_t send_count</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-347-222-26-0)" font-size="12px"><text x="1178.5" y="359.5">uint32_t flags</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-373-222-26-0)" font-size="12px"><text x="1178.5" y="385.5">void (*del_callback)</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-399-222-26-0)" font-size="12px"><text x="1178.5" y="411.5">int (*rcv_callback)</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-425-222-26-0)" font-size="12px"><text x="1178.5" y="437.5">int (*snd_callback)</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-451-222-26-0)" font-size="12px"><text x="1178.5" y="463.5">void *data</text></g><path d="M 1403 485 L 1423 485 L 1423 10 L 1279 10 L 1279 22.63" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 1279 27.88 L 1275.5 20.88 L 1279 22.63 L 1282.5 20.88 Z" fill="#000000" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1177-477-222-26-0)" font-size="12px"><text x="1178.5" y="489.5">struct pollinfo *next</text></g><path d="M 1523 121 L 1523 95 L 1773 95 L 1773 121" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 1523 121 L 1523 615 L 1773 615 L 1773 121" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 1523 121 L 1773 121" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><g fill="#000000" font-family="Helvetica" text-anchor="middle" font-size="12px"><text x="1647.5" y="112.5">POLLJOB</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-126-242-26-0)" font-size="12px"><text x="1528.5" y="138.5">size_t slots</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-152-242-26-0)" font-size="12px"><text x="1528.5" y="164.5">size_t used</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-178-242-26-0)" font-size="12px"><text x="1528.5" y="190.5">size_t min</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-204-242-26-0)" font-size="12px"><text x="1528.5" y="216.5">size_t max</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-230-242-26-0)" font-size="12px"><text x="1528.5" y="242.5">size_t limit</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-256-242-26-0)" font-size="12px"><text x="1528.5" y="268.5">time_t complete_request_timeout</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-282-242-26-0)" font-size="12px"><text x="1528.5" y="294.5">time_t idle_timeout</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-308-242-26-0)" font-size="12px"><text x="1528.5" y="320.5">time_t check_every</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-334-242-26-0)" font-size="12px"><text x="1528.5" y="346.5">time_t timer_milliseconds</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-360-242-26-0)" font-size="12px"><text x="1528.5" y="372.5">void *timer_data</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-386-242-26-0)" font-size="12px"><text x="1528.5" y="398.5">struct run *fds</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-412-242-26-0)" font-size="12px"><text x="1528.5" y="424.5">struct pollinfo *inf</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-438-242-26-0)" font-size="12px"><text x="1528.5" y="450.5">struct pollinfo *first_free</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-464-242-26-0)" font-size="12px"><text x="1528.5" y="476.5">SIMPLE_PATTERN *access_list</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-490-242-26-0)" font-size="12px"><text x="1528.5" y="502.5">void *(*add_callback)</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-516-242-26-0)" font-size="12px"><text x="1528.5" y="528.5">void (*dell_callback)</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-542-242-26-0)" font-size="12px"><text x="1528.5" y="554.5">int (*rcv_callback)</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-568-242-26-0)" font-size="12px"><text x="1528.5" y="580.5">int (*snd_callback)</text></g><g fill="#000000" font-family="Helvetica" clip-path="url(#mx-clip-1527-594-242-26-0)" font-size="12px"><text x="1528.5" y="606.5">void (*tmr_callback)</text></g><path d="M 1403 69 L 1648 69 L 1648 88.63" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 1648 93.88 L 1644.5 86.88 L 1648 88.63 L 1651.5 86.88 Z" fill="#000000" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 1773 420 L 1793 420 L 1793 10 L 1279 10 L 1279 22.63" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 1279 27.88 L 1275.5 20.88 L 1279 22.63 L 1282.5 20.88 Z" fill="#000000" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 1773 446 L 1793 446 L 1793 10 L 1278 10 L 1278 22.63" fill="none" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/><path d="M 1278 27.88 L 1274.5 20.88 L 1278 22.63 L 1281.5 20.88 Z" fill="#000000" stroke="#000000" stroke-miterlimit="10" pointer-events="none"/></svg>
\ No newline at end of file
src/collectors/apps.plugin/apps_pid.c
+3
-1
@@ -53,7 +53,9 @@ size_t all_pids_count(void) {
53
}
54
55
void apps_pids_init(void) {
56
- pids.all_pids.aral = aral_create("pid_stat", sizeof(struct pid_stat), 1, 0, NULL, NULL, NULL, false, true);
56
+ pids.all_pids.aral = aral_create("pid_stat", sizeof(struct pid_stat),
57
+ 1, 0, NULL, NULL, NULL,
58
+ false, true, false);
59
simple_hashtable_init_PID(&pids.all_pids.ht, 1024);
60
}
61
src/collectors/ebpf.plugin/ebpf.c
+3
-2
@@ -739,8 +739,9 @@ ARAL *ebpf_allocate_pid_aral(char *name, size_t size)
739
}
740
741
return aral_create(name, size,
742
- 0, 0,
743
- NULL, NULL, NULL, false, false);
742
+ 0, 0,
743
+ NULL, NULL, NULL,
744
+ false, false, false);
745
}
746
747
/*****************************************************************
src/collectors/statsd.plugin/statsd.c
+11
-13
@@ -261,7 +261,7 @@ static struct statsd {
261
size_t udp_packets_received;
262
size_t udp_bytes_read;
263
264
- int32_t update_every;
264
+ time_t update_every;
265
bool enabled;
266
bool private_charts_hidden;
267
SIMPLE_PATTERN *charts_for;
@@ -873,12 +873,12 @@ struct statsd_udp {
873
};
874
875
// new TCP client connected
876
-static void *statsd_add_callback(POLLINFO *pi, short int *events, void *data) {
876
+static void *statsd_add_callback(POLLINFO *pi, nd_poll_event_t *events, void *data) {
877
(void)pi;
878
(void)data;
879
880
worker_is_busy(WORKER_JOB_TYPE_TCP_CONNECTED);
881
- *events = POLLIN;
881
+ *events = ND_POLL_READ;
882
883
struct statsd_tcp *t = (struct statsd_tcp *)callocz(sizeof(struct statsd_tcp) + STATSD_TCP_BUFFER_SIZE, 1);
884
t->type = STATSD_SOCKET_DATA_TYPE_TCP;
@@ -916,11 +916,11 @@ static void statsd_del_callback(POLLINFO *pi) {
916
}
917
918
// Receive data
919
-static int statsd_rcv_callback(POLLINFO *pi, short int *events) {
919
+static int statsd_rcv_callback(POLLINFO *pi, nd_poll_event_t *events) {
920
int retval = -1;
921
worker_is_busy(WORKER_JOB_TYPE_RCV_DATA);
922
923
- *events = POLLIN;
923
+ *events = ND_POLL_READ;
924
925
int fd = pi->fd;
926
@@ -1064,10 +1064,7 @@ cleanup:
1064
return retval;
1065
}
1066
1067
-static int statsd_snd_callback(POLLINFO *pi, short int *events) {
1068
- (void)pi;
1069
- (void)events;
1070
-
1067
+static int statsd_snd_callback(POLLINFO *pi __maybe_unused, nd_poll_event_t *events __maybe_unused) {
1068
worker_is_busy(WORKER_JOB_TYPE_SND_DATA);
1069
netdata_log_error("STATSD: snd_callback() called, but we never requested to send data to statsd clients.");
1070
worker_is_idle();
@@ -2487,11 +2484,12 @@ void *statsd_main(void *ptr) {
2484
2485
statsd.enabled = config_get_boolean(CONFIG_SECTION_PLUGINS, "statsd", statsd.enabled);
2486
2490
- statsd.update_every = default_rrd_update_every;
2487
+ statsd.update_every = nd_profile.update_every;
2488
statsd.update_every = (int)config_get_duration_seconds(CONFIG_SECTION_STATSD, "update every (flushInterval)", statsd.update_every);
2492
- if(statsd.update_every < default_rrd_update_every) {
2493
- collector_error("STATSD: minimum flush interval %d given, but the minimum is the update every of netdata. Using %d", statsd.update_every, default_rrd_update_every);
2494
- statsd.update_every = default_rrd_update_every;
2489
+ if(statsd.update_every < nd_profile.update_every) {
2490
+ collector_error("STATSD: minimum flush interval %d given, but the minimum is the update every of netdata. Using %d",
2491
+ (int)statsd.update_every, (int)nd_profile.update_every);
2492
+ statsd.update_every = nd_profile.update_every;
2493
config_set_duration_seconds(CONFIG_SECTION_STATSD, "update every (flushInterval)", statsd.update_every);
2494
}
2495
src/collectors/systemd-journal.plugin/systemd-journal-annotations.c
+2
@@ -610,6 +610,8 @@ static void netdata_systemd_journal_message_ids_init(void) {
610
msgid_into_dict("6e2e3839067648968b646045dbf28d66", "Netdata connection to parent");
611
msgid_into_dict("9ce0cb58ab8b44df82c4bf1ad9ee22de", "Netdata alert transition");
612
msgid_into_dict("6db0018e83e34320ae2a659d78019fb7", "Netdata alert notification");
613
+ msgid_into_dict("23e93dfccbf64e11aac858b9410d8a82", "Netdata fatal message");
614
+
615
msgid_into_dict("8ddaf5ba33a74078b609250db1e951f3", "Sensor state transition");
616
}
617
src/collectors/windows-events.plugin/windows-events-providers.c
+2
-2
@@ -99,8 +99,8 @@ static inline bool provider_cache_compar(PROVIDER_KEY *a, PROVIDER_KEY *b) {
99
100
void provider_cache_init(void) {
101
simple_hashtable_init_PROVIDER(&pbc.hashtable, 100000);
102
- pbc.aral_providers = aral_create("wevt_providers", sizeof(PROVIDER), 0, 4096, NULL, NULL, NULL, false, true);
103
- pbc.aral_handles = aral_create("wevt_handles", sizeof(PROVIDER_META_HANDLE), 0, 4096, NULL, NULL, NULL, false, true);
102
+ pbc.aral_providers = aral_create("wevt_providers", sizeof(PROVIDER), 0, 4096, NULL, NULL, NULL, false, true, false);
103
+ pbc.aral_handles = aral_create("wevt_handles", sizeof(PROVIDER_META_HANDLE), 0, 4096, NULL, NULL, NULL, false, true, false);
104
}
105
106
static bool provider_property_get(PROVIDER_META_HANDLE *h, WEVT_VARIANT *content, EVT_PUBLISHER_METADATA_PROPERTY_ID property_id) {
src/daemon/analytics.c
+1
-1
@@ -518,7 +518,7 @@ void analytics_gather_mutable_meta_data(void)
518
analytics_alarms_notifications();
519
520
analytics_set_data(
521
- &analytics_data.netdata_config_is_parent, (rrdhost_hosts_available() > 1 || stream_conf_is_parent(false)) ? "true" : "false");
521
+ &analytics_data.netdata_config_is_parent, (rrdhost_hosts_available() > 1 || netdata_conf_is_parent()) ? "true" : "false");
522
523
analytics_set_data(&analytics_data.netdata_host_agent_claimed, is_agent_claimed() ? "true" : "false");
524
src/daemon/config/netdata-conf-cloud.c
+1
-1
@@ -5,7 +5,7 @@
5
6
size_t netdata_conf_cloud_query_threads(void) {
7
size_t cpus = MIN(netdata_conf_cpus(), 256); // max 256 cores
8
- size_t threads = MIN(cpus * (stream_conf_is_parent(false) ? 2 : 1), (size_t)libuv_worker_threads / 2);
8
+ size_t threads = MIN(cpus * (netdata_conf_is_parent() ? 2 : 1), (size_t)libuv_worker_threads / 2);
9
threads = MAX(threads, 6);
10
11
threads = config_get_number(CONFIG_SECTION_CLOUD, "query threads", threads);
src/daemon/config/netdata-conf-db.c
+37
-29
@@ -2,11 +2,9 @@
2
3
#include "netdata-conf-db.h"
4
5
-int default_rrd_update_every = UPDATE_EVERY;
5
int default_rrd_history_entries = RRD_DEFAULT_HISTORY_ENTRIES;
6
7
bool dbengine_enabled = false; // will become true if and when dbengine is initialized
9
-size_t storage_tiers = 3;
8
bool dbengine_use_direct_io = true;
9
static size_t storage_tiers_grouping_iterations[RRD_STORAGE_TIERS] = {1, 60, 60, 60, 60};
10
static double storage_tiers_retention_days[RRD_STORAGE_TIERS] = {14, 90, 2 * 365, 2 * 365, 2 * 365};
@@ -16,7 +14,7 @@ time_t rrdhost_free_orphan_time_s = 3600;
14
time_t rrdhost_free_ephemeral_time_s = 86400;
15
16
size_t get_tier_grouping(size_t tier) {
19
- if(unlikely(tier >= storage_tiers)) tier = storage_tiers - 1;
17
+ if(unlikely(tier >= nd_profile.storage_tiers)) tier = nd_profile.storage_tiers - 1;
18
19
size_t grouping = 1;
20
// first tier is always 1 iteration of whatever update every the chart has
@@ -191,20 +189,20 @@ void netdata_conf_dbengine_init(const char *hostname) {
189
config_set_number(CONFIG_SECTION_DB, "dbengine pages per extent", rrdeng_pages_per_extent);
190
}
191
194
- storage_tiers = config_get_number(CONFIG_SECTION_DB, "storage tiers", storage_tiers);
195
- if(storage_tiers < 1) {
192
+ nd_profile.storage_tiers = config_get_number(CONFIG_SECTION_DB, "storage tiers", nd_profile.storage_tiers);
193
+ if(nd_profile.storage_tiers < 1) {
194
nd_log(NDLS_DAEMON, NDLP_WARNING, "At least 1 storage tier is required. Assuming 1.");
195
198
- storage_tiers = 1;
199
- config_set_number(CONFIG_SECTION_DB, "storage tiers", storage_tiers);
196
+ nd_profile.storage_tiers = 1;
197
+ config_set_number(CONFIG_SECTION_DB, "storage tiers", nd_profile.storage_tiers);
198
}
201
- if(storage_tiers > RRD_STORAGE_TIERS) {
199
+ if(nd_profile.storage_tiers > RRD_STORAGE_TIERS) {
200
nd_log(NDLS_DAEMON, NDLP_WARNING,
201
"Up to %d storage tier are supported. Assuming %d.",
202
RRD_STORAGE_TIERS, RRD_STORAGE_TIERS);
203
206
- storage_tiers = RRD_STORAGE_TIERS;
207
- config_set_number(CONFIG_SECTION_DB, "storage tiers", storage_tiers);
204
+ nd_profile.storage_tiers = RRD_STORAGE_TIERS;
205
+ config_set_number(CONFIG_SECTION_DB, "storage tiers", nd_profile.storage_tiers);
206
}
207
208
new_dbengine_defaults =
@@ -221,10 +219,10 @@ void netdata_conf_dbengine_init(const char *hostname) {
219
default_backfill = get_dbengine_backfill(RRD_BACKFILL_NEW);
220
char dbengineconfig[200 + 1];
221
224
- size_t grouping_iterations = default_rrd_update_every;
225
- storage_tiers_grouping_iterations[0] = default_rrd_update_every;
222
+ size_t grouping_iterations = nd_profile.update_every;
223
+ storage_tiers_grouping_iterations[0] = nd_profile.update_every;
224
227
- for (size_t tier = 1; tier < storage_tiers; tier++) {
225
+ for (size_t tier = 1; tier < nd_profile.storage_tiers; tier++) {
226
grouping_iterations = storage_tiers_grouping_iterations[tier];
227
snprintfz(dbengineconfig, sizeof(dbengineconfig) - 1, "dbengine tier %zu update every iterations", tier);
228
grouping_iterations = config_get_number(CONFIG_SECTION_DB, dbengineconfig, grouping_iterations);
@@ -250,7 +248,7 @@ void netdata_conf_dbengine_init(const char *hostname) {
248
// fails on Windows.
249
bool parallel_initialization = false;
250
#else
253
- bool parallel_initialization = (storage_tiers <= netdata_conf_cpus()) ? true : false;
251
+ bool parallel_initialization = (nd_profile.storage_tiers <= netdata_conf_cpus()) ? true : false;
252
#endif
253
254
struct dbengine_initialization tiers_init[RRD_STORAGE_TIERS] = {};
@@ -258,7 +256,7 @@ void netdata_conf_dbengine_init(const char *hostname) {
256
size_t created_tiers = 0;
257
char dbenginepath[FILENAME_MAX + 1];
258
261
- for (size_t tier = 0; tier < storage_tiers; tier++) {
259
+ for (size_t tier = 0; tier < nd_profile.storage_tiers; tier++) {
260
261
if (tier == 0)
262
snprintfz(dbenginepath, FILENAME_MAX, "%s/dbengine", netdata_configured_cache_dir);
@@ -294,7 +292,7 @@ void netdata_conf_dbengine_init(const char *hostname) {
292
dbengine_tier_init(&tiers_init[tier]);
293
}
294
297
- for(size_t tier = 0; tier < storage_tiers ;tier++) {
295
+ for(size_t tier = 0; tier < nd_profile.storage_tiers;tier++) {
296
if(parallel_initialization)
297
nd_thread_join(tiers_init[tier].thread);
298
@@ -307,31 +305,32 @@ void netdata_conf_dbengine_init(const char *hostname) {
305
created_tiers++;
306
}
307
310
- if(created_tiers && created_tiers < storage_tiers) {
308
+ if(created_tiers && created_tiers < nd_profile.storage_tiers) {
309
nd_log(NDLS_DAEMON, NDLP_WARNING,
310
"DBENGINE on '%s': Managed to create %zu tiers instead of %zu. Continuing with %zu available.",
313
- hostname, created_tiers, storage_tiers, created_tiers);
311
+ hostname, created_tiers,
312
+ nd_profile.storage_tiers, created_tiers);
313
315
- storage_tiers = created_tiers;
314
+ nd_profile.storage_tiers = created_tiers;
315
}
316
else if(!created_tiers)
317
fatal("DBENGINE on '%s', failed to initialize databases at '%s'.", hostname, netdata_configured_cache_dir);
318
320
- for(size_t tier = 0; tier < storage_tiers ;tier++)
319
+ for(size_t tier = 0; tier < nd_profile.storage_tiers;tier++)
320
rrdeng_readiness_wait(multidb_ctx[tier]);
321
322
calculate_tier_disk_space_percentage();
323
324
dbengine_enabled = true;
325
#else
327
- storage_tiers = config_get_number(CONFIG_SECTION_DB, "storage tiers", 1);
328
- if(storage_tiers != 1) {
326
+ nd_profile.storage_tiers = config_get_number(CONFIG_SECTION_DB, "storage tiers", 1);
327
+ if(nd_profile.storage_tiers != 1) {
328
nd_log(NDLS_DAEMON, NDLP_WARNING,
329
"DBENGINE is not available on '%s', so only 1 database tier can be supported.",
330
hostname);
331
333
- storage_tiers = 1;
334
- config_set_number(CONFIG_SECTION_DB, "storage tiers", storage_tiers);
332
+ nd_profile.storage_tiers = 1;
333
+ config_set_number(CONFIG_SECTION_DB, "storage tiers", nd_profile.storage_tiers);
334
}
335
dbengine_enabled = false;
336
#endif
@@ -350,11 +349,20 @@ void netdata_conf_section_db(void) {
349
// ------------------------------------------------------------------------
350
// get default database update frequency
351
353
- default_rrd_update_every = (int) config_get_duration_seconds(CONFIG_SECTION_DB, "update every", UPDATE_EVERY);
354
- if(default_rrd_update_every < 1 || default_rrd_update_every > 600) {
355
- netdata_log_error("Invalid data collection frequency (update every) %d given. Defaulting to %d.", default_rrd_update_every, UPDATE_EVERY);
356
- default_rrd_update_every = UPDATE_EVERY;
357
- config_set_duration_seconds(CONFIG_SECTION_DB, "update every", default_rrd_update_every);
352
+ nd_profile.update_every = (int) config_get_duration_seconds(CONFIG_SECTION_DB, "update every", nd_profile.update_every);
353
+ if(nd_profile.update_every < UPDATE_EVERY_MIN) {
354
+ nd_log(NDLS_DAEMON, NDLP_WARNING,
355
+ "Data collection frequency in netdata.conf ([" CONFIG_SECTION_DB "].update every), changed from %d to %d",
356
+ (int)nd_profile.update_every, UPDATE_EVERY_MIN);
357
+ nd_profile.update_every = UPDATE_EVERY_MIN;
358
+ config_set_duration_seconds(CONFIG_SECTION_DB, "update every", nd_profile.update_every);
359
+ }
360
+ if(nd_profile.update_every > UPDATE_EVERY_MAX) {
361
+ nd_log(NDLS_DAEMON, NDLP_WARNING,
362
+ "Data collection frequency in netdata.conf ([" CONFIG_SECTION_DB "].update every), changed from %d to %d",
363
+ (int)nd_profile.update_every, UPDATE_EVERY_MIN);
364
+ nd_profile.update_every = UPDATE_EVERY_MAX;
365
+ config_set_duration_seconds(CONFIG_SECTION_DB, "update every", nd_profile.update_every);
366
}
367
368
// ------------------------------------------------------------------------
src/daemon/config/netdata-conf-db.h
-2
@@ -6,10 +6,8 @@
6
#include "libnetdata/libnetdata.h"
7
8
extern bool dbengine_enabled;
9
-extern size_t storage_tiers;
9
extern bool dbengine_use_direct_io;
10
12
-extern int default_rrd_update_every;
11
extern int default_rrd_history_entries;
12
extern int gap_when_lost_iterations_above;
13
extern time_t rrdset_free_obsolete_time_s;
src/daemon/config/netdata-conf-global.c
+27
-8
@@ -55,15 +55,33 @@ static int get_hostname(char *buf, size_t buf_size) {
55
return rc;
56
}
57
58
-static void glibc_initialize(void) {
59
- const char *pmax = config_get(CONFIG_SECTION_GLOBAL, "glibc malloc arena max for plugins", "1");
60
- if(pmax && *pmax)
61
- setenv("MALLOC_ARENA_MAX", pmax, 1);
58
+void netdata_conf_glibc_malloc_initialize(size_t wanted_arenas, size_t trim_threshold __maybe_unused) {
59
+ wanted_arenas = config_get_number(CONFIG_SECTION_GLOBAL, "glibc malloc arena max for plugins", wanted_arenas);
60
+ if(wanted_arenas < 1 || wanted_arenas > os_get_system_cpus_cached(true)) {
61
+ if(wanted_arenas < 1) wanted_arenas = 1;
62
+ else wanted_arenas = os_get_system_cpus_cached(true);
63
+ config_set_number(CONFIG_SECTION_GLOBAL, "glibc malloc arena max for plugins", wanted_arenas);
64
+ nd_log(NDLS_DAEMON, NDLP_NOTICE,
65
+ "malloc arenas can be from 1 to %zu. Setting it to %zu",
66
+ os_get_system_cpus_cached(true), wanted_arenas);
67
+ }
68
+
69
+ char buf[32];
70
+ snprintfz(buf, sizeof(buf), "%zu", wanted_arenas);
71
+ setenv("MALLOC_ARENA_MAX", buf, true);
72
73
#if defined(HAVE_C_MALLOPT)
64
- int i = (int)config_get_number(CONFIG_SECTION_GLOBAL, "glibc malloc arena max for netdata", 1);
65
- if(i > 0)
66
- mallopt(M_ARENA_MAX, 1);
74
+ wanted_arenas = config_get_number(CONFIG_SECTION_GLOBAL, "glibc malloc arena max for netdata", wanted_arenas);
75
+ if(wanted_arenas < 1 || wanted_arenas > os_get_system_cpus_cached(true)) {
76
+ if(wanted_arenas < 1) wanted_arenas = 1;
77
+ else wanted_arenas = os_get_system_cpus_cached(true);
78
+ config_set_number(CONFIG_SECTION_GLOBAL, "glibc malloc arena max for netdata", wanted_arenas);
79
+ nd_log(NDLS_DAEMON, NDLP_NOTICE,
80
+ "malloc arenas can be from 1 to %zu. Setting it to %zu",
81
+ os_get_system_cpus_cached(true), wanted_arenas);
82
+ }
83
+ mallopt(M_ARENA_MAX, (int)wanted_arenas);
84
+ mallopt(M_TRIM_THRESHOLD, (int)trim_threshold);
85
86
#ifdef NETDATA_INTERNAL_CHECKS
87
mallopt(M_PERTURB, 0x5A);
@@ -108,6 +126,8 @@ void netdata_conf_section_global(void) {
126
netdata_log_debug(D_OPTIONS, "hostname set to '%s'", netdata_configured_hostname);
127
128
netdata_conf_section_directories();
129
+
130
+ nd_profile_setup(); // required for configuring the database
131
netdata_conf_section_db();
132
133
// --------------------------------------------------------------------
@@ -116,7 +136,6 @@ void netdata_conf_section_global(void) {
136
os_get_system_cpus_uncached();
137
os_get_system_pid_max();
138
119
- glibc_initialize();
139
libuv_initialize();
140
}
141
src/daemon/config/netdata-conf-global.h
+2
@@ -10,6 +10,8 @@ void netdata_conf_section_global_run_as_user(const char **user);
10
11
size_t netdata_conf_cpus(void);
12
13
+void netdata_conf_glibc_malloc_initialize(size_t wanted_arenas, size_t trim_threshold);
14
+
15
#include "netdata-conf.h"
16
17
#endif //NETDATA_NETDATA_CONF_GLOBAL_H
src/daemon/config/netdata-conf-profile.c
new
+155
@@ -0,0 +1,155 @@
1
+// SPDX-License-Identifier: GPL-3.0-or-later
2
+
3
+#include "netdata-conf-profile.h"
4
+#include "streaming/stream-conf.h"
5
+#include "netdata-conf.h"
6
+
7
+ENUM_STR_MAP_DEFINE(ND_PROFILE) = {
8
+ { .id = ND_PROFILE_STANDALONE, .name = "standalone" },
9
+ { .id = ND_PROFILE_PARENT, .name = "parent" },
10
+ { .id = ND_PROFILE_CHILD, .name = "child" },
11
+ { .id = ND_PROFILE_IOT, .name = "iot" },
12
+
13
+ // terminator
14
+ { . id = 0, .name = NULL }
15
+};
16
+
17
+BITMAP_STR_DEFINE_FUNCTIONS(ND_PROFILE, ND_PROFILE_NONE, "");
18
+
19
+static inline ND_PROFILE prefer_profile(ND_PROFILE setting, ND_PROFILE preferred, ND_PROFILE out_of) {
20
+ if(setting & preferred) {
21
+ setting &= ~out_of;
22
+ setting |= preferred;
23
+ }
24
+ return setting;
25
+}
26
+
27
+ND_PROFILE nd_profile_detect_and_configure(bool recheck) {
28
+ static ND_PROFILE profile = ND_PROFILE_NONE;
29
+ if(!recheck && profile != ND_PROFILE_NONE)
30
+ return profile;
31
+
32
+ // required for detecting the profile
33
+ stream_conf_load();
34
+
35
+ ND_PROFILE def_profile = ND_PROFILE_NONE;
36
+
37
+ OS_SYSTEM_MEMORY mem = os_system_memory(true);
38
+ size_t cpus = os_get_system_cpus_uncached();
39
+
40
+ if(cpus <= 1 || (mem.ram_total_bytes && mem.ram_total_bytes < 1ULL * 1024 * 1024 * 1024))
41
+ def_profile = ND_PROFILE_IOT;
42
+
43
+ else if(stream_conf_is_parent(true))
44
+ def_profile = ND_PROFILE_PARENT;
45
+
46
+ else if(stream_conf_is_child())
47
+ def_profile = ND_PROFILE_CHILD;
48
+
49
+ else
50
+ def_profile = ND_PROFILE_STANDALONE;
51
+
52
+ CLEAN_BUFFER *wb = buffer_create(0, NULL);
53
+ ND_PROFILE_2buffer(wb, def_profile, " ");
54
+
55
+ CLEAN_CHAR_P *s = strdupz(config_get(CONFIG_SECTION_GLOBAL, "profile", buffer_tostring(wb)));
56
+
57
+ char *words[100];
58
+ size_t n = quoted_strings_splitter(s, words, _countof(words), isspace_map_whitespace);
59
+ ND_PROFILE pt = ND_PROFILE_NONE;
60
+ for(size_t i = 0; i < n ;i++) {
61
+ ND_PROFILE ptt = ND_PROFILE_2id_one(words[i]);
62
+ if(ptt == ND_PROFILE_NONE)
63
+ nd_log(NDLS_DAEMON, NDLP_ERR, "Cannot understand netdata.conf [global].profile = %s", words[i]);
64
+ pt |= ptt;
65
+ }
66
+
67
+ // sanity checks
68
+
69
+ ND_PROFILE started = pt;
70
+
71
+ if(!(pt & ND_CONF_PROFILES_SYSTEM))
72
+ // system profile is missing from the settings
73
+ pt |= (def_profile & ND_CONF_PROFILES_SYSTEM);
74
+
75
+ pt = prefer_profile(pt, ND_PROFILE_PARENT, ND_CONF_PROFILES_SYSTEM);
76
+ pt = prefer_profile(pt, ND_PROFILE_STANDALONE, ND_CONF_PROFILES_SYSTEM);
77
+ pt = prefer_profile(pt, ND_PROFILE_CHILD, ND_CONF_PROFILES_SYSTEM);
78
+ pt = prefer_profile(pt, ND_PROFILE_IOT, ND_CONF_PROFILES_SYSTEM);
79
+
80
+ if(pt != started) {
81
+ buffer_flush(wb);
82
+ ND_PROFILE_2buffer(wb, pt, " ");
83
+ config_set(CONFIG_SECTION_GLOBAL, "profile", buffer_tostring(wb));
84
+
85
+ nd_log(NDLS_DAEMON, NDLP_WARNING,
86
+ "The netdata.conf setting [global].profile has been overwritten to '%s'",
87
+ buffer_tostring(wb));
88
+ }
89
+
90
+ profile = pt;
91
+ return profile;
92
+}
93
+
94
+struct nd_profile_t nd_profile = { 0 };
95
+
96
+void nd_profile_setup(void) {
97
+ static bool run = false;
98
+ if(run) return;
99
+ run = true;
100
+
101
+ ND_PROFILE profile = nd_profile_detect_and_configure(true); (void)profile;
102
+ if(netdata_conf_is_iot()) {
103
+ nd_profile.storage_tiers = 3; // MUST BE 1
104
+ nd_profile.update_every = 1; // MUST BE 2
105
+ nd_profile.malloc_arenas = 1;
106
+ nd_profile.malloc_trim = 32 * 1024;
107
+ // web server threads = 6
108
+ // aclk query threads = 6
109
+ // backfill threads = 0
110
+ // replication threads = 1 // can we disable them completely?
111
+ // ml enabled = false
112
+ // health enabled = true if there is no parent, otherwise false
113
+ // disable internal plugins: tc, idlejitter, cgroups
114
+ // disable external plugins: all except apps, network_viewer, systemd-journal
115
+ // disable sqlite
116
+ }
117
+ else if(netdata_conf_is_parent()) {
118
+ nd_profile.storage_tiers = 3;
119
+ nd_profile.update_every = 1;
120
+ nd_profile.malloc_arenas = os_get_system_cpus_cached(true);
121
+ nd_profile.malloc_trim = 256 * 1024;
122
+ // web server threads = dynamic
123
+ // aclk query threads = dynamic
124
+ // backfill threads = dynamic
125
+ // replication threads = dynamic
126
+ // ml enabled = true
127
+ // health enabled = true
128
+ }
129
+ else if(netdata_conf_is_child()) {
130
+ nd_profile.storage_tiers = 3;
131
+ nd_profile.update_every = 1;
132
+ nd_profile.malloc_arenas = 1;
133
+ nd_profile.malloc_trim = 32 * 1024;
134
+ // web server threads = 6
135
+ // aclk query threads = 6
136
+ // backfill threads = 0
137
+ // replication threads = 1
138
+ // ml enabled = false
139
+ // health enabled = false
140
+ }
141
+ else /* if(netdata_conf_is_standalone()) */ {
142
+ nd_profile.storage_tiers = 3;
143
+ nd_profile.update_every = 1;
144
+ nd_profile.malloc_arenas = 1;
145
+ nd_profile.malloc_trim = 64 * 1024;
146
+ // web server threads = 6
147
+ // aclk query threads = 6
148
+ // backfill threads = 0
149
+ // replication threads = 1 // can we disable them completely?
150
+ // ml enabled = true
151
+ // health enabled = true
152
+ }
153
+
154
+ netdata_conf_glibc_malloc_initialize(nd_profile.malloc_arenas, nd_profile.malloc_trim);
155
+}
src/daemon/config/netdata-conf-profile.h
new
+54
@@ -0,0 +1,54 @@
1
+// SPDX-License-Identifier: GPL-3.0-or-later
2
+
3
+#ifndef NETDATA_NETDATA_CONF_PROFILE_H
4
+#define NETDATA_NETDATA_CONF_PROFILE_H
5
+
6
+#include "libnetdata/libnetdata.h"
7
+
8
+typedef enum {
9
+ ND_PROFILE_NONE = (0),
10
+
11
+ // system profiles
12
+ ND_PROFILE_PARENT = (1 << 30),
13
+ ND_PROFILE_STANDALONE = (1 << 29),
14
+ ND_PROFILE_CHILD = (1 << 28),
15
+ ND_PROFILE_IOT = (1 << 27),
16
+
17
+ // optional attributed to profiles
18
+
19
+} ND_PROFILE;
20
+
21
+#define ND_CONF_PROFILES_SYSTEM (ND_PROFILE_STANDALONE | ND_PROFILE_PARENT | ND_PROFILE_CHILD | ND_PROFILE_IOT)
22
+
23
+BITMAP_STR_DEFINE_FUNCTIONS_EXTERN(ND_PROFILE);
24
+
25
+ND_PROFILE nd_profile_detect_and_configure(bool recheck);
26
+
27
+void nd_profile_setup(void);
28
+
29
+static inline bool netdata_conf_is_iot(void) {
30
+ return (nd_profile_detect_and_configure(false) & ND_CONF_PROFILES_SYSTEM) == ND_PROFILE_IOT;
31
+}
32
+
33
+static inline bool netdata_conf_is_standalone(void) {
34
+ return (nd_profile_detect_and_configure(false) & ND_CONF_PROFILES_SYSTEM) == ND_PROFILE_STANDALONE;
35
+}
36
+
37
+static inline bool netdata_conf_is_child(void) {
38
+ return (nd_profile_detect_and_configure(false) & ND_CONF_PROFILES_SYSTEM) == ND_PROFILE_CHILD;
39
+}
40
+
41
+static inline bool netdata_conf_is_parent(void) {
42
+ return (nd_profile_detect_and_configure(false) & ND_CONF_PROFILES_SYSTEM) == ND_PROFILE_PARENT;
43
+}
44
+
45
+struct nd_profile_t {
46
+ size_t storage_tiers;
47
+ time_t update_every;
48
+ size_t malloc_arenas;
49
+ size_t malloc_trim;
50
+};
51
+
52
+extern struct nd_profile_t nd_profile;
53
+
54
+#endif //NETDATA_NETDATA_CONF_PROFILE_H
src/daemon/config/netdata-conf-web.c
+1
-1
@@ -12,7 +12,7 @@ size_t netdata_conf_web_query_threads(void) {
12
}
13
14
size_t cpus = MIN(netdata_conf_cpus(), 256); // max 256 cores
15
- size_t threads = cpus * (stream_conf_is_parent(false) ? 2 : 1);
15
+ size_t threads = cpus * (netdata_conf_is_parent() ? 2 : 1);
16
threads = MAX(threads, 6);
17
18
threads = config_get_number(CONFIG_SECTION_WEB, "web server threads", threads);
src/daemon/config/netdata-conf.h
+1
@@ -12,6 +12,7 @@ bool netdata_conf_load(char *filename, char overwrite_used, const char **user);
12
#include "netdata-conf-db.h"
13
#include "netdata-conf-directories.h"
14
#include "netdata-conf-global.h"
15
+#include "netdata-conf-profile.h"
16
#include "netdata-conf-logs.h"
17
#include "netdata-conf-web.h"
18
#include "netdata-conf-cloud.h"
src/daemon/daemon-shutdown.c
+5
-5
@@ -81,7 +81,7 @@ static void rrdeng_flush_everything_and_wait(bool wait_flush, bool wait_collecto
81
return;
82
83
nd_log(NDLS_DAEMON, NDLP_INFO, "Flushing DBENGINE dirty pages...");
84
- for (size_t tier = 0; tier < storage_tiers; tier++)
84
+ for (size_t tier = 0; tier < nd_profile.storage_tiers; tier++)
85
rrdeng_quiesce(multidb_ctx[tier]);
86
87
struct pgc_statistics pgc_main_stats = pgc_get_statistics(main_cache);
@@ -94,7 +94,7 @@ static void rrdeng_flush_everything_and_wait(bool wait_flush, bool wait_collecto
94
size_t count = 10;
95
while (running && count) {
96
running = 0;
97
- for (size_t tier = 0; tier < storage_tiers; tier++)
97
+ for (size_t tier = 0; tier < nd_profile.storage_tiers; tier++)
98
running += rrdeng_collectors_running(multidb_ctx[tier]);
99
100
if (running) {
@@ -236,14 +236,14 @@ void netdata_cleanup_and_exit(int ret, const char *action, const char *action_re
236
rrdeng_flush_everything_and_wait(true, true);
237
watcher_step_complete(WATCHER_STEP_ID_WAIT_FOR_DBENGINE_COLLECTORS_TO_FINISH);
238
239
- ND_THREAD *th[storage_tiers];
240
- for (size_t tier = 0; tier < storage_tiers; tier++)
239
+ ND_THREAD *th[nd_profile.storage_tiers];
240
+ for (size_t tier = 0; tier < nd_profile.storage_tiers; tier++)
241
th[tier] = nd_thread_create("rrdeng-exit", NETDATA_THREAD_OPTION_JOINABLE, rrdeng_exit_background, multidb_ctx[tier]);
242
243
// flush anything remaining again - just in case
244
rrdeng_flush_everything_and_wait(true, false);
245
246
- for (size_t tier = 0; tier < storage_tiers; tier++)
246
+ for (size_t tier = 0; tier < nd_profile.storage_tiers; tier++)
247
nd_thread_join(th[tier]);
248
249
rrdeng_enq_cmd(NULL, RRDENG_OPCODE_SHUTDOWN_EVLOOP, NULL, NULL, STORAGE_PRIORITY_INTERNAL_DBENGINE, NULL, NULL);
src/daemon/environment.c
+1
-1
@@ -36,7 +36,7 @@ static const char *verify_or_create_required_private_directory(const char *dir)
36
void set_environment_for_plugins_and_scripts(void) {
37
{
38
char b[16];
39
- snprintfz(b, sizeof(b) - 1, "%d", default_rrd_update_every);
39
+ snprintfz(b, sizeof(b) - 1, "%d", (int)nd_profile.update_every);
40
nd_setenv("NETDATA_UPDATE_EVERY", b, 1);
41
}
42
src/daemon/main.c
+14
-8
@@ -212,10 +212,10 @@ int windows_perflib_dump(const char *key);
212
int unittest_prepare_rrd(const char **user) {
213
netdata_conf_section_global_run_as_user(user);
214
netdata_conf_section_global();
215
- default_rrd_update_every = 1;
215
+ nd_profile.update_every = 1;
216
default_rrd_memory_mode = RRD_MEMORY_MODE_RAM;
217
health_plugin_disable();
218
- storage_tiers = 1;
218
+ nd_profile.storage_tiers = 1;
219
registry_init();
220
if(rrd_init("unittest", NULL, true)) {
221
fprintf(stderr, "rrd_init failed for unittest\n");
@@ -383,6 +383,7 @@ int netdata_main(int argc, char **argv) {
383
if (ctx_unittest()) return 1;
384
if (uuid_unittest()) return 1;
385
if (dyncfg_unittest()) return 1;
386
+ if (unittest_waiting_queue()) return 1;
387
sqlite_library_shutdown();
388
fprintf(stderr, "\n\nALL TESTS PASSED\n\n");
389
return 0;
@@ -398,6 +399,10 @@ int netdata_main(int argc, char **argv) {
399
unittest_running = true;
400
return aral_unittest(10000);
401
}
402
+ else if(strcmp(optarg, "waitqtest") == 0) {
403
+ unittest_running = true;
404
+ return unittest_waiting_queue();
405
+ }
406
else if(strcmp(optarg, "stringtest") == 0) {
407
unittest_running = true;
408
return string_unittest(10000);
@@ -463,7 +468,7 @@ int netdata_main(int argc, char **argv) {
468
unsigned history_seconds = strtoul(optarg, NULL, 0);
469
netdata_conf_section_global_run_as_user(&user);
470
netdata_conf_section_global();
466
- default_rrd_update_every = 1;
471
+ nd_profile.update_every = 1;
472
registry_init();
473
if(rrd_init("dbengine-dataset", NULL, true)) {
474
fprintf(stderr, "rrd_init failed for unittest\n");
@@ -768,12 +773,14 @@ int netdata_main(int argc, char **argv) {
773
workers_utilization_enable();
774
775
// ----------------------------------------------------------------------------------------------------------------
771
- // streaming, replication, backfilling
776
+ // profiles
777
773
- stream_conf_load();
774
- check_local_streaming_capabilities();
775
- replication_initialize();
778
+ nd_profile_setup();
779
780
+ // ----------------------------------------------------------------------------------------------------------------
781
+ // streaming, replication, functions initialization
782
+
783
+ replication_initialize();
784
rrd_functions_inflight_init();
785
786
{
@@ -967,7 +974,6 @@ int netdata_main(int argc, char **argv) {
974
delta_startup_time("start the static threads");
975
976
netdata_conf_section_web();
970
- backfill_threads_detect_from_stream_conf();
977
978
set_late_analytics_variables(system_info);
979
for (i = 0; static_threads[i].name != NULL ; i++) {
src/daemon/pulse/pulse-aral.c
+1
-1
@@ -167,7 +167,7 @@ void pulse_aral_do(bool extended) {
167
ai->rd_utilization = rrddim_add(ai->st_utilization, "utilization", NULL, 1, 1000, RRD_ALGORITHM_ABSOLUTE);
168
}
169
170
- rrddim_set_by_pointer(ai->st_utilization, ai->rd_utilization, (collected_number)utilization * 1000LL);
170
+ rrddim_set_by_pointer(ai->st_utilization, ai->rd_utilization, (collected_number)(utilization * 1000.0));
171
rrdset_done(ai->st_utilization);
172
}
173
}
src/daemon/pulse/pulse-daemon-memory.c
+8
-8
@@ -2,7 +2,7 @@
2
3
#define PULSE_INTERNALS 1
4
#include "pulse-daemon-memory.h"
5
-#include "streaming/replication.h"
5
+#include "streaming/stream-replication-sender.h"
6
7
#define dictionary_stats_memory_total(stats) \
8
((stats).memory.dict + (stats).memory.values + (stats).memory.index)
@@ -93,13 +93,13 @@ void pulse_daemon_memory_do(bool extended) {
93
netdata_buffers_statistics.buffers_streaming +
94
netdata_buffers_statistics.cbuffers_streaming +
95
netdata_buffers_statistics.buffers_web +
96
- replication_allocated_buffers() + aral_by_size_free_bytes() + judy_aral_free_bytes();
96
+ replication_sender_allocated_buffers() + aral_by_size_free_bytes() + judy_aral_free_bytes();
97
98
int sqlite3_memory_used_current = 0, sqlite3_memory_used_highwater = 0;
99
sqlite3_status(SQLITE_STATUS_MEMORY_USED, &sqlite3_memory_used_current, &sqlite3_memory_used_highwater, 1);
100
101
- size_t strings = 0;
102
- string_statistics(NULL, NULL, NULL, NULL, NULL, &strings, NULL, NULL);
101
+ size_t strings_memory = 0, strings_index = 0;
102
+ string_statistics(NULL, NULL, NULL, NULL, NULL, &strings_memory, &strings_index, NULL, NULL);
103
104
rrddim_set_by_pointer(st_memory, rd_db_dbengine, (collected_number)pulse_dbengine_total_memory);
105
rrddim_set_by_pointer(st_memory, rd_db_rrd, (collected_number)pulse_rrd_memory_size);
@@ -128,7 +128,7 @@ void pulse_daemon_memory_do(bool extended) {
128
(collected_number)dictionary_stats_memory_total(dictionary_stats_category_functions));
129
130
rrddim_set_by_pointer(st_memory, rd_replication,
131
- (collected_number)dictionary_stats_memory_total(dictionary_stats_category_replication) + (collected_number)replication_allocated_memory());
131
+ (collected_number)dictionary_stats_memory_total(dictionary_stats_category_replication) + (collected_number)replication_sender_allocated_memory());
132
#else
133
uint64_t metadata =
134
aral_by_size_structures_bytes() + aral_by_size_used_bytes() +
@@ -140,7 +140,7 @@ void pulse_daemon_memory_do(bool extended) {
140
dictionary_stats_category_functions.memory.dict + dictionary_stats_category_functions.memory.index +
141
dictionary_stats_category_replication.memory.dict + dictionary_stats_category_replication.memory.index +
142
netdata_buffers_statistics.rrdhost_allocations_size +
143
- replication_allocated_memory();
143
+ replication_sender_allocated_memory();
144
145
rrddim_set_by_pointer(st_memory, rd_metadata, (collected_number)metadata);
146
#endif
@@ -153,7 +153,7 @@ void pulse_daemon_memory_do(bool extended) {
153
(collected_number)pulse_ml_get_current_memory_usage());
154
155
rrddim_set_by_pointer(st_memory, rd_strings,
156
- (collected_number)strings);
156
+ (collected_number)(strings_memory + strings_index));
157
158
rrddim_set_by_pointer(st_memory, rd_streaming,
159
(collected_number)netdata_buffers_statistics.rrdhost_senders + (collected_number)netdata_buffers_statistics.rrdhost_receivers);
@@ -243,7 +243,7 @@ void pulse_daemon_memory_do(bool extended) {
243
rrddim_set_by_pointer(st_memory_buffers, rd_buffers_streaming, (collected_number)netdata_buffers_statistics.buffers_streaming);
244
rrddim_set_by_pointer(st_memory_buffers, rd_cbuffers_streaming, (collected_number)netdata_buffers_statistics.cbuffers_streaming);
245
rrddim_set_by_pointer(st_memory_buffers, rd_buffers_web, (collected_number)netdata_buffers_statistics.buffers_web);
246
- rrddim_set_by_pointer(st_memory_buffers, rd_buffers_replication, (collected_number)replication_allocated_buffers());
246
+ rrddim_set_by_pointer(st_memory_buffers, rd_buffers_replication, (collected_number)replication_sender_allocated_buffers());
247
rrddim_set_by_pointer(st_memory_buffers, rd_buffers_aral, (collected_number)aral_by_size_free_bytes());
248
rrddim_set_by_pointer(st_memory_buffers, rd_buffers_judy, (collected_number)judy_aral_free_bytes());
249
src/daemon/pulse/pulse-db-dbengine.c
+140
-38
@@ -7,6 +7,13 @@ size_t pulse_dbengine_total_memory = 0;
7
8
#if defined(ENABLE_DBENGINE)
9
10
+static usec_t time_and_count_delta_average(struct time_and_count *prev, struct time_and_count *latest) {
11
+ if(latest->count > prev->count && latest->usec > prev->usec)
12
+ return (latest->usec - prev->usec) / (latest->count - prev->count);
13
+
14
+ return 0;
15
+}
16
+
17
struct dbengine2_cache_pointers {
18
RRDSET *st_cache_hit_ratio;
19
RRDDIM *rd_hit_ratio_closest;
@@ -573,18 +580,18 @@ static void dbengine2_cache_statistics_charts(struct dbengine2_cache_pointers *p
580
priority++;
581
}
582
576
- rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_evict_relocated, (collected_number)pgc_stats->waste_evict_relocated);
577
- rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_flushes_cancelled, (collected_number)pgc_stats->waste_flushes_cancelled);
578
- rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_acquire_spins, (collected_number)pgc_stats->waste_acquire_spins);
579
- rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_release_spins, (collected_number)pgc_stats->waste_release_spins);
580
- rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_insert_spins, (collected_number)pgc_stats->waste_insert_spins);
581
- rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_delete_spins, (collected_number)pgc_stats->waste_delete_spins);
582
- rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_evict_spins, (collected_number)pgc_stats->waste_evict_useless_spins);
583
- rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_evict_thread_signals, (collected_number)pgc_stats->waste_evict_thread_signals);
584
- rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_evict_inline_on_add, (collected_number)pgc_stats->waste_evictions_inline_on_add);
585
- rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_evict_inline_on_release, (collected_number)pgc_stats->waste_evictions_inline_on_release);
586
- rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_flush_inline_on_add, (collected_number)pgc_stats->waste_flush_on_add);
587
- rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_flush_inline_on_release, (collected_number)pgc_stats->waste_flush_on_release);
583
+ rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_evict_relocated, (collected_number)pgc_stats->p2_waste_evict_relocated);
584
+ rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_flushes_cancelled, (collected_number)pgc_stats->p2_waste_flushes_cancelled);
585
+ rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_acquire_spins, (collected_number)pgc_stats->p2_waste_acquire_spins);
586
+ rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_release_spins, (collected_number)pgc_stats->p2_waste_release_spins);
587
+ rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_insert_spins, (collected_number)pgc_stats->p2_waste_insert_spins);
588
+ rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_delete_spins, (collected_number)pgc_stats->p2_waste_delete_spins);
589
+ rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_evict_spins, (collected_number)pgc_stats->p2_waste_evict_useless_spins);
590
+ rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_evict_thread_signals, (collected_number)pgc_stats->p2_waste_evict_thread_signals);
591
+ rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_evict_inline_on_add, (collected_number)pgc_stats->p2_waste_evictions_inline_on_add);
592
+ rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_evict_inline_on_release, (collected_number)pgc_stats->p2_waste_evictions_inline_on_release);
593
+ rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_flush_inline_on_add, (collected_number)pgc_stats->p2_waste_flush_on_add);
594
+ rrddim_set_by_pointer(ptrs->st_pgc_waste, ptrs->rd_pgc_waste_flush_inline_on_release, (collected_number)pgc_stats->p2_waste_flush_on_release);
595
596
rrdset_done(ptrs->st_pgc_waste);
597
}
@@ -627,12 +634,12 @@ static void dbengine2_cache_statistics_charts(struct dbengine2_cache_pointers *p
634
priority++;
635
}
636
630
- rrddim_set_by_pointer(ptrs->st_pgc_workers, ptrs->rd_pgc_workers_searchers, (collected_number)pgc_stats->workers_search);
631
- rrddim_set_by_pointer(ptrs->st_pgc_workers, ptrs->rd_pgc_workers_adders, (collected_number)pgc_stats->workers_add);
632
- rrddim_set_by_pointer(ptrs->st_pgc_workers, ptrs->rd_pgc_workers_evictors, (collected_number)pgc_stats->workers_evict);
633
- rrddim_set_by_pointer(ptrs->st_pgc_workers, ptrs->rd_pgc_workers_flushers, (collected_number)pgc_stats->workers_flush);
634
- rrddim_set_by_pointer(ptrs->st_pgc_workers, ptrs->rd_pgc_workers_hot2dirty, (collected_number)pgc_stats->workers_hot2dirty);
635
- rrddim_set_by_pointer(ptrs->st_pgc_workers, ptrs->rd_pgc_workers_jv2_flushers, (collected_number)pgc_stats->workers_jv2_flush);
637
+ rrddim_set_by_pointer(ptrs->st_pgc_workers, ptrs->rd_pgc_workers_searchers, (collected_number)pgc_stats->p2_workers_search);
638
+ rrddim_set_by_pointer(ptrs->st_pgc_workers, ptrs->rd_pgc_workers_adders, (collected_number)pgc_stats->p2_workers_add);
639
+ rrddim_set_by_pointer(ptrs->st_pgc_workers, ptrs->rd_pgc_workers_evictors, (collected_number)pgc_stats->p0_workers_evict);
640
+ rrddim_set_by_pointer(ptrs->st_pgc_workers, ptrs->rd_pgc_workers_flushers, (collected_number)pgc_stats->p2_workers_flush);
641
+ rrddim_set_by_pointer(ptrs->st_pgc_workers, ptrs->rd_pgc_workers_hot2dirty, (collected_number)pgc_stats->p2_workers_hot2dirty);
642
+ rrddim_set_by_pointer(ptrs->st_pgc_workers, ptrs->rd_pgc_workers_jv2_flushers, (collected_number)pgc_stats->p2_workers_jv2_flush);
643
644
rrdset_done(ptrs->st_pgc_workers);
645
}
@@ -1012,9 +1019,9 @@ void pulse_dbengine_do(bool extended) {
1019
}
1020
priority++;
1021
1015
- rrddim_set_by_pointer(st_queries, rd_total, (collected_number)cache_efficiency_stats.queries);
1016
- rrddim_set_by_pointer(st_queries, rd_open, (collected_number)cache_efficiency_stats.queries_open);
1017
- rrddim_set_by_pointer(st_queries, rd_jv2, (collected_number)cache_efficiency_stats.queries_journal_v2);
1022
+ rrddim_set_by_pointer(st_queries, rd_total, (collected_number)cache_efficiency_stats.prep_time_in_main_cache_lookup.count);
1023
+ rrddim_set_by_pointer(st_queries, rd_open, (collected_number)cache_efficiency_stats.prep_time_in_open_cache_lookup.count);
1024
+ rrddim_set_by_pointer(st_queries, rd_jv2, (collected_number)cache_efficiency_stats.prep_time_in_journal_v2_lookup.count);
1025
rrddim_set_by_pointer(st_queries, rd_planned_with_gaps, (collected_number)cache_efficiency_stats.queries_planned_with_gaps);
1026
rrddim_set_by_pointer(st_queries, rd_executed_with_gaps, (collected_number)cache_efficiency_stats.queries_executed_with_gaps);
1027
@@ -1303,7 +1310,9 @@ void pulse_dbengine_do(bool extended) {
1310
1311
{
1312
static RRDSET *st_prep_timings = NULL;
1306
- static RRDDIM *rd_routing = NULL;
1313
+ static RRDDIM *rd_routing_sync = NULL;
1314
+ static RRDDIM *rd_routing_syncfirst = NULL;
1315
+ static RRDDIM *rd_routing_async = NULL;
1316
static RRDDIM *rd_main_cache = NULL;
1317
static RRDDIM *rd_open_cache = NULL;
1318
static RRDDIM *rd_journal_v2 = NULL;
@@ -1316,7 +1325,7 @@ void pulse_dbengine_do(bool extended) {
1325
NULL,
1326
"dbengine query router",
1327
NULL,
1319
- "Netdata Query Preparation Timings",
1328
+ "Netdata Query Planning Timings",
1329
"usec/s",
1330
"netdata",
1331
"pulse",
@@ -1324,7 +1333,9 @@ void pulse_dbengine_do(bool extended) {
1333
localhost->rrd_update_every,
1334
RRDSET_TYPE_STACKED);
1335
1327
- rd_routing = rrddim_add(st_prep_timings, "routing", NULL, 1, 1, RRD_ALGORITHM_INCREMENTAL);
1336
+ rd_routing_sync = rrddim_add(st_prep_timings, "pdc sync", NULL, 1, 1, RRD_ALGORITHM_INCREMENTAL);
1337
+ rd_routing_syncfirst = rrddim_add(st_prep_timings, "pdc syncfirst", NULL, 1, 1, RRD_ALGORITHM_INCREMENTAL);
1338
+ rd_routing_async = rrddim_add(st_prep_timings, "pdc async", NULL, 1, 1, RRD_ALGORITHM_INCREMENTAL);
1339
rd_main_cache = rrddim_add(st_prep_timings, "main cache", NULL, 1, 1, RRD_ALGORITHM_INCREMENTAL);
1340
rd_open_cache = rrddim_add(st_prep_timings, "open cache", NULL, 1, 1, RRD_ALGORITHM_INCREMENTAL);
1341
rd_journal_v2 = rrddim_add(st_prep_timings, "journal v2", NULL, 1, 1, RRD_ALGORITHM_INCREMENTAL);
@@ -1332,11 +1343,59 @@ void pulse_dbengine_do(bool extended) {
1343
}
1344
priority++;
1345
1335
- rrddim_set_by_pointer(st_prep_timings, rd_routing, (collected_number)cache_efficiency_stats.prep_time_to_route);
1336
- rrddim_set_by_pointer(st_prep_timings, rd_main_cache, (collected_number)cache_efficiency_stats.prep_time_in_main_cache_lookup);
1337
- rrddim_set_by_pointer(st_prep_timings, rd_open_cache, (collected_number)cache_efficiency_stats.prep_time_in_open_cache_lookup);
1338
- rrddim_set_by_pointer(st_prep_timings, rd_journal_v2, (collected_number)cache_efficiency_stats.prep_time_in_journal_v2_lookup);
1339
- rrddim_set_by_pointer(st_prep_timings, rd_pass4, (collected_number)cache_efficiency_stats.prep_time_in_pass4_lookup);
1346
+ rrddim_set_by_pointer(st_prep_timings, rd_routing_sync, (collected_number)cache_efficiency_stats.prep_time_to_route_sync.usec);
1347
+ rrddim_set_by_pointer(st_prep_timings, rd_routing_syncfirst, (collected_number)cache_efficiency_stats.prep_time_to_route_syncfirst.usec);
1348
+ rrddim_set_by_pointer(st_prep_timings, rd_routing_async, (collected_number)cache_efficiency_stats.prep_time_to_route_async.usec);
1349
+ rrddim_set_by_pointer(st_prep_timings, rd_main_cache, (collected_number)cache_efficiency_stats.prep_time_in_main_cache_lookup.usec);
1350
+ rrddim_set_by_pointer(st_prep_timings, rd_open_cache, (collected_number)cache_efficiency_stats.prep_time_in_open_cache_lookup.usec);
1351
+ rrddim_set_by_pointer(st_prep_timings, rd_journal_v2, (collected_number)cache_efficiency_stats.prep_time_in_journal_v2_lookup.usec);
1352
+ rrddim_set_by_pointer(st_prep_timings, rd_pass4, (collected_number)cache_efficiency_stats.prep_time_in_pass4_lookup.usec);
1353
+
1354
+ rrdset_done(st_prep_timings);
1355
+ }
1356
+
1357
+ {
1358
+ static RRDSET *st_prep_timings = NULL;
1359
+ static RRDDIM *rd_routing_sync = NULL;
1360
+ static RRDDIM *rd_routing_syncfirst = NULL;
1361
+ static RRDDIM *rd_routing_async = NULL;
1362
+ static RRDDIM *rd_main_cache = NULL;
1363
+ static RRDDIM *rd_open_cache = NULL;
1364
+ static RRDDIM *rd_journal_v2 = NULL;
1365
+ static RRDDIM *rd_pass4 = NULL;
1366
+
1367
+ if (unlikely(!st_prep_timings)) {
1368
+ st_prep_timings = rrdset_create_localhost(
1369
+ "netdata",
1370
+ "dbengine_prep_average_timings",
1371
+ NULL,
1372
+ "dbengine query router",
1373
+ NULL,
1374
+ "Netdata Query Planning Average Timings",
1375
+ "usec/s",
1376
+ "netdata",
1377
+ "pulse",
1378
+ priority + 1,
1379
+ localhost->rrd_update_every,
1380
+ RRDSET_TYPE_STACKED);
1381
+
1382
+ rd_routing_sync = rrddim_add(st_prep_timings, "pdc sync", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
1383
+ rd_routing_syncfirst = rrddim_add(st_prep_timings, "pdc syncfirst", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
1384
+ rd_routing_async = rrddim_add(st_prep_timings, "pdc async", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
1385
+ rd_main_cache = rrddim_add(st_prep_timings, "main cache", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
1386
+ rd_open_cache = rrddim_add(st_prep_timings, "open cache", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
1387
+ rd_journal_v2 = rrddim_add(st_prep_timings, "journal v2", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
1388
+ rd_pass4 = rrddim_add(st_prep_timings, "pass4", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
1389
+ }
1390
+ priority++;
1391
+
1392
+ rrddim_set_by_pointer(st_prep_timings,rd_routing_sync, (collected_number)time_and_count_delta_average(&cache_efficiency_stats_old.prep_time_to_route_sync, &cache_efficiency_stats.prep_time_to_route_sync));
1393
+ rrddim_set_by_pointer(st_prep_timings,rd_routing_syncfirst, (collected_number)time_and_count_delta_average(&cache_efficiency_stats_old.prep_time_to_route_syncfirst, &cache_efficiency_stats.prep_time_to_route_syncfirst));
1394
+ rrddim_set_by_pointer(st_prep_timings,rd_routing_async, (collected_number)time_and_count_delta_average(&cache_efficiency_stats_old.prep_time_to_route_async, &cache_efficiency_stats.prep_time_to_route_async));
1395
+ rrddim_set_by_pointer(st_prep_timings, rd_main_cache, (collected_number)time_and_count_delta_average(&cache_efficiency_stats_old.prep_time_in_main_cache_lookup, &cache_efficiency_stats.prep_time_in_main_cache_lookup));
1396
+ rrddim_set_by_pointer(st_prep_timings, rd_open_cache, (collected_number)time_and_count_delta_average(&cache_efficiency_stats_old.prep_time_in_open_cache_lookup, &cache_efficiency_stats.prep_time_in_open_cache_lookup));
1397
+ rrddim_set_by_pointer(st_prep_timings, rd_journal_v2, (collected_number)time_and_count_delta_average(&cache_efficiency_stats_old.prep_time_in_journal_v2_lookup, &cache_efficiency_stats.prep_time_in_journal_v2_lookup));
1398
+ rrddim_set_by_pointer(st_prep_timings, rd_pass4, (collected_number)time_and_count_delta_average(&cache_efficiency_stats_old.prep_time_in_pass4_lookup, &cache_efficiency_stats.prep_time_in_pass4_lookup));
1399
1400
rrdset_done(st_prep_timings);
1401
}
@@ -1365,8 +1424,8 @@ void pulse_dbengine_do(bool extended) {
1424
localhost->rrd_update_every,
1425
RRDSET_TYPE_STACKED);
1426
1368
- rd_init = rrddim_add(st_query_timings, "init", NULL, 1, 1, RRD_ALGORITHM_INCREMENTAL);
1369
- rd_prep_wait = rrddim_add(st_query_timings, "prep wait", NULL, 1, 1, RRD_ALGORITHM_INCREMENTAL);
1427
+ rd_init = rrddim_add(st_query_timings, "plan", NULL, 1, 1, RRD_ALGORITHM_INCREMENTAL);
1428
+ rd_prep_wait = rrddim_add(st_query_timings, "async wait", NULL, 1, 1, RRD_ALGORITHM_INCREMENTAL);
1429
rd_next_page_disk_fast = rrddim_add(st_query_timings, "next page disk fast", NULL, 1, 1, RRD_ALGORITHM_INCREMENTAL);
1430
rd_next_page_disk_slow = rrddim_add(st_query_timings, "next page disk slow", NULL, 1, 1, RRD_ALGORITHM_INCREMENTAL);
1431
rd_next_page_preload_fast = rrddim_add(st_query_timings, "next page preload fast", NULL, 1, 1, RRD_ALGORITHM_INCREMENTAL);
@@ -1374,16 +1433,59 @@ void pulse_dbengine_do(bool extended) {
1433
}
1434
priority++;
1435
1377
- rrddim_set_by_pointer(st_query_timings, rd_init, (collected_number)cache_efficiency_stats.query_time_init);
1378
- rrddim_set_by_pointer(st_query_timings, rd_prep_wait, (collected_number)cache_efficiency_stats.query_time_wait_for_prep);
1379
- rrddim_set_by_pointer(st_query_timings, rd_next_page_disk_fast, (collected_number)cache_efficiency_stats.query_time_to_fast_disk_next_page);
1380
- rrddim_set_by_pointer(st_query_timings, rd_next_page_disk_slow, (collected_number)cache_efficiency_stats.query_time_to_slow_disk_next_page);
1381
- rrddim_set_by_pointer(st_query_timings, rd_next_page_preload_fast, (collected_number)cache_efficiency_stats.query_time_to_fast_preload_next_page);
1382
- rrddim_set_by_pointer(st_query_timings, rd_next_page_preload_slow, (collected_number)cache_efficiency_stats.query_time_to_slow_preload_next_page);
1436
+ rrddim_set_by_pointer(st_query_timings, rd_init, (collected_number)cache_efficiency_stats.query_time_init.usec);
1437
+ rrddim_set_by_pointer(st_query_timings, rd_prep_wait, (collected_number)cache_efficiency_stats.query_time_wait_for_prep.usec);
1438
+ rrddim_set_by_pointer(st_query_timings, rd_next_page_disk_fast, (collected_number)cache_efficiency_stats.query_time_to_fast_disk_next_page.usec);
1439
+ rrddim_set_by_pointer(st_query_timings, rd_next_page_disk_slow, (collected_number)cache_efficiency_stats.query_time_to_slow_disk_next_page.usec);
1440
+ rrddim_set_by_pointer(st_query_timings, rd_next_page_preload_fast, (collected_number)cache_efficiency_stats.query_time_to_fast_preload_next_page.usec);
1441
+ rrddim_set_by_pointer(st_query_timings, rd_next_page_preload_slow, (collected_number)cache_efficiency_stats.query_time_to_slow_preload_next_page.usec);
1442
1443
rrdset_done(st_query_timings);
1444
}
1445
1446
+ {
1447
+ static RRDSET *st_query_timings_average = NULL;
1448
+ static RRDDIM *rd_init = NULL;
1449
+ static RRDDIM *rd_prep_wait = NULL;
1450
+ static RRDDIM *rd_next_page_disk_fast = NULL;
1451
+ static RRDDIM *rd_next_page_disk_slow = NULL;
1452
+ static RRDDIM *rd_next_page_preload_fast = NULL;
1453
+ static RRDDIM *rd_next_page_preload_slow = NULL;
1454
+
1455
+ if (unlikely(!st_query_timings_average)) {
1456
+ st_query_timings_average = rrdset_create_localhost(
1457
+ "netdata",
1458
+ "dbengine_query_timings_average",
1459
+ NULL,
1460
+ "dbengine query router",
1461
+ NULL,
1462
+ "Netdata Query Average Timings",
1463
+ "usec/s",
1464
+ "netdata",
1465
+ "pulse",
1466
+ priority + 1,
1467
+ localhost->rrd_update_every,
1468
+ RRDSET_TYPE_STACKED);
1469
+
1470
+ rd_init = rrddim_add(st_query_timings_average, "plan", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
1471
+ rd_prep_wait = rrddim_add(st_query_timings_average, "async wait", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
1472
+ rd_next_page_disk_fast = rrddim_add(st_query_timings_average, "next page disk fast", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
1473
+ rd_next_page_disk_slow = rrddim_add(st_query_timings_average, "next page disk slow", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
1474
+ rd_next_page_preload_fast = rrddim_add(st_query_timings_average, "next page preload fast", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
1475
+ rd_next_page_preload_slow = rrddim_add(st_query_timings_average, "next page preload slow", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
1476
+ }
1477
+ priority++;
1478
+
1479
+ rrddim_set_by_pointer(st_query_timings_average, rd_init, (collected_number)time_and_count_delta_average(&cache_efficiency_stats_old.query_time_init, &cache_efficiency_stats.query_time_init));
1480
+ rrddim_set_by_pointer(st_query_timings_average, rd_prep_wait, (collected_number)time_and_count_delta_average(&cache_efficiency_stats_old.query_time_wait_for_prep, &cache_efficiency_stats.query_time_wait_for_prep));
1481
+ rrddim_set_by_pointer(st_query_timings_average, rd_next_page_disk_fast, (collected_number)time_and_count_delta_average(&cache_efficiency_stats_old.query_time_to_fast_disk_next_page, &cache_efficiency_stats.query_time_to_fast_disk_next_page));
1482
+ rrddim_set_by_pointer(st_query_timings_average, rd_next_page_disk_slow, (collected_number)time_and_count_delta_average(&cache_efficiency_stats_old.query_time_to_slow_disk_next_page, &cache_efficiency_stats.query_time_to_slow_disk_next_page));
1483
+ rrddim_set_by_pointer(st_query_timings_average, rd_next_page_preload_fast, (collected_number)time_and_count_delta_average(&cache_efficiency_stats_old.query_time_to_fast_preload_next_page, &cache_efficiency_stats.query_time_to_fast_preload_next_page));
1484
+ rrddim_set_by_pointer(st_query_timings_average, rd_next_page_preload_slow, (collected_number)time_and_count_delta_average(&cache_efficiency_stats_old.query_time_to_slow_preload_next_page, &cache_efficiency_stats.query_time_to_slow_preload_next_page));
1485
+
1486
+ rrdset_done(st_query_timings_average);
1487
+ }
1488
+
1489
if(netdata_rwlock_tryrdlock(&rrd_rwlock) == 0) {
1490
priority = 135400;
1491
@@ -1396,7 +1498,7 @@ void pulse_dbengine_do(bool extended) {
1498
if (!rrdhost_flag_check(host, RRDHOST_FLAG_ARCHIVED)) {
1499
1500
/* get localhost's DB engine's statistics for each tier */
1399
- for(size_t tier = 0; tier < storage_tiers ;tier++) {
1501
+ for(size_t tier = 0; tier < nd_profile.storage_tiers;tier++) {
1502
if(host->db[tier].mode != RRD_MEMORY_MODE_DBENGINE) continue;
1503
if(!host->db[tier].si) continue;
1504
src/daemon/pulse/pulse-gorilla.c
+5
-9
@@ -6,15 +6,11 @@
6
static struct gorilla_statistics {
7
bool enabled;
8
9
- CACHE_LINE_PADDING();
10
- uint64_t tier0_hot_gorilla_buffers;
11
-
12
- CACHE_LINE_PADDING();
13
- uint64_t gorilla_tier0_disk_actual_bytes;
14
- CACHE_LINE_PADDING();
15
- uint64_t gorilla_tier0_disk_optimal_bytes;
16
- CACHE_LINE_PADDING();
17
- uint64_t gorilla_tier0_disk_original_bytes;
9
+ PAD64(uint64_t) tier0_hot_gorilla_buffers;
10
+
11
+ PAD64(uint64_t) gorilla_tier0_disk_actual_bytes;
12
+ PAD64(uint64_t) gorilla_tier0_disk_optimal_bytes;
13
+ PAD64(uint64_t) gorilla_tier0_disk_original_bytes;
14
} gorilla_statistics = { 0 };
15
16
void pulse_gorilla_hot_buffer_added() {
src/daemon/pulse/pulse-http-api.c
+11
-20
@@ -8,26 +8,17 @@
8
static struct web_statistics {
9
bool extended;
10
11
- CACHE_LINE_PADDING();
12
- uint16_t connected_clients;
13
- CACHE_LINE_PADDING();
14
- uint64_t web_client_count; // oops! this is used for giving unique IDs to web_clients!
15
-
16
- CACHE_LINE_PADDING();
17
- uint64_t web_requests;
18
- CACHE_LINE_PADDING();
19
- uint64_t web_usec;
20
- CACHE_LINE_PADDING();
21
- uint64_t web_usec_max;
22
- CACHE_LINE_PADDING();
23
- uint64_t bytes_received;
24
- CACHE_LINE_PADDING();
25
- uint64_t bytes_sent;
26
-
27
- CACHE_LINE_PADDING();
28
- uint64_t content_size_uncompressed;
29
- CACHE_LINE_PADDING();
30
- uint64_t content_size_compressed;
11
+ PAD64(uint16_t) connected_clients;
12
+ PAD64(uint64_t) web_client_count; // oops! this is used for giving unique IDs to web_clients!
13
+
14
+ PAD64(uint64_t) web_requests;
15
+ PAD64(uint64_t) web_usec;
16
+ PAD64(uint64_t) web_usec_max;
17
+ PAD64(uint64_t) bytes_received;
18
+ PAD64(uint64_t) bytes_sent;
19
+
20
+ PAD64(uint64_t) content_size_uncompressed;
21
+ PAD64(uint64_t) content_size_compressed;
22
} web_statistics;
23
24
uint64_t pulse_web_client_connected(void) {
src/daemon/pulse/pulse-ingestion.c
+4
-4
@@ -8,14 +8,14 @@ static struct ingest_statistics {
8
} ingest_statistics;
9
10
void pulse_queries_rrdset_collection_completed(size_t *points_read_per_tier_array) {
11
- for(size_t tier = 0; tier < storage_tiers ;tier++) {
11
+ for(size_t tier = 0; tier < nd_profile.storage_tiers;tier++) {
12
__atomic_fetch_add(&ingest_statistics.db_points_stored_per_tier[tier], points_read_per_tier_array[tier], __ATOMIC_RELAXED);
13
points_read_per_tier_array[tier] = 0;
14
}
15
}
16
17
static inline void pulse_ingestion_copy(struct ingest_statistics *gs) {
18
- for(size_t tier = 0; tier < storage_tiers ;tier++)
18
+ for(size_t tier = 0; tier < nd_profile.storage_tiers;tier++)
19
gs->db_points_stored_per_tier[tier] = __atomic_load_n(&ingest_statistics.db_points_stored_per_tier[tier], __ATOMIC_RELAXED);
20
}
21
@@ -43,14 +43,14 @@ void pulse_ingestion_do(bool extended __maybe_unused) {
43
, RRDSET_TYPE_STACKED
44
);
45
46
- for(size_t tier = 0; tier < storage_tiers ;tier++) {
46
+ for(size_t tier = 0; tier < nd_profile.storage_tiers;tier++) {
47
char buf[30 + 1];
48
snprintfz(buf, sizeof(buf) - 1, "tier%zu", tier);
49
rds[tier] = rrddim_add(st_points_stored, buf, NULL, 1, 1, RRD_ALGORITHM_INCREMENTAL);
50
}
51
}
52
53
- for(size_t tier = 0; tier < storage_tiers ;tier++)
53
+ for(size_t tier = 0; tier < nd_profile.storage_tiers;tier++)
54
rrddim_set_by_pointer(st_points_stored, rds[tier], (collected_number)gs.db_points_stored_per_tier[tier]);
55
56
rrdset_done(st_points_stored);
src/daemon/pulse/pulse-ml.c
+8
-16
@@ -4,22 +4,14 @@
4
#include "pulse-ml.h"
5
6
static struct ml_statistics {
7
- CACHE_LINE_PADDING();
8
- uint64_t ml_models_consulted;
9
- CACHE_LINE_PADDING();
10
- uint64_t ml_models_received;
11
- CACHE_LINE_PADDING();
12
- uint64_t ml_models_ignored;
13
- CACHE_LINE_PADDING();
14
- uint64_t ml_models_sent;
15
- CACHE_LINE_PADDING();
16
- uint64_t ml_models_deserialization_failures;
17
- CACHE_LINE_PADDING();
18
- uint64_t ml_memory_consumption;
19
- CACHE_LINE_PADDING();
20
- uint64_t ml_memory_new;
21
- CACHE_LINE_PADDING();
22
- uint64_t ml_memory_delete;
7
+ PAD64(uint64_t) ml_models_consulted;
8
+ PAD64(uint64_t) ml_models_received;
9
+ PAD64(uint64_t) ml_models_ignored;
10
+ PAD64(uint64_t) ml_models_sent;
11
+ PAD64(uint64_t) ml_models_deserialization_failures;
12
+ PAD64(uint64_t) ml_memory_consumption;
13
+ PAD64(uint64_t) ml_memory_new;
14
+ PAD64(uint64_t) ml_memory_delete;
15
} ml_statistics = {0};
16
17
void pulse_ml_models_received()
src/daemon/pulse/pulse-queries.c
+26
-45
@@ -2,53 +2,34 @@
2
3
#define PULSE_INTERNALS 1
4
#include "pulse-queries.h"
5
-#include "streaming/replication.h"
5
+#include "streaming/stream-replication-sender.h"
6
7
static struct query_statistics {
8
- CACHE_LINE_PADDING();
9
- uint64_t api_data_queries_made;
10
- CACHE_LINE_PADDING();
11
- uint64_t api_data_db_points_read;
12
- CACHE_LINE_PADDING();
13
- uint64_t api_data_result_points_generated;
14
-
15
- CACHE_LINE_PADDING();
16
- uint64_t api_weights_queries_made;
17
- CACHE_LINE_PADDING();
18
- uint64_t api_weights_db_points_read;
19
- CACHE_LINE_PADDING();
20
- uint64_t api_weights_result_points_generated;
21
-
22
- CACHE_LINE_PADDING();
23
- uint64_t api_badges_queries_made;
24
- CACHE_LINE_PADDING();
25
- uint64_t api_badges_db_points_read;
26
- CACHE_LINE_PADDING();
27
- uint64_t api_badges_result_points_generated;
28
-
29
- CACHE_LINE_PADDING();
30
- uint64_t health_queries_made;
31
- CACHE_LINE_PADDING();
32
- uint64_t health_db_points_read;
33
- CACHE_LINE_PADDING();
34
- uint64_t health_result_points_generated;
35
-
36
- CACHE_LINE_PADDING();
37
- uint64_t ml_queries_made;
38
- CACHE_LINE_PADDING();
39
- uint64_t ml_db_points_read;
40
- CACHE_LINE_PADDING();
41
- uint64_t ml_result_points_generated;
42
-
43
- CACHE_LINE_PADDING();
44
- uint64_t backfill_queries_made;
45
- CACHE_LINE_PADDING();
46
- uint64_t backfill_db_points_read;
47
-
48
- CACHE_LINE_PADDING();
49
- uint64_t exporters_queries_made;
50
- CACHE_LINE_PADDING();
51
- uint64_t exporters_db_points_read;
8
+ PAD64(uint64_t) api_data_queries_made;
9
+ PAD64(uint64_t) api_data_db_points_read;
10
+ PAD64(uint64_t) api_data_result_points_generated;
11
+
12
+ PAD64(uint64_t) api_weights_queries_made;
13
+ PAD64(uint64_t) api_weights_db_points_read;
14
+ PAD64(uint64_t) api_weights_result_points_generated;
15
+
16
+ PAD64(uint64_t) api_badges_queries_made;
17
+ PAD64(uint64_t) api_badges_db_points_read;
18
+ PAD64(uint64_t) api_badges_result_points_generated;
19
+
20
+ PAD64(uint64_t) health_queries_made;
21
+ PAD64(uint64_t) health_db_points_read;
22
+ PAD64(uint64_t) health_result_points_generated;
23
+
24
+ PAD64(uint64_t) ml_queries_made;
25
+ PAD64(uint64_t) ml_db_points_read;
26
+ PAD64(uint64_t) ml_result_points_generated;
27
+
28
+ PAD64(uint64_t) backfill_queries_made;
29
+ PAD64(uint64_t) backfill_db_points_read;
30
+
31
+ PAD64(uint64_t) exporters_queries_made;
32
+ PAD64(uint64_t) exporters_db_points_read;
33
} query_statistics;
34
35
void pulse_queries_ml_query_completed(size_t points_read) {
src/daemon/pulse/pulse-sqlite3.c
+15
-29
@@ -6,34 +6,20 @@
6
static struct sqlite3_statistics {
7
bool enabled;
8
9
- CACHE_LINE_PADDING();
10
- uint64_t sqlite3_queries_made;
11
- CACHE_LINE_PADDING();
12
- uint64_t sqlite3_queries_ok;
13
- CACHE_LINE_PADDING();
14
- uint64_t sqlite3_queries_failed;
15
- CACHE_LINE_PADDING();
16
- uint64_t sqlite3_queries_failed_busy;
17
- CACHE_LINE_PADDING();
18
- uint64_t sqlite3_queries_failed_locked;
19
- CACHE_LINE_PADDING();
20
- uint64_t sqlite3_rows;
21
- CACHE_LINE_PADDING();
22
- uint64_t sqlite3_metadata_cache_hit;
23
- CACHE_LINE_PADDING();
24
- uint64_t sqlite3_context_cache_hit;
25
- CACHE_LINE_PADDING();
26
- uint64_t sqlite3_metadata_cache_miss;
27
- CACHE_LINE_PADDING();
28
- uint64_t sqlite3_context_cache_miss;
29
- CACHE_LINE_PADDING();
30
- uint64_t sqlite3_metadata_cache_spill;
31
- CACHE_LINE_PADDING();
32
- uint64_t sqlite3_context_cache_spill;
33
- CACHE_LINE_PADDING();
34
- uint64_t sqlite3_metadata_cache_write;
35
- CACHE_LINE_PADDING();
36
- uint64_t sqlite3_context_cache_write;
9
+ PAD64(uint64_t) sqlite3_queries_made;
10
+ PAD64(uint64_t) sqlite3_queries_ok;
11
+ PAD64(uint64_t) sqlite3_queries_failed;
12
+ PAD64(uint64_t) sqlite3_queries_failed_busy;
13
+ PAD64(uint64_t) sqlite3_queries_failed_locked;
14
+ PAD64(uint64_t) sqlite3_rows;
15
+ PAD64(uint64_t) sqlite3_metadata_cache_hit;
16
+ PAD64(uint64_t) sqlite3_context_cache_hit;
17
+ PAD64(uint64_t) sqlite3_metadata_cache_miss;
18
+ PAD64(uint64_t) sqlite3_context_cache_miss;
19
+ PAD64(uint64_t) sqlite3_metadata_cache_spill;
20
+ PAD64(uint64_t) sqlite3_context_cache_spill;
21
+ PAD64(uint64_t) sqlite3_metadata_cache_write;
22
+ PAD64(uint64_t) sqlite3_context_cache_write;
23
} sqlite3_statistics = { };
24
25
void pulse_sqlite3_query_completed(bool success, bool busy, bool locked) {
@@ -71,7 +57,7 @@ static inline void sqlite3_statistics_copy(struct sqlite3_statistics *gs) {
57
gs->sqlite3_queries_failed_locked = __atomic_load_n(&sqlite3_statistics.sqlite3_queries_failed_locked, __ATOMIC_RELAXED);
58
gs->sqlite3_rows = __atomic_load_n(&sqlite3_statistics.sqlite3_rows, __ATOMIC_RELAXED);
59
74
- usec_t timeout = default_rrd_update_every * USEC_PER_SEC + default_rrd_update_every * USEC_PER_SEC / 3;
60
+ usec_t timeout = nd_profile.update_every * USEC_PER_SEC + nd_profile.update_every * USEC_PER_SEC / 3;
61
usec_t now = now_monotonic_usec();
62
if(!last_run)
63
last_run = now;
src/daemon/pulse/pulse-string.c
+7
-4
@@ -10,13 +10,14 @@ void pulse_string_do(bool extended) {
10
static RRDDIM *rd_ops_inserts = NULL, *rd_ops_deletes = NULL;
11
static RRDDIM *rd_entries_entries = NULL;
12
static RRDDIM *rd_mem = NULL;
13
+ static RRDDIM *rd_mem_idx = NULL;
14
#ifdef NETDATA_INTERNAL_CHECKS
15
static RRDDIM *rd_entries_refs = NULL, *rd_ops_releases = NULL, *rd_ops_duplications = NULL, *rd_ops_searches = NULL;
16
#endif
17
17
- size_t inserts, deletes, searches, entries, references, memory, duplications, releases;
18
+ size_t inserts, deletes, searches, entries, references, memory, memory_index, duplications, releases;
19
19
- string_statistics(&inserts, &deletes, &searches, &entries, &references, &memory, &duplications, &releases);
20
+ string_statistics(&inserts, &deletes, &searches, &entries, &references, &memory, &memory_index, &duplications, &releases);
21
22
if (unlikely(!st_ops)) {
23
st_ops = rrdset_create_localhost(
@@ -91,11 +92,13 @@ void pulse_string_do(bool extended) {
92
, "pulse"
93
, 910001
94
, localhost->rrd_update_every
94
- , RRDSET_TYPE_AREA);
95
+ , RRDSET_TYPE_STACKED);
96
96
- rd_mem = rrddim_add(st_mem, "memory", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
97
+ rd_mem = rrddim_add(st_mem, "memory", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
98
+ rd_mem_idx = rrddim_add(st_mem, "index", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
99
}
100
101
rrddim_set_by_pointer(st_mem, rd_mem, (collected_number)memory);
102
+ rrddim_set_by_pointer(st_mem, rd_mem_idx, (collected_number)memory_index);
103
rrdset_done(st_mem);
104
}
src/daemon/pulse/pulse-workers.c
+74
-19
@@ -20,11 +20,14 @@ struct worker_job_type_gs {
20
STRING *name;
21
STRING *units;
22
23
- size_t jobs_started;
24
- usec_t busy_time;
23
+ struct {
24
+ size_t jobs_started;
25
+ usec_t busy_time;
26
+ } data[2];
27
28
RRDDIM *rd_jobs_started;
29
RRDDIM *rd_busy_time;
30
+ RRDDIM *rd_avg_time;
31
32
WORKER_METRIC_TYPE type;
33
NETDATA_DOUBLE min_value;
@@ -102,7 +105,8 @@ struct worker_utilization {
105
RRDDIM *rd_workers_threads_busy;
106
107
RRDSET *st_workers_jobs_per_job_type;
105
- RRDSET *st_workers_busy_per_job_type;
108
+ RRDSET *st_workers_time_per_job_type;
109
+ RRDSET *st_workers_avg_time_per_job_type;
110
111
RRDDIM *rd_total_cpu_utilizaton;
112
@@ -403,7 +407,7 @@ static void workers_utilization_update_chart(struct worker_utilization *wu) {
407
}
408
#endif
409
406
- // ----------------------------------------------------------------------
410
+ // ----------------------------------------------------------------------------------------------------------------
411
412
if(unlikely(!wu->st_workers_jobs_per_job_type)) {
413
char name[RRD_ID_LENGTH_MAX + 1];
@@ -439,23 +443,23 @@ static void workers_utilization_update_chart(struct worker_utilization *wu) {
443
if(unlikely(!wu->per_job_type[i].rd_jobs_started))
444
wu->per_job_type[i].rd_jobs_started = rrddim_add(wu->st_workers_jobs_per_job_type, string2str(wu->per_job_type[i].name), NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
445
442
- rrddim_set_by_pointer(wu->st_workers_jobs_per_job_type, wu->per_job_type[i].rd_jobs_started, (collected_number)(wu->per_job_type[i].jobs_started));
446
+ rrddim_set_by_pointer(wu->st_workers_jobs_per_job_type, wu->per_job_type[i].rd_jobs_started, (collected_number)(wu->per_job_type[i].data[0].jobs_started));
447
}
448
}
449
}
450
451
rrdset_done(wu->st_workers_jobs_per_job_type);
452
449
- // ----------------------------------------------------------------------
453
+ // ----------------------------------------------------------------------------------------------------------------
454
451
- if(unlikely(!wu->st_workers_busy_per_job_type)) {
455
+ if(unlikely(!wu->st_workers_time_per_job_type)) {
456
char name[RRD_ID_LENGTH_MAX + 1];
457
snprintfz(name, RRD_ID_LENGTH_MAX, "workers_busy_time_by_type_%s", wu->name_lowercase);
458
459
char context[RRD_ID_LENGTH_MAX + 1];
460
snprintf(context, RRD_ID_LENGTH_MAX, "netdata.workers.%s.time_by_type", wu->name_lowercase);
461
458
- wu->st_workers_busy_per_job_type = rrdset_create_localhost(
462
+ wu->st_workers_time_per_job_type = rrdset_create_localhost(
463
"netdata"
464
, name
465
, NULL
@@ -480,16 +484,62 @@ static void workers_utilization_update_chart(struct worker_utilization *wu) {
484
if (wu->per_job_type[i].name) {
485
486
if(unlikely(!wu->per_job_type[i].rd_busy_time))
483
- wu->per_job_type[i].rd_busy_time = rrddim_add(wu->st_workers_busy_per_job_type, string2str(wu->per_job_type[i].name), NULL, 1, USEC_PER_MS, RRD_ALGORITHM_ABSOLUTE);
487
+ wu->per_job_type[i].rd_busy_time = rrddim_add(wu->st_workers_time_per_job_type, string2str(wu->per_job_type[i].name), NULL, 1, USEC_PER_MS, RRD_ALGORITHM_ABSOLUTE);
488
485
- rrddim_set_by_pointer(wu->st_workers_busy_per_job_type, wu->per_job_type[i].rd_busy_time, (collected_number)(wu->per_job_type[i].busy_time));
489
+ rrddim_set_by_pointer(wu->st_workers_time_per_job_type, wu->per_job_type[i].rd_busy_time, (collected_number)(wu->per_job_type[i].data[0].busy_time));
490
}
491
}
492
}
493
490
- rrdset_done(wu->st_workers_busy_per_job_type);
494
+ rrdset_done(wu->st_workers_time_per_job_type);
495
492
- // ----------------------------------------------------------------------
496
+ // ----------------------------------------------------------------------------------------------------------------
497
+
498
+ if(unlikely(!wu->st_workers_avg_time_per_job_type)) {
499
+ char name[RRD_ID_LENGTH_MAX + 1];
500
+ snprintfz(name, RRD_ID_LENGTH_MAX, "workers_avg_time_by_type_%s", wu->name_lowercase);
501
+
502
+ char context[RRD_ID_LENGTH_MAX + 1];
503
+ snprintf(context, RRD_ID_LENGTH_MAX, "netdata.workers.%s.avg_time_by_type", wu->name_lowercase);
504
+
505
+ wu->st_workers_avg_time_per_job_type = rrdset_create_localhost(
506
+ "netdata"
507
+ , name
508
+ , NULL
509
+ , wu->family
510
+ , context
511
+ , "Netdata Workers Average Time by Type"
512
+ , "ms"
513
+ , "netdata"
514
+ , "pulse"
515
+ , wu->priority + 4
516
+ , localhost->rrd_update_every
517
+ , RRDSET_TYPE_STACKED
518
+ );
519
+ }
520
+
521
+ {
522
+ size_t i;
523
+ for(i = 0; i <= wu->workers_max_job_id ;i++) {
524
+ if(unlikely(wu->per_job_type[i].type != WORKER_METRIC_IDLE_BUSY))
525
+ continue;
526
+
527
+ if (wu->per_job_type[i].name) {
528
+
529
+ if(unlikely(!wu->per_job_type[i].rd_avg_time))
530
+ wu->per_job_type[i].rd_avg_time = rrddim_add(wu->st_workers_avg_time_per_job_type, string2str(wu->per_job_type[i].name), NULL, 1, USEC_PER_MS, RRD_ALGORITHM_ABSOLUTE);
531
+
532
+ ssize_t jobs_delta = (ssize_t)wu->per_job_type[i].data[0].jobs_started;
533
+ susec_t time_delta = (susec_t)wu->per_job_type[i].data[0].busy_time;
534
+ susec_t average = (jobs_delta != 0) ? time_delta / jobs_delta : 0;
535
+ rrddim_set_by_pointer(wu->st_workers_avg_time_per_job_type, wu->per_job_type[i].rd_avg_time, (collected_number)average);
536
+ }
537
+ }
538
+ }
539
+
540
+ rrdset_done(wu->st_workers_avg_time_per_job_type);
541
+
542
+ // ----------------------------------------------------------------------------------------------------------------
543
544
if(wu->st_workers_threads || wu->workers_registered > 1) {
545
if(unlikely(!wu->st_workers_threads)) {
@@ -509,7 +559,7 @@ static void workers_utilization_update_chart(struct worker_utilization *wu) {
559
, "threads"
560
, "netdata"
561
, "pulse"
512
- , wu->priority + 4
562
+ , wu->priority + 5
563
, localhost->rrd_update_every
564
, RRDSET_TYPE_STACKED
565
);
@@ -544,7 +594,7 @@ static void workers_utilization_update_chart(struct worker_utilization *wu) {
594
, "locks"
595
, "netdata"
596
, "pulse"
547
- , wu->priority + 5
597
+ , wu->priority + 6
598
, localhost->rrd_update_every
599
, RRDSET_TYPE_LINE
600
);
@@ -582,7 +632,7 @@ static void workers_utilization_update_chart(struct worker_utilization *wu) {
632
, "spins"
633
, "netdata"
634
, "pulse"
585
- , wu->priority + 6
635
+ , wu->priority + 7
636
, localhost->rrd_update_every
637
, RRDSET_TYPE_LINE
638
);
@@ -745,8 +795,13 @@ static void workers_utilization_reset_statistics(struct worker_utilization *wu)
795
for( ; *s ; s++) *s = tolower(*s);
796
}
797
748
- wu->per_job_type[i].jobs_started = 0;
749
- wu->per_job_type[i].busy_time = 0;
798
+ // copy the usage data
799
+ wu->per_job_type[i].data[1].jobs_started = wu->per_job_type[i].data[0].jobs_started;
800
+ wu->per_job_type[i].data[1].busy_time = wu->per_job_type[i].data[0].busy_time;
801
+
802
+ // reset them for the next collection
803
+ wu->per_job_type[i].data[0].jobs_started = 0;
804
+ wu->per_job_type[i].data[0].busy_time = 0;
805
806
wu->per_job_type[i].min_value = NAN;
807
wu->per_job_type[i].max_value = NAN;
@@ -910,8 +965,8 @@ static void worker_utilization_charts_callback(void *ptr
965
966
wu->per_job_type[i].type = job_types_metric_types[i];
967
913
- wu->per_job_type[i].jobs_started += job_types_jobs_started[i];
914
- wu->per_job_type[i].busy_time += job_types_busy_time[i];
968
+ wu->per_job_type[i].data[0].jobs_started += job_types_jobs_started[i];
969
+ wu->per_job_type[i].data[0].busy_time += job_types_busy_time[i];
970
971
NETDATA_DOUBLE value = job_types_custom_metrics[i];
972
if(netdata_double_isnumber(value)) {
src/daemon/pulse/pulse.h
+33
@@ -28,4 +28,37 @@ extern bool pulse_extended_enabled;
28
void *pulse_thread_main(void *ptr);
29
void *pulse_thread_sqlite3_main(void *ptr);
30
31
+#define p1_add_fetch(variable, value) __atomic_add_fetch(variable, value, __ATOMIC_RELAXED)
32
+#define p1_sub_fetch(variable, value) __atomic_sub_fetch(variable, value, __ATOMIC_RELAXED)
33
+
34
+#define p1_fetch_add(variable, value) __atomic_fetch_add(variable, value, __ATOMIC_RELAXED)
35
+#define p1_fetch_sub(variable, value) __atomic_fetch_sub(variable, value, __ATOMIC_RELAXED)
36
+
37
+#define p1_store(variable, value) __atomic_store_n(variable, value, __ATOMIC_RELAXED)
38
+#define p1_load(variable) __atomic_load_n(variable, value, __ATOMIC_RELAXED)
39
+
40
+#if !defined(PULSE_EXTENDED_STATISTICS) && (defined(NETDATA_INTERNAL_CHECKS) || defined(NETDATA_GOD_MODE))
41
+#define PULSE_EXTENDED_STATISTICS
42
+#endif
43
+
44
+#if defined(PULSE_EXTENDED_STATISTICS)
45
+#define p2_add_fetch(variable, value) __atomic_add_fetch(variable, value, __ATOMIC_RELAXED)
46
+#define p2_sub_fetch(variable, value) __atomic_sub_fetch(variable, value, __ATOMIC_RELAXED)
47
+
48
+#define p2_fetch_add(variable, value) __atomic_fetch_add(variable, value, __ATOMIC_RELAXED)
49
+#define p2_fetch_sub(variable, value) __atomic_fetch_sub(variable, value, __ATOMIC_RELAXED)
50
+
51
+#define p2_store(variable, value) __atomic_store_n(variable, value, __ATOMIC_RELAXED)
52
+#define p2_load(variable) __atomic_load_n(variable, value, __ATOMIC_RELAXED)
53
+#else
54
+#define p2_add_fetch(variable, value) debug_dummy()
55
+#define p2_sub_fetch(variable, value) debug_dummy()
56
+
57
+#define p2_fetch_add(variable, value) debug_dummy()
58
+#define p2_fetch_sub(variable, value) debug_dummy()
59
+
60
+#define p2_store(variable, value) debug_dummy()
61
+#define p2_load(variable) debug_dummy()
62
+#endif
63
+
64
#endif /* NETDATA_PULSE_H */
src/daemon/static_threads.c
+1
-1
@@ -174,7 +174,7 @@ const struct netdata_static_thread static_threads_common[] = {
174
.name = "BACKFILL",
175
.config_section = NULL,
176
.config_name = NULL,
177
- .enable_routine = backfill_threads_detect_from_stream_conf,
177
+ .enable_routine = netdata_conf_is_parent,
178
.enabled = 0,
179
.thread = NULL,
180
.init_routine = NULL,
src/daemon/unit_test.c
+2
-2
@@ -1269,7 +1269,7 @@ int run_test(struct test *test)
1269
fprintf(stderr, "\nRunning test '%s':\n%s\n", test->name, test->description);
1270
1271
default_rrd_memory_mode = RRD_MEMORY_MODE_ALLOC;
1272
- default_rrd_update_every = test->update_every;
1272
+ nd_profile.update_every = test->update_every;
1273
1274
char name[101];
1275
snprintfz(name, sizeof(name) - 1, "unittest-%s", test->name);
@@ -1538,7 +1538,7 @@ int unit_test(long delay, long shift)
1538
1539
//debug_flags = 0xffffffff;
1540
default_rrd_memory_mode = RRD_MEMORY_MODE_ALLOC;
1541
- default_rrd_update_every = 1;
1541
+ nd_profile.update_every = 1;
1542
1543
int do_abs = 1;
1544
int do_inc = 1;
src/database/contexts/api_v2_contexts_agents.c
+1
-1
@@ -104,7 +104,7 @@ void buffer_json_agents_v2(BUFFER *wb, struct query_timings *timings, time_t now
104
105
buffer_json_member_add_array(wb, "db_size");
106
size_t group_seconds;
107
- for (size_t tier = 0; tier < storage_tiers; tier++) {
107
+ for (size_t tier = 0; tier < nd_profile.storage_tiers; tier++) {
108
STORAGE_ENGINE *eng = localhost->db[tier].eng;
109
if (!eng) continue;
110
src/database/contexts/query_target.c
+6
-6
@@ -219,7 +219,7 @@ static inline void query_metric_release(QUERY_TARGET *qt, QUERY_METRIC *qm) {
219
qm->plan.used = 0;
220
221
// reset the tiers
222
- for(size_t tier = 0; tier < storage_tiers ;tier++) {
222
+ for(size_t tier = 0; tier < nd_profile.storage_tiers;tier++) {
223
if(qm->tiers[tier].smh) {
224
STORAGE_ENGINE *eng = query_metric_storage_engine(qt, qm, tier);
225
eng->api.metric_release(qm->tiers[tier].smh);
@@ -244,9 +244,9 @@ static bool query_metric_add(QUERY_TARGET_LOCALS *qtl, QUERY_NODE *qn, QUERY_CON
244
time_t db_first_time_s;
245
time_t db_last_time_s;
246
time_t db_update_every_s;
247
- } tier_retention[storage_tiers];
247
+ } tier_retention[nd_profile.storage_tiers];
248
249
- for (size_t tier = 0; tier < storage_tiers; tier++) {
249
+ for (size_t tier = 0; tier < nd_profile.storage_tiers; tier++) {
250
STORAGE_ENGINE *eng = qn->rrdhost->db[tier].eng;
251
tier_retention[tier].eng = eng;
252
tier_retention[tier].db_update_every_s = (time_t) (qn->rrdhost->db[tier].tier_grouping * ri->update_every_s);
@@ -284,7 +284,7 @@ static bool query_metric_add(QUERY_TARGET_LOCALS *qtl, QUERY_NODE *qn, QUERY_CON
284
}
285
}
286
287
- for (size_t tier = 0; tier < storage_tiers; tier++) {
287
+ for (size_t tier = 0; tier < nd_profile.storage_tiers; tier++) {
288
if(!qt->db.tiers[tier].update_every || (tier_retention[tier].db_update_every_s && tier_retention[tier].db_update_every_s < qt->db.tiers[tier].update_every))
289
qt->db.tiers[tier].update_every = tier_retention[tier].db_update_every_s;
290
@@ -328,7 +328,7 @@ static bool query_metric_add(QUERY_TARGET_LOCALS *qtl, QUERY_NODE *qn, QUERY_CON
328
if (!qt->db.last_time_s || common_last_time_s > qt->db.last_time_s)
329
qt->db.last_time_s = common_last_time_s;
330
331
- for (size_t tier = 0; tier < storage_tiers; tier++) {
331
+ for (size_t tier = 0; tier < nd_profile.storage_tiers; tier++) {
332
internal_fatal(tier_retention[tier].eng != query_metric_storage_engine(qt, qm, tier), "QUERY TARGET: storage engine mismatch");
333
qm->tiers[tier].smh = tier_retention[tier].smh;
334
qm->tiers[tier].db_first_time_s = tier_retention[tier].db_first_time_s;
@@ -340,7 +340,7 @@ static bool query_metric_add(QUERY_TARGET_LOCALS *qtl, QUERY_NODE *qn, QUERY_CON
340
}
341
342
// cleanup anything we allocated to the retention we will not use
343
- for(size_t tier = 0; tier < storage_tiers ;tier++) {
343
+ for(size_t tier = 0; tier < nd_profile.storage_tiers;tier++) {
344
if (tier_retention[tier].smh) {
345
tier_retention[tier].eng->api.metric_release(tier_retention[tier].smh);
346
tier_retention[tier].smh = NULL;
src/database/contexts/worker.c
+3
-3
@@ -248,12 +248,12 @@ bool rrdmetric_update_retention(RRDMETRIC *rm) {
248
}
249
else {
250
RRDHOST *rrdhost = rm->ri->rc->rrdhost;
251
- for (size_t tier = 0; tier < storage_tiers; tier++) {
251
+ for (size_t tier = 0; tier < nd_profile.storage_tiers; tier++) {
252
STORAGE_ENGINE *eng = rrdhost->db[tier].eng;
253
254
- time_t first_time_t, last_time_t;
254
+ time_t first_time_t = 0, last_time_t = 0;
255
if (eng->api.metric_retention_by_uuid(rrdhost->db[tier].si, &rm->uuid, &first_time_t, &last_time_t)) {
256
- if (first_time_t < min_first_time_t)
256
+ if (first_time_t > 0 && first_time_t < min_first_time_t)
257
min_first_time_t = first_time_t;
258
259
if (last_time_t > max_last_time_t)
src/database/engine/cache.c
+188
-190
@@ -15,7 +15,7 @@
15
*/
16
17
typedef int32_t REFCOUNT;
18
-#define REFCOUNT_DELETING (-100)
18
+#define REFCOUNT_DICONNECTED (-100)
19
20
// to use ARAL uncomment the following line:
21
#if !defined(FSANITIZE_ADDRESS)
@@ -225,49 +225,27 @@ static inline size_t pgc_indexing_partition(PGC *cache, Word_t metric_id) {
225
return last_partition;
226
}
227
228
-static inline void pgc_index_read_lock(PGC *cache, size_t partition) {
229
- rw_spinlock_read_lock(&cache->index[partition].rw_spinlock);
230
-}
231
-static inline void pgc_index_read_unlock(PGC *cache, size_t partition) {
232
- rw_spinlock_read_unlock(&cache->index[partition].rw_spinlock);
233
-}
234
-static inline void pgc_index_write_lock(PGC *cache, size_t partition) {
235
- rw_spinlock_write_lock(&cache->index[partition].rw_spinlock);
236
-}
237
-static inline void pgc_index_write_unlock(PGC *cache, size_t partition) {
238
- rw_spinlock_write_unlock(&cache->index[partition].rw_spinlock);
239
-}
240
-static inline bool pgc_index_trywrite_lock(PGC *cache, size_t partition, bool force) {
241
- if(force) {
242
- rw_spinlock_write_lock(&cache->index[partition].rw_spinlock);
243
- return true;
244
- }
245
- return rw_spinlock_trywrite_lock(&cache->index[partition].rw_spinlock);
246
-}
247
-
248
-static inline bool pgc_queue_trylock(PGC *cache __maybe_unused, struct pgc_queue *ll) {
249
- return spinlock_trylock(&ll->spinlock);
250
-}
251
-
252
-static inline void pgc_queue_lock(PGC *cache __maybe_unused, struct pgc_queue *ll) {
253
- spinlock_lock(&ll->spinlock);
254
-}
255
-
256
-static inline void pgc_queue_unlock(PGC *cache __maybe_unused, struct pgc_queue *ll) {
257
- spinlock_unlock(&ll->spinlock);
258
-}
259
-
260
-static inline bool page_transition_trylock(PGC *cache __maybe_unused, PGC_PAGE *page) {
261
- return spinlock_trylock(&page->transition_spinlock);
262
-}
263
-
264
-static inline void page_transition_lock(PGC *cache __maybe_unused, PGC_PAGE *page) {
265
- spinlock_lock(&page->transition_spinlock);
266
-}
267
-
268
-static inline void page_transition_unlock(PGC *cache __maybe_unused, PGC_PAGE *page) {
269
- spinlock_unlock(&page->transition_spinlock);
270
-}
228
+#define pgc_index_read_lock(cache, partition) rw_spinlock_read_lock(&(cache)->index[partition].rw_spinlock)
229
+#define pgc_index_read_unlock(cache, partition) rw_spinlock_read_unlock(&(cache)->index[partition].rw_spinlock)
230
+#define pgc_index_write_lock(cache, partition) rw_spinlock_write_lock(&(cache)->index[partition].rw_spinlock)
231
+#define pgc_index_write_unlock(cache, partition) rw_spinlock_write_unlock(&(cache)->index[partition].rw_spinlock)
232
+#define pgc_index_trywrite_lock(cache, partition, force) ({ \
233
+ bool _result; \
234
+ if (force) { \
235
+ rw_spinlock_write_lock(&(cache)->index[partition].rw_spinlock); \
236
+ _result = true; \
237
+ } else \
238
+ _result = rw_spinlock_trywrite_lock(&(cache)->index[partition].rw_spinlock); \
239
+ _result; \
240
+})
241
+
242
+#define pgc_queue_trylock(cache, ll) spinlock_trylock(&(ll)->spinlock)
243
+#define pgc_queue_lock(cache, ll) spinlock_lock(&(ll)->spinlock)
244
+#define pgc_queue_unlock(cache, ll) spinlock_unlock(&(ll)->spinlock)
245
+
246
+#define page_transition_trylock(cache, page) spinlock_trylock(&(page)->transition_spinlock)
247
+#define page_transition_lock(cache, page) spinlock_lock(&(page)->transition_spinlock)
248
+#define page_transition_unlock(cache, page) spinlock_unlock(&(page)->transition_spinlock)
249
250
// ----------------------------------------------------------------------------
251
// size histogram
@@ -457,7 +435,7 @@ static inline size_t cache_usage_per1000(PGC *cache, size_t *size_to_evict) {
435
436
if(signal) {
437
completion_mark_complete_a_job(&cache->evictor.completion);
460
- __atomic_add_fetch(&cache->stats.waste_evict_thread_signals, 1, __ATOMIC_RELAXED);
438
+ p2_add_fetch(&cache->stats.p2_waste_evict_thread_signals, 1);
439
}
440
}
441
@@ -481,13 +459,13 @@ static bool evict_pages_with_filter(PGC *cache, size_t max_skip, size_t max_evic
459
static inline bool flushing_critical(PGC *cache);
460
static bool flush_pages(PGC *cache, size_t max_flushes, Word_t section, bool wait, bool all_of_them);
461
484
-static void signal_evict_thread_or_evict_inline(PGC *cache, bool on_release) {
462
+static void evict_pages_inline(PGC *cache, bool on_release) {
463
const size_t per1000 = cache_usage_per1000(cache, NULL);
464
465
if(!(cache->config.options & PGC_OPTIONS_EVICT_PAGES_NO_INLINE)) {
466
if (per1000 > cache->config.aggressive_evict_per1000 && !on_release) {
467
// the threads that add pages, turn into evictors when the cache needs evictions aggressively
490
- __atomic_add_fetch(&cache->stats.waste_evictions_inline_on_add, 1, __ATOMIC_RELAXED);
468
+ p2_add_fetch(&cache->stats.p2_waste_evictions_inline_on_add, 1);
469
evict_pages(cache,
470
cache->config.max_skip_pages_per_inline_eviction,
471
cache->config.max_pages_per_inline_eviction,
@@ -495,7 +473,7 @@ static void signal_evict_thread_or_evict_inline(PGC *cache, bool on_release) {
473
}
474
else if (per1000 > cache->config.severe_pressure_per1000 && on_release) {
475
// the threads that are releasing pages, turn into evictors when the cache is critical
498
- __atomic_add_fetch(&cache->stats.waste_evictions_inline_on_release, 1, __ATOMIC_RELAXED);
476
+ p2_add_fetch(&cache->stats.p2_waste_evictions_inline_on_release, 1);
477
478
evict_pages(cache,
479
cache->config.max_skip_pages_per_inline_eviction,
@@ -506,19 +484,19 @@ static void signal_evict_thread_or_evict_inline(PGC *cache, bool on_release) {
484
}
485
486
static inline void evict_on_clean_page_added(PGC *cache) {
509
- signal_evict_thread_or_evict_inline(cache, false);
487
+ evict_pages_inline(cache, false);
488
}
489
490
static inline void evict_on_page_release_when_permitted(PGC *cache) {
513
- signal_evict_thread_or_evict_inline(cache, true);
491
+ evict_pages_inline(cache, true);
492
}
493
494
static inline void flush_inline(PGC *cache, bool on_release) {
495
if(!(cache->config.options & PGC_OPTIONS_FLUSH_PAGES_NO_INLINE) && flushing_critical(cache)) {
496
if (on_release)
519
- __atomic_add_fetch(&cache->stats.waste_flush_on_release, 1, __ATOMIC_RELAXED);
497
+ p2_add_fetch(&cache->stats.p2_waste_flush_on_release, 1);
498
else
521
- __atomic_add_fetch(&cache->stats.waste_flush_on_add, 1, __ATOMIC_RELAXED);
499
+ p2_add_fetch(&cache->stats.p2_waste_flush_on_add, 1);
500
501
flush_pages(cache, cache->config.max_flushes_inline, PGC_SECTION_ALL, false, false);
502
}
@@ -582,7 +560,8 @@ static void pgc_section_pages_static_aral_init(void) {
560
561
if(!pgc_sections_aral) {
562
pgc_sections_aral = aral_create(
585
- "pgc-sections", sizeof(struct section_pages), 0, 0, &pgc_aral_statistics, NULL, NULL, false, false);
563
+ "pgc-sections", sizeof(struct section_pages), 0, 0, &pgc_aral_statistics,
564
+ NULL, NULL, false, false, false);
565
566
pulse_aral_register_statistics(&pgc_aral_statistics, "pgc");
567
}
@@ -885,7 +864,7 @@ static inline bool page_acquire(PGC *cache, PGC_PAGE *page) {
864
} while(!__atomic_compare_exchange_n(&page->refcount, &expected, desired, false, __ATOMIC_ACQUIRE, __ATOMIC_RELAXED));
865
866
if(unlikely(spins > 1))
888
- __atomic_add_fetch(&cache->stats.waste_acquire_spins, spins - 1, __ATOMIC_RELAXED);
867
+ p2_add_fetch(&cache->stats.p2_waste_acquire_spins, spins - 1);
868
869
if(desired == 1)
870
PGC_REFERENCED_PAGES_PLUS1(cache, page);
@@ -913,7 +892,7 @@ static inline void page_release(PGC *cache, PGC_PAGE *page, bool evict_if_necess
892
} while(!__atomic_compare_exchange_n(&page->refcount, &expected, desired, false, __ATOMIC_RELEASE, __ATOMIC_RELAXED));
893
894
if(unlikely(spins > 1))
916
- __atomic_add_fetch(&cache->stats.waste_release_spins, spins - 1, __ATOMIC_RELAXED);
895
+ p2_add_fetch(&cache->stats.p2_waste_release_spins, spins - 1);
896
897
if(desired == 0) {
898
PGC_REFERENCED_PAGES_MINUS1(cache, assumed_size);
@@ -939,7 +918,7 @@ static inline bool non_acquired_page_get_for_deletion___while_having_clean_locke
918
spins++;
919
920
if (expected == 0) {
942
- desired = REFCOUNT_DELETING;
921
+ desired = REFCOUNT_DICONNECTED;
922
delete_it = true;
923
}
924
else {
@@ -958,7 +937,7 @@ static inline bool non_acquired_page_get_for_deletion___while_having_clean_locke
937
}
938
939
if(unlikely(spins > 1))
961
- __atomic_add_fetch(&cache->stats.waste_delete_spins, spins - 1, __ATOMIC_RELAXED);
940
+ p2_add_fetch(&cache->stats.p2_waste_delete_spins, spins - 1);
941
942
return delete_it;
943
}
@@ -982,7 +961,7 @@ static inline bool acquired_page_get_for_deletion_or_release_it(PGC *cache __may
961
962
if (expected == 1) {
963
// we are the only one having this page referenced
985
- desired = REFCOUNT_DELETING;
964
+ desired = REFCOUNT_DICONNECTED;
965
delete_it = true;
966
}
967
else {
@@ -1004,7 +983,7 @@ static inline bool acquired_page_get_for_deletion_or_release_it(PGC *cache __may
983
}
984
985
if(unlikely(spins > 1))
1007
- __atomic_add_fetch(&cache->stats.waste_delete_spins, spins - 1, __ATOMIC_RELAXED);
986
+ p2_add_fetch(&cache->stats.p2_waste_delete_spins, spins - 1);
987
988
return delete_it;
989
}
@@ -1149,9 +1128,9 @@ static bool evict_pages_with_filter(PGC *cache, size_t max_skip, size_t max_evic
1128
return false;
1129
1130
bool under_sever_pressure = per1000 >= cache->config.severe_pressure_per1000;
1152
- size_t workers_running = __atomic_add_fetch(&cache->stats.workers_evict, 1, __ATOMIC_RELAXED);
1131
+ size_t workers_running = __atomic_add_fetch(&cache->stats.p0_workers_evict, 1, __ATOMIC_RELAXED);
1132
if(!wait && !all_of_them && workers_running > cache->config.max_workers_evict_inline && !under_sever_pressure) {
1154
- __atomic_sub_fetch(&cache->stats.workers_evict, 1, __ATOMIC_RELAXED);
1133
+ __atomic_sub_fetch(&cache->stats.p0_workers_evict, 1, __ATOMIC_RELAXED);
1134
return false;
1135
}
1136
@@ -1186,18 +1165,17 @@ static bool evict_pages_with_filter(PGC *cache, size_t max_skip, size_t max_evic
1165
else if(unlikely(wait)) {
1166
// evict as many as necessary for the cache to go at the predefined threshold
1167
per1000 = cache_usage_per1000(cache, &max_size_to_evict);
1189
- max_size_to_evict /= 3; // do it in 3 steps
1168
if(per1000 >= cache->config.severe_pressure_per1000) {
1169
under_sever_pressure = true;
1192
- max_pages_to_evict = max_pages_to_evict ? max_pages_to_evict * 2 : 512;
1193
- if(max_pages_to_evict > 4096)
1194
- max_pages_to_evict = 4096;
1170
+ max_pages_to_evict = max_pages_to_evict ? max_pages_to_evict * 2 : 16;
1171
+ if(max_pages_to_evict > 64)
1172
+ max_pages_to_evict = 64;
1173
}
1174
else if(per1000 >= cache->config.aggressive_evict_per1000) {
1175
under_sever_pressure = false;
1198
- max_pages_to_evict = max_pages_to_evict ? max_pages_to_evict * 2 : 32;
1199
- if(max_pages_to_evict > 1024)
1200
- max_pages_to_evict = 1024;
1176
+ max_pages_to_evict = max_pages_to_evict ? max_pages_to_evict * 2 : 4;
1177
+ if(max_pages_to_evict > 16)
1178
+ max_pages_to_evict = 16;
1179
}
1180
else {
1181
under_sever_pressure = false;
@@ -1220,7 +1198,7 @@ static bool evict_pages_with_filter(PGC *cache, size_t max_skip, size_t max_evic
1198
}
1199
1200
if(++spins > 1 && !this_loop_evicted)
1223
- __atomic_add_fetch(&cache->stats.waste_evict_useless_spins, 1, __ATOMIC_RELAXED);
1201
+ p2_add_fetch(&cache->stats.p2_waste_evict_useless_spins, 1);
1202
1203
this_loop_evicted = 0;
1204
@@ -1423,9 +1401,9 @@ static bool evict_pages_with_filter(PGC *cache, size_t max_skip, size_t max_evic
1401
1402
premature_exit:
1403
if(unlikely(total_pages_relocated))
1426
- __atomic_add_fetch(&cache->stats.waste_evict_relocated, total_pages_relocated, __ATOMIC_RELAXED);
1404
+ p2_add_fetch(&cache->stats.p2_waste_evict_relocated, total_pages_relocated);
1405
1428
- __atomic_sub_fetch(&cache->stats.workers_evict, 1, __ATOMIC_RELAXED);
1406
+ __atomic_sub_fetch(&cache->stats.p0_workers_evict, 1, __ATOMIC_RELAXED);
1407
1408
return stopped_before_finishing;
1409
}
@@ -1434,7 +1412,7 @@ static PGC_PAGE *page_add(PGC *cache, PGC_ENTRY *entry, bool *added) {
1412
internal_fatal(entry->start_time_s < 0 || entry->end_time_s < 0,
1413
"DBENGINE CACHE: timestamps are negative");
1414
1437
- __atomic_add_fetch(&cache->stats.workers_add, 1, __ATOMIC_RELAXED);
1415
+ p2_add_fetch(&cache->stats.p2_workers_add, 1);
1416
1417
size_t partition = pgc_indexing_partition(cache, entry->metric_id);
1418
@@ -1451,8 +1429,7 @@ static PGC_PAGE *page_add(PGC *cache, PGC_ENTRY *entry, bool *added) {
1429
entry->end_time_s = 0;
1430
1431
do {
1454
- if(++spins > 1)
1455
- __atomic_add_fetch(&cache->stats.waste_insert_spins, 1, __ATOMIC_RELAXED);
1432
+ spins++;
1433
1434
pgc_index_write_lock(cache, partition);
1435
@@ -1553,7 +1530,10 @@ static PGC_PAGE *page_add(PGC *cache, PGC_ENTRY *entry, bool *added) {
1530
aral_freez(cache->index[partition].aral, allocation);
1531
#endif
1532
1556
- __atomic_sub_fetch(&cache->stats.workers_add, 1, __ATOMIC_RELAXED);
1533
+ if(spins > 1)
1534
+ p2_add_fetch(&cache->stats.p2_waste_insert_spins, spins - 1);
1535
+
1536
+ p2_sub_fetch(&cache->stats.p2_workers_add, 1);
1537
1538
if(!entry->hot)
1539
evict_on_clean_page_added(cache);
@@ -1563,9 +1543,95 @@ static PGC_PAGE *page_add(PGC *cache, PGC_ENTRY *entry, bool *added) {
1543
return page;
1544
}
1545
1566
-static PGC_PAGE *page_find_and_acquire_once(PGC *cache, Word_t section, Word_t metric_id, time_t start_time_s, PGC_SEARCH method, bool *retry) {
1567
- *retry = false;
1546
+static inline PGC_PAGE *page_find_and_acquire_exact_unsafe(PGC *cache, Pvoid_t *pages_judy_pptr, time_t start_time_s) {
1547
+ Pvoid_t *page_ptr = JudyLGet(*pages_judy_pptr, start_time_s, PJE0);
1548
+ if(!page_ptr)
1549
+ return NULL;
1550
+
1551
+ if (unlikely(page_ptr == PJERR))
1552
+ fatal("DBENGINE CACHE: corrupted page in pages judy array");
1553
+
1554
+ PGC_PAGE *page = *page_ptr;
1555
+ if(page && page_acquire(cache, page))
1556
+ // we have our page acquired
1557
+ return page;
1558
+
1559
+ return NULL;
1560
+}
1561
+
1562
+static inline PGC_PAGE *page_find_and_acquire_first_unsafe(PGC *cache, Pvoid_t *pages_judy_pptr, time_t start_time_s) {
1563
+ Word_t time = start_time_s;
1564
+ for(Pvoid_t *page_ptr = JudyLFirst(*pages_judy_pptr, &time, PJE0);
1565
+ page_ptr ;
1566
+ page_ptr = JudyLNext(*pages_judy_pptr, &time, PJE0)) {
1567
+
1568
+ if (unlikely(page_ptr == PJERR))
1569
+ fatal("DBENGINE CACHE: corrupted page in pages judy array");
1570
+
1571
+ PGC_PAGE *page = *page_ptr;
1572
+ if(page && page_acquire(cache, page))
1573
+ // we have our page acquired
1574
+ return page;
1575
+ }
1576
+
1577
+ return NULL;
1578
+}
1579
+
1580
+static inline PGC_PAGE *page_find_and_acquire_next_unsafe(PGC *cache, Pvoid_t *pages_judy_pptr, time_t start_time_s) {
1581
+ Word_t time = start_time_s;
1582
+ for(Pvoid_t *page_ptr = JudyLNext(*pages_judy_pptr, &time, PJE0);
1583
+ page_ptr ;
1584
+ page_ptr = JudyLNext(*pages_judy_pptr, &time, PJE0)) {
1585
+
1586
+ if (unlikely(page_ptr == PJERR))
1587
+ fatal("DBENGINE CACHE: corrupted page in pages judy array");
1588
+
1589
+ PGC_PAGE *page = *page_ptr;
1590
+ if(page && page_acquire(cache, page))
1591
+ // we have our page acquired
1592
+ return page;
1593
+ }
1594
+
1595
+ return NULL;
1596
+}
1597
+
1598
+static inline PGC_PAGE *page_find_and_acquire_last_unsafe(PGC *cache, Pvoid_t *pages_judy_pptr, time_t start_time_s) {
1599
+ Word_t time = start_time_s;
1600
+ for(Pvoid_t *page_ptr = JudyLLast(*pages_judy_pptr, &time, PJE0);
1601
+ page_ptr ;
1602
+ page_ptr = JudyLPrev(*pages_judy_pptr, &time, PJE0)) {
1603
+
1604
+ if (unlikely(page_ptr == PJERR))
1605
+ fatal("DBENGINE CACHE: corrupted page in pages judy array");
1606
+
1607
+ PGC_PAGE *page = *page_ptr;
1608
+ if(page && page_acquire(cache, page))
1609
+ // we have our page acquired
1610
+ return page;
1611
+ }
1612
+
1613
+ return NULL;
1614
+}
1615
+
1616
+static inline PGC_PAGE *page_find_and_acquire_prev_unsafe(PGC *cache, Pvoid_t *pages_judy_pptr, time_t start_time_s) {
1617
+ Word_t time = start_time_s;
1618
+ for(Pvoid_t *page_ptr = JudyLPrev(*pages_judy_pptr, &time, PJE0);
1619
+ page_ptr ;
1620
+ page_ptr = JudyLPrev(*pages_judy_pptr, &time, PJE0)) {
1621
1622
+ if (unlikely(page_ptr == PJERR))
1623
+ fatal("DBENGINE CACHE: corrupted page in pages judy array");
1624
+
1625
+ PGC_PAGE *page = *page_ptr;
1626
+ if(page && page_acquire(cache, page))
1627
+ // we have our page acquired
1628
+ return page;
1629
+ }
1630
+
1631
+ return NULL;
1632
+}
1633
+
1634
+static PGC_PAGE *page_find_and_acquire_once(PGC *cache, Word_t section, Word_t metric_id, time_t start_time_s, PGC_SEARCH method) {
1635
PGC_PAGE *page = NULL;
1636
size_t partition = pgc_indexing_partition(cache, metric_id);
1637
@@ -1592,105 +1658,47 @@ static PGC_PAGE *page_find_and_acquire_once(PGC *cache, Word_t section, Word_t m
1658
switch(method) {
1659
default:
1660
case PGC_SEARCH_CLOSEST: {
1595
- Pvoid_t *page_ptr = JudyLGet(*pages_judy_pptr, start_time_s, PJE0);
1596
- if (unlikely(page_ptr == PJERR))
1597
- fatal("DBENGINE CACHE: corrupted page in pages judy array");
1598
-
1599
- if (page_ptr)
1600
- page = *page_ptr;
1601
-
1602
- else {
1603
- Word_t time = start_time_s;
1604
-
1605
- // find the previous page
1606
- page_ptr = JudyLPrev(*pages_judy_pptr, &time, PJE0);
1607
- if(unlikely(page_ptr == PJERR))
1608
- fatal("DBENGINE CACHE: corrupted page in pages judy array #2");
1609
-
1610
- if(page_ptr) {
1661
+ page = page_find_and_acquire_exact_unsafe(cache, pages_judy_pptr, start_time_s);
1662
+ if(!page) {
1663
+ page = page_find_and_acquire_prev_unsafe(cache, pages_judy_pptr, start_time_s);
1664
+ if(page && start_time_s > page->end_time_s) {
1665
// found a page starting before our timestamp
1612
- // check if our timestamp is included
1613
- page = *page_ptr;
1614
- if(start_time_s > page->end_time_s)
1615
- // it is not good for us
1616
- page = NULL;
1666
+ // but our timestamp is not included in it
1667
+ page_release(cache, page, false);
1668
+ page = NULL;
1669
}
1670
1619
- if(!page) {
1620
- // find the next page then...
1621
- time = start_time_s;
1622
- page_ptr = JudyLNext(*pages_judy_pptr, &time, PJE0);
1623
- if(page_ptr)
1624
- page = *page_ptr;
1625
- }
1671
+ if(!page)
1672
+ page = page_find_and_acquire_next_unsafe(cache, pages_judy_pptr, start_time_s);
1673
}
1674
}
1675
break;
1676
1630
- case PGC_SEARCH_EXACT: {
1631
- Pvoid_t *page_ptr = JudyLGet(*pages_judy_pptr, start_time_s, PJE0);
1632
- if (unlikely(page_ptr == PJERR))
1633
- fatal("DBENGINE CACHE: corrupted page in pages judy array");
1634
-
1635
- if (page_ptr)
1636
- page = *page_ptr;
1637
- }
1638
- break;
1639
-
1640
- case PGC_SEARCH_FIRST: {
1641
- Word_t time = start_time_s;
1642
- Pvoid_t *page_ptr = JudyLFirst(*pages_judy_pptr, &time, PJE0);
1643
- if (unlikely(page_ptr == PJERR))
1644
- fatal("DBENGINE CACHE: corrupted page in pages judy array");
1645
-
1646
- if (page_ptr)
1647
- page = *page_ptr;
1648
- }
1649
- break;
1650
-
1651
- case PGC_SEARCH_NEXT: {
1652
- Word_t time = start_time_s;
1653
- Pvoid_t *page_ptr = JudyLNext(*pages_judy_pptr, &time, PJE0);
1654
- if (unlikely(page_ptr == PJERR))
1655
- fatal("DBENGINE CACHE: corrupted page in pages judy array");
1656
-
1657
- if (page_ptr)
1658
- page = *page_ptr;
1659
- }
1660
- break;
1677
+ case PGC_SEARCH_EXACT:
1678
+ page = page_find_and_acquire_exact_unsafe(cache, pages_judy_pptr, start_time_s);
1679
+ break;
1680
1662
- case PGC_SEARCH_LAST: {
1663
- Word_t time = start_time_s;
1664
- Pvoid_t *page_ptr = JudyLLast(*pages_judy_pptr, &time, PJE0);
1665
- if (unlikely(page_ptr == PJERR))
1666
- fatal("DBENGINE CACHE: corrupted page in pages judy array");
1681
+ case PGC_SEARCH_FIRST:
1682
+ page = page_find_and_acquire_first_unsafe(cache, pages_judy_pptr, start_time_s);
1683
+ break;
1684
1668
- if (page_ptr)
1669
- page = *page_ptr;
1670
- }
1671
- break;
1685
+ case PGC_SEARCH_NEXT:
1686
+ page = page_find_and_acquire_next_unsafe(cache, pages_judy_pptr, start_time_s);
1687
+ break;
1688
1673
- case PGC_SEARCH_PREV: {
1674
- Word_t time = start_time_s;
1675
- Pvoid_t *page_ptr = JudyLPrev(*pages_judy_pptr, &time, PJE0);
1676
- if (unlikely(page_ptr == PJERR))
1677
- fatal("DBENGINE CACHE: corrupted page in pages judy array");
1689
+ case PGC_SEARCH_LAST:
1690
+ page = page_find_and_acquire_last_unsafe(cache, pages_judy_pptr, start_time_s);
1691
+ break;
1692
1679
- if (page_ptr)
1680
- page = *page_ptr;
1681
- }
1682
- break;
1693
+ case PGC_SEARCH_PREV:
1694
+ page = page_find_and_acquire_prev_unsafe(cache, pages_judy_pptr, start_time_s);
1695
+ break;
1696
}
1697
1685
- if(page) {
1698
+#ifdef NETDATA_PGC_POINTER_CHECK
1699
+ if(page)
1700
pointer_check(cache, page);
1687
-
1688
- if(!page_acquire(cache, page)) {
1689
- // this page is not good to use
1690
- *retry = true;
1691
- page = NULL;
1692
- }
1693
- }
1701
+#endif
1702
1703
cleanup:
1704
pgc_index_read_unlock(cache, partition);
@@ -1752,7 +1760,7 @@ static bool flush_pages(PGC *cache, size_t max_flushes, Word_t section, bool wai
1760
return false;
1761
}
1762
1755
- __atomic_add_fetch(&cache->stats.workers_flush, 1, __ATOMIC_RELAXED);
1763
+ p2_add_fetch(&cache->stats.p2_workers_flush, 1);
1764
1765
bool have_dirty_lock = true;
1766
@@ -1865,8 +1873,8 @@ static bool flush_pages(PGC *cache, size_t max_flushes, Word_t section, bool wai
1873
// page ptr may be invalid now
1874
}
1875
1868
- __atomic_add_fetch(&cache->stats.waste_flushes_cancelled, pages_cancelled, __ATOMIC_RELAXED);
1869
- __atomic_add_fetch(&cache->stats.flushes_cancelled_size, pages_cancelled_size, __ATOMIC_RELAXED);
1876
+ p2_add_fetch(&cache->stats.p2_waste_flushes_cancelled, pages_cancelled);
1877
+ p2_add_fetch(&cache->stats.flushes_cancelled_size, pages_cancelled_size);
1878
1879
internal_fatal(pages_added != pages_cancelled || pages_added_size != pages_cancelled_size,
1880
"DBENGINE CACHE: flushing cancel pages mismatch");
@@ -1938,7 +1946,7 @@ static bool flush_pages(PGC *cache, size_t max_flushes, Word_t section, bool wai
1946
pgc_queue_unlock(cache, &cache->dirty);
1947
}
1948
1941
- __atomic_sub_fetch(&cache->stats.workers_flush, 1, __ATOMIC_RELAXED);
1949
+ p2_sub_fetch(&cache->stats.p2_workers_flush, 1);
1950
1951
return stopped_before_finishing;
1952
}
@@ -2066,10 +2074,8 @@ PGC *pgc_create(const char *name,
2074
0,
2075
0,
2076
&pgc_aral_statistics,
2069
- NULL,
2070
- NULL,
2071
- false,
2072
- false);
2077
+ NULL, NULL,
2078
+ false, false, false);
2079
}
2080
}
2081
#endif
@@ -2165,7 +2171,7 @@ void pgc_page_release(PGC *cache, PGC_PAGE *page) {
2171
}
2172
2173
void pgc_page_hot_to_dirty_and_release(PGC *cache, PGC_PAGE *page, bool never_flush) {
2168
- __atomic_add_fetch(&cache->stats.workers_hot2dirty, 1, __ATOMIC_RELAXED);
2174
+ p2_add_fetch(&cache->stats.p2_workers_hot2dirty, 1);
2175
2176
//#ifdef NETDATA_INTERNAL_CHECKS
2177
// page_transition_lock(cache, page);
@@ -2180,7 +2186,7 @@ void pgc_page_hot_to_dirty_and_release(PGC *cache, PGC_PAGE *page, bool never_fl
2186
page_release(cache, page, true);
2187
// page ptr may be invalid now
2188
2183
- __atomic_sub_fetch(&cache->stats.workers_hot2dirty, 1, __ATOMIC_RELAXED);
2189
+ p2_sub_fetch(&cache->stats.p2_workers_hot2dirty, 1);
2190
2191
// flush, if we have to
2192
if(!never_flush)
@@ -2190,7 +2196,7 @@ void pgc_page_hot_to_dirty_and_release(PGC *cache, PGC_PAGE *page, bool never_fl
2196
bool pgc_page_to_clean_evict_or_release(PGC *cache, PGC_PAGE *page) {
2197
bool ret;
2198
2193
- __atomic_add_fetch(&cache->stats.workers_hot2dirty, 1, __ATOMIC_RELAXED);
2199
+ p2_add_fetch(&cache->stats.p2_workers_hot2dirty, 1);
2200
2201
// prevent accesses from increasing the accesses counter
2202
page_flag_set(page, PGC_PAGE_HAS_NO_DATA_IGNORE_ACCESSES);
@@ -2208,7 +2214,7 @@ bool pgc_page_to_clean_evict_or_release(PGC *cache, PGC_PAGE *page) {
2214
ret = false;
2215
}
2216
2211
- __atomic_sub_fetch(&cache->stats.workers_hot2dirty, 1, __ATOMIC_RELAXED);
2217
+ p2_sub_fetch(&cache->stats.p2_workers_hot2dirty, 1);
2218
2219
return ret;
2220
}
@@ -2371,7 +2377,7 @@ void pgc_page_hot_set_end_time_s(PGC *cache __maybe_unused, PGC_PAGE *page, time
2377
PGC_PAGE *pgc_page_get_and_acquire(PGC *cache, Word_t section, Word_t metric_id, time_t start_time_s, PGC_SEARCH method) {
2378
PGC_PAGE *page = NULL;
2379
2374
- __atomic_add_fetch(&cache->stats.workers_search, 1, __ATOMIC_RELAXED);
2380
+ p2_add_fetch(&cache->stats.p2_workers_search, 1);
2381
2382
size_t *stats_hit_ptr, *stats_miss_ptr;
2383
@@ -2386,17 +2392,7 @@ PGC_PAGE *pgc_page_get_and_acquire(PGC *cache, Word_t section, Word_t metric_id,
2392
stats_miss_ptr = &cache->stats.searches_exact_misses;
2393
}
2394
2389
- while(1) {
2390
- bool retry = false;
2391
-
2392
- page = page_find_and_acquire_once(cache, section, metric_id, start_time_s, method, &retry);
2393
-
2394
- if(page || !retry)
2395
- break;
2396
-
2397
- tinysleep();
2398
- }
2399
-
2395
+ page = page_find_and_acquire_once(cache, section, metric_id, start_time_s, method);
2396
if(page) {
2397
__atomic_add_fetch(stats_hit_ptr, 1, __ATOMIC_RELAXED);
2398
page_has_been_accessed(cache, page);
@@ -2404,7 +2400,7 @@ PGC_PAGE *pgc_page_get_and_acquire(PGC *cache, Word_t section, Word_t metric_id,
2400
else
2401
__atomic_add_fetch(stats_miss_ptr, 1, __ATOMIC_RELAXED);
2402
2407
- __atomic_sub_fetch(&cache->stats.workers_search, 1, __ATOMIC_RELAXED);
2403
+ p2_sub_fetch(&cache->stats.p2_workers_search, 1);
2404
2405
return page;
2406
}
@@ -2427,7 +2423,7 @@ size_t pgc_hot_and_dirty_entries(PGC *cache) {
2423
2424
void pgc_open_cache_to_journal_v2(PGC *cache, Word_t section, unsigned datafile_fileno, uint8_t type, migrate_to_v2_callback cb, void *data) {
2425
__atomic_add_fetch(&rrdeng_cache_efficiency_stats.journal_v2_indexing_started, 1, __ATOMIC_RELAXED);
2430
- __atomic_add_fetch(&cache->stats.workers_jv2_flush, 1, __ATOMIC_RELAXED);
2426
+ p2_add_fetch(&cache->stats.p2_workers_jv2_flush, 1);
2427
2428
pgc_queue_lock(cache, &cache->hot);
2429
@@ -2559,7 +2555,8 @@ void pgc_open_cache_to_journal_v2(PGC *cache, Word_t section, unsigned datafile_
2555
page_release(cache, page, false);
2556
}
2557
2562
- pgc_queue_lock(cache, &cache->hot);
2558
+ yield_the_processor(); // do not lock too aggressively
2559
+ pgc_queue_lock(cache, &cache->hot);
2560
}
2561
2562
spinlock_unlock(&sp->migration_to_v2_spinlock);
@@ -2582,6 +2579,7 @@ void pgc_open_cache_to_journal_v2(PGC *cache, Word_t section, unsigned datafile_
2579
struct jv2_page_info *pi = *PValue2;
2580
2581
// balance-parents: transition from hot to clean directly
2582
+ yield_the_processor(); // do not lock too aggressively
2583
page_set_clean(cache, pi->page, true, false);
2584
page_transition_unlock(cache, pi->page);
2585
page_release(cache, pi->page, true);
@@ -2616,7 +2614,7 @@ void pgc_open_cache_to_journal_v2(PGC *cache, Word_t section, unsigned datafile_
2614
aral_by_size_release(ar_pi);
2615
aral_by_size_release(ar_mi);
2616
2619
- __atomic_sub_fetch(&cache->stats.workers_jv2_flush, 1, __ATOMIC_RELAXED);
2617
+ p2_sub_fetch(&cache->stats.p2_workers_jv2_flush, 1);
2618
2619
// balance-parents: do not flush, there is nothing dirty
2620
// flush_pages(cache, cache->config.max_flushes_inline, PGC_SECTION_ALL, false, false);
src/database/engine/cache.h
+64
-128
@@ -48,180 +48,116 @@ struct pgc_size_histogram {
48
};
49
50
struct pgc_queue_statistics {
51
- CACHE_LINE_PADDING();
51
struct pgc_size_histogram size_histogram;
52
54
- CACHE_LINE_PADDING();
55
- size_t entries;
56
- CACHE_LINE_PADDING();
57
- size_t size;
53
+ PAD64(size_t) entries;
54
+ PAD64(size_t) size;
55
59
- CACHE_LINE_PADDING();
60
- size_t max_entries;
61
- CACHE_LINE_PADDING();
62
- size_t max_size;
56
+ PAD64(size_t) max_entries;
57
+ PAD64(size_t) max_size;
58
64
- CACHE_LINE_PADDING();
65
- size_t added_entries;
66
- CACHE_LINE_PADDING();
67
- size_t added_size;
59
+ PAD64(size_t) added_entries;
60
+ PAD64(size_t) added_size;
61
69
- CACHE_LINE_PADDING();
70
- size_t removed_entries;
71
- CACHE_LINE_PADDING();
72
- size_t removed_size;
73
-
74
- CACHE_LINE_PADDING();
62
+ PAD64(size_t) removed_entries;
63
+ PAD64(size_t) removed_size;
64
};
65
66
struct pgc_statistics {
78
- CACHE_LINE_PADDING();
79
- size_t wanted_cache_size;
80
- CACHE_LINE_PADDING();
81
- size_t current_cache_size;
82
- CACHE_LINE_PADDING();
67
+ PAD64(size_t) wanted_cache_size;
68
+ PAD64(size_t) current_cache_size;
69
70
// ----------------------------------------------------------------------------------------------------------------
71
// volume
72
87
- CACHE_LINE_PADDING();
88
- size_t entries; // all the entries (includes clean, dirty, hot)
89
- CACHE_LINE_PADDING();
90
- size_t size; // all the entries (includes clean, dirty, hot)
73
+ PAD64(size_t) entries; // all the entries (includes clean, dirty, hot)
74
+ PAD64(size_t) size; // all the entries (includes clean, dirty, hot)
75
92
- CACHE_LINE_PADDING();
93
- size_t referenced_entries; // all the entries currently referenced
94
- CACHE_LINE_PADDING();
95
- size_t referenced_size; // all the entries currently referenced
76
+ PAD64(size_t) referenced_entries; // all the entries currently referenced
77
+ PAD64(size_t) referenced_size; // all the entries currently referenced
78
97
- CACHE_LINE_PADDING();
98
- size_t added_entries;
99
- CACHE_LINE_PADDING();
100
- size_t added_size;
79
+ PAD64(size_t) added_entries;
80
+ PAD64(size_t) added_size;
81
102
- CACHE_LINE_PADDING();
103
- size_t removed_entries;
104
- CACHE_LINE_PADDING();
105
- size_t removed_size;
82
+ PAD64(size_t) removed_entries;
83
+ PAD64(size_t) removed_size;
84
85
#ifdef PGC_COUNT_POINTS_COLLECTED
108
- CACHE_LINE_PADDING();
109
- size_t points_collected;
86
+ PAD64(size_t) points_collected;
87
#endif
88
89
// ----------------------------------------------------------------------------------------------------------------
90
// migrations
91
115
- CACHE_LINE_PADDING();
116
- size_t evicting_entries;
117
- CACHE_LINE_PADDING();
118
- size_t evicting_size;
92
+ PAD64(size_t) evicting_entries;
93
+ PAD64(size_t) evicting_size;
94
120
- CACHE_LINE_PADDING();
121
- size_t flushing_entries;
122
- CACHE_LINE_PADDING();
123
- size_t flushing_size;
95
+ PAD64(size_t) flushing_entries;
96
+ PAD64(size_t) flushing_size;
97
125
- CACHE_LINE_PADDING();
126
- size_t hot2dirty_entries;
127
- CACHE_LINE_PADDING();
128
- size_t hot2dirty_size;
98
+ PAD64(size_t) hot2dirty_entries;
99
+ PAD64(size_t) hot2dirty_size;
100
130
- CACHE_LINE_PADDING();
131
- size_t hot_empty_pages_evicted_immediately;
132
- CACHE_LINE_PADDING();
133
- size_t hot_empty_pages_evicted_later;
101
+ PAD64(size_t) hot_empty_pages_evicted_immediately;
102
+ PAD64(size_t) hot_empty_pages_evicted_later;
103
104
// ----------------------------------------------------------------------------------------------------------------
105
// workload
106
138
- CACHE_LINE_PADDING();
139
- size_t acquires;
140
- CACHE_LINE_PADDING();
141
- size_t releases;
142
-
143
- CACHE_LINE_PADDING();
144
- size_t acquires_for_deletion;
145
-
146
- CACHE_LINE_PADDING();
147
- size_t searches_exact;
148
- CACHE_LINE_PADDING();
149
- size_t searches_exact_hits;
150
- CACHE_LINE_PADDING();
151
- size_t searches_exact_misses;
152
-
153
- CACHE_LINE_PADDING();
154
- size_t searches_closest;
155
- CACHE_LINE_PADDING();
156
- size_t searches_closest_hits;
157
- CACHE_LINE_PADDING();
158
- size_t searches_closest_misses;
159
-
160
- CACHE_LINE_PADDING();
161
- size_t flushes_completed;
162
- CACHE_LINE_PADDING();
163
- size_t flushes_completed_size;
164
- CACHE_LINE_PADDING();
165
- size_t flushes_cancelled_size;
107
+ PAD64(size_t) acquires;
108
+ PAD64(size_t) releases;
109
+
110
+ PAD64(size_t) acquires_for_deletion;
111
+
112
+ PAD64(size_t) searches_exact;
113
+ PAD64(size_t) searches_exact_hits;
114
+ PAD64(size_t) searches_exact_misses;
115
+
116
+ PAD64(size_t) searches_closest;
117
+ PAD64(size_t) searches_closest_hits;
118
+ PAD64(size_t) searches_closest_misses;
119
+
120
+ PAD64(size_t) flushes_completed;
121
+ PAD64(size_t) flushes_completed_size;
122
+ PAD64(size_t) flushes_cancelled_size;
123
124
// ----------------------------------------------------------------------------------------------------------------
125
// critical events
126
170
- CACHE_LINE_PADDING();
171
- size_t events_cache_under_severe_pressure;
172
- CACHE_LINE_PADDING();
173
- size_t events_cache_needs_space_aggressively;
174
- CACHE_LINE_PADDING();
175
- size_t events_flush_critical;
127
+ PAD64(size_t) events_cache_under_severe_pressure;
128
+ PAD64(size_t) events_cache_needs_space_aggressively;
129
+ PAD64(size_t) events_flush_critical;
130
131
// ----------------------------------------------------------------------------------------------------------------
132
// worker threads
133
180
- CACHE_LINE_PADDING();
181
- size_t workers_search;
182
- CACHE_LINE_PADDING();
183
- size_t workers_add;
184
- CACHE_LINE_PADDING();
185
- size_t workers_evict;
186
- CACHE_LINE_PADDING();
187
- size_t workers_flush;
188
- CACHE_LINE_PADDING();
189
- size_t workers_jv2_flush;
190
- CACHE_LINE_PADDING();
191
- size_t workers_hot2dirty;
134
+ PAD64(size_t) p2_workers_search;
135
+ PAD64(size_t) p2_workers_add;
136
+ PAD64(size_t) p0_workers_evict; // priority 0, we always need this when inline evictions are enabled
137
+ PAD64(size_t) p2_workers_flush;
138
+ PAD64(size_t) p2_workers_jv2_flush;
139
+ PAD64(size_t) p2_workers_hot2dirty;
140
141
// ----------------------------------------------------------------------------------------------------------------
142
// waste events
143
144
// waste events - spins
197
- CACHE_LINE_PADDING();
198
- size_t waste_insert_spins;
199
- CACHE_LINE_PADDING();
200
- size_t waste_evict_useless_spins;
201
- CACHE_LINE_PADDING();
202
- size_t waste_release_spins;
203
- CACHE_LINE_PADDING();
204
- size_t waste_acquire_spins;
205
- CACHE_LINE_PADDING();
206
- size_t waste_delete_spins;
145
+ PAD64(size_t) p2_waste_insert_spins;
146
+ PAD64(size_t) p2_waste_evict_useless_spins;
147
+ PAD64(size_t) p2_waste_release_spins;
148
+ PAD64(size_t) p2_waste_acquire_spins;
149
+ PAD64(size_t) p2_waste_delete_spins;
150
151
// waste events - eviction
209
- CACHE_LINE_PADDING();
210
- size_t waste_evict_relocated;
211
- CACHE_LINE_PADDING();
212
- size_t waste_evict_thread_signals;
213
- CACHE_LINE_PADDING();
214
- size_t waste_evictions_inline_on_add;
215
- CACHE_LINE_PADDING();
216
- size_t waste_evictions_inline_on_release;
152
+ PAD64(size_t) p2_waste_evict_relocated;
153
+ PAD64(size_t) p2_waste_evict_thread_signals;
154
+ PAD64(size_t) p2_waste_evictions_inline_on_add;
155
+ PAD64(size_t) p2_waste_evictions_inline_on_release;
156
157
// waste events - flushing
219
- CACHE_LINE_PADDING();
220
- size_t waste_flush_on_add;
221
- CACHE_LINE_PADDING();
222
- size_t waste_flush_on_release;
223
- CACHE_LINE_PADDING();
224
- size_t waste_flushes_cancelled;
158
+ PAD64(size_t) p2_waste_flush_on_add;
159
+ PAD64(size_t) p2_waste_flush_on_release;
160
+ PAD64(size_t) p2_waste_flushes_cancelled;
161
162
// ----------------------------------------------------------------------------------------------------------------
163
// per queue statistics
src/database/engine/datafile.c
+1
-1
@@ -33,7 +33,7 @@ static struct rrdengine_datafile *datafile_alloc_and_init(struct rrdengine_insta
33
34
spinlock_init(&datafile->users.spinlock);
35
spinlock_init(&datafile->writers.spinlock);
36
- spinlock_init(&datafile->extent_queries.spinlock);
36
+ rw_spinlock_init(&datafile->extent_epdl.spinlock);
37
38
return datafile;
39
}
src/database/engine/datafile.h
+9
-3
@@ -38,6 +38,12 @@ typedef enum __attribute__ ((__packed__)) {
38
DATAFILE_ACQUIRE_MAX,
39
} DATAFILE_ACQUIRE_REASONS;
40
41
+struct extent_page_details_list;
42
+typedef struct {
43
+ SPINLOCK spinlock;
44
+ struct extent_page_details_list *base;
45
+} EPDL_EXTENT;
46
+
47
/* only one event loop is supported for now */
48
struct rrdengine_datafile {
49
unsigned tier;
@@ -70,9 +76,9 @@ struct rrdengine_datafile {
76
} users;
77
78
struct {
73
- SPINLOCK spinlock;
74
- Pvoid_t pending_epdl_by_extent_offset_judyL;
75
- } extent_queries;
79
+ RW_SPINLOCK spinlock;
80
+ Pvoid_t epdl_per_extent;
81
+ } extent_epdl;
82
};
83
84
bool datafile_acquire(struct rrdengine_datafile *df, DATAFILE_ACQUIRE_REASONS reason);
src/database/engine/dbengine-stresstest.c
+1
-1
@@ -18,7 +18,7 @@ static RRDHOST *dbengine_rrdhost_find_or_create(char *name) {
18
netdata_configured_utc_offset,
19
program_name,
20
NETDATA_VERSION,
21
- default_rrd_update_every,
21
+ nd_profile.update_every,
22
default_rrd_history_entries,
23
RRD_MEMORY_MODE_DBENGINE,
24
health_plugin_enabled(),
src/database/engine/dbengine-unittest.c
+1
-1
@@ -104,7 +104,7 @@ static RRDHOST *dbengine_rrdhost_find_or_create(char *name) {
104
netdata_configured_utc_offset,
105
program_name,
106
NETDATA_VERSION,
107
- default_rrd_update_every,
107
+ nd_profile.update_every,
108
default_rrd_history_entries,
109
RRD_MEMORY_MODE_DBENGINE,
110
health_plugin_enabled(),
src/database/engine/journalfile.c
+2
-2
@@ -296,7 +296,7 @@ static bool journalfile_v2_mounted_data_unmount(struct rrdengine_journalfile *jo
296
void journalfile_v2_data_unmount_cleanup(time_t now_s) {
297
// DO NOT WAIT ON ANY LOCK!!!
298
299
- for(size_t tier = 0; tier < (size_t)storage_tiers ;tier++) {
299
+ for(size_t tier = 0; tier < (size_t)nd_profile.storage_tiers;tier++) {
300
struct rrdengine_instance *ctx = multidb_ctx[tier];
301
if(!ctx) continue;
302
@@ -1338,7 +1338,7 @@ void journalfile_migrate_to_v2_callback(Word_t section, unsigned datafile_fileno
1338
total_file_size += sizeof(struct journal_v2_block_trailer);
1339
1340
int fd_v2;
1341
- uint8_t *data_start = netdata_mmap(path, total_file_size, MAP_SHARED, 0, false, &fd_v2);
1341
+ uint8_t *data_start = netdata_mmap(path, total_file_size, MAP_SHARED, 0, false, true, &fd_v2);
1342
uint8_t *data = data_start;
1343
1344
memset(data_start, 0, extent_offset);
src/database/engine/metric.c
+8
-7
@@ -5,7 +5,7 @@
5
#include "rrddiskprotocol.h"
6
7
typedef int32_t REFCOUNT;
8
-#define REFCOUNT_DELETING (-100)
8
+#define REFCOUNT_DICONNECTED (-100)
9
10
struct metric {
11
nd_uuid_t uuid; // never changes
@@ -192,7 +192,7 @@ static inline void acquired_for_deletion_metric_delete(MRG *mrg, METRIC *metric)
192
193
rc = JudyHSDel(&mrg->index[partition].uuid_judy, &metric->uuid, sizeof(nd_uuid_t), PJE0);
194
195
- int64_t judy_mem = JudyAllocThreadPulseGetAndReset();
195
+ judy_mem = JudyAllocThreadPulseGetAndReset();
196
197
if(unlikely(!rc))
198
fatal("DBENGINE METRIC: cannot delete UUID from JudyHS");
@@ -242,22 +242,22 @@ static inline bool metric_release(MRG *mrg, METRIC *metric) {
242
}
243
244
if(expected == 1 && !acquired_metric_has_retention(mrg, metric))
245
- desired = REFCOUNT_DELETING;
245
+ desired = REFCOUNT_DICONNECTED;
246
else
247
desired = expected - 1;
248
249
} while(!__atomic_compare_exchange_n(&metric->refcount, &expected, desired, false, __ATOMIC_RELEASE, __ATOMIC_RELAXED));
250
251
- if(desired == 0 || desired == REFCOUNT_DELETING) {
251
+ if(desired == 0 || desired == REFCOUNT_DICONNECTED) {
252
__atomic_sub_fetch(&mrg->index[partition].stats.entries_acquired, 1, __ATOMIC_RELAXED);
253
254
- if(desired == REFCOUNT_DELETING)
254
+ if(desired == REFCOUNT_DICONNECTED)
255
acquired_for_deletion_metric_delete(mrg, metric);
256
}
257
258
__atomic_sub_fetch(&mrg->index[partition].stats.current_references, 1, __ATOMIC_RELAXED);
259
260
- return desired == REFCOUNT_DELETING;
260
+ return desired == REFCOUNT_DICONNECTED;
261
}
262
263
static inline METRIC *metric_add_and_acquire(MRG *mrg, MRG_ENTRY *entry, bool *ret) {
@@ -386,7 +386,8 @@ inline MRG *mrg_create(ssize_t partitions) {
386
char buf[ARAL_MAX_NAME + 1];
387
snprintfz(buf, ARAL_MAX_NAME, "mrg[%zu]", i);
388
389
- mrg->index[i].aral = aral_create(buf, sizeof(METRIC), 0, 16384, &mrg_aral_statistics, NULL, NULL, false, false);
389
+ mrg->index[i].aral = aral_create(buf, sizeof(METRIC), 0, 16384, &mrg_aral_statistics, NULL, NULL,
390
+ false, false, true);
391
}
392
pulse_aral_register_statistics(&mrg_aral_statistics, "mrg");
393
src/database/engine/metric.h
+6
-13
@@ -30,21 +30,14 @@ struct mrg_statistics {
30
31
// --- atomic --- multiple readers / writers
32
33
- CACHE_LINE_PADDING();
34
- ssize_t entries_acquired;
33
+ PAD64(ssize_t) entries_acquired;
34
+ PAD64(ssize_t) current_references;
35
36
- CACHE_LINE_PADDING();
37
- ssize_t current_references;
36
+ PAD64(size_t) search_hits;
37
+ PAD64(size_t) search_misses;
38
39
- CACHE_LINE_PADDING();
40
- size_t search_hits;
41
- CACHE_LINE_PADDING();
42
- size_t search_misses;
43
-
44
- CACHE_LINE_PADDING();
45
- size_t writers;
46
- CACHE_LINE_PADDING();
47
- size_t writers_conflicts;
39
+ PAD64(size_t) writers;
40
+ PAD64(size_t) writers_conflicts;
41
};
42
43
MRG *mrg_create(ssize_t partitions);
src/database/engine/page.c
+17
-1
@@ -99,6 +99,13 @@ static size_t aral_sizes[] = {
99
// our structures
100
sizeof(gorilla_writer_t),
101
sizeof(PGD),
102
+
103
+ 512, 1024, 1536, 2048, 5 * 512, 6 * 512, 7 * 512,
104
+ 1 * 4096, 2 * 4096, 3 * 4096, 4 * 4096, 5 * 4096,
105
+ 6 * 4096, 7 * 4096, 8 * 4096, 9 * 4096, 10 * 4096,
106
+ 11 * 4096, 12 * 4096, 13 * 4096, 14 * 4096, 15 * 4096,
107
+ 16 * 4096, 17 * 4096, 18 * 4096, 19 * 4096, 20 * 4096,
108
+ 21 * 4096, 22 * 4096, 23 * 4096, 24 * 4096, 25 * 4096,
109
};
110
static ARAL **arals = NULL;
111
@@ -131,6 +138,15 @@ void pgd_init_arals(void) {
138
for(size_t i = 0; i < RRD_STORAGE_TIERS ;i++)
139
aral_sizes[i] = tier_page_size[i];
140
141
+ if(!netdata_conf_is_parent()) {
142
+ // this agent is not a parent
143
+ // do not use ARAL for sizes above 4KiB
144
+ for(size_t i = RRD_STORAGE_TIERS ; i < _countof(aral_sizes) ;i++) {
145
+ if(aral_sizes[i] > 4096)
146
+ aral_sizes[i] = 0;
147
+ }
148
+ }
149
+
150
size_t max_delta = 0;
151
for(size_t i = 0; i < aral_sizes_count ;i++) {
152
size_t wanted = aral_sizes[i];
@@ -181,7 +197,7 @@ void pgd_init_arals(void) {
197
0,
198
0,
199
&pgd_aral_statistics,
184
- NULL, NULL, false, false);
200
+ NULL, NULL, false, false, true);
201
}
202
}
203
src/database/engine/pagecache.c
+40
-27
@@ -227,7 +227,7 @@ static size_t get_page_list_from_pgc(PGC *cache, METRIC *metric, struct rrdengin
227
uint32_t dt_s = mrg_metric_get_update_every_s(main_mrg, metric);
228
229
if(!dt_s)
230
- dt_s = default_rrd_update_every;
230
+ dt_s = nd_profile.update_every;
231
232
time_t previous_page_end_time_s = now_s - dt_s;
233
bool first = true;
@@ -402,7 +402,7 @@ static size_t list_has_time_gaps(
402
time_t now_s = wanted_start_time_s;
403
time_t dt_s = mrg_metric_get_update_every_s(main_mrg, metric);
404
if(!dt_s)
405
- dt_s = default_rrd_update_every;
405
+ dt_s = nd_profile.update_every;
406
407
size_t pages_pass2 = 0, pages_pass3 = 0;
408
while((pd = pdc_find_page_for_time(
@@ -656,7 +656,7 @@ static Pvoid_t get_page_list(
656
pages_total = 0;
657
658
size_t cache_gaps = 0, query_gaps = 0;
659
- bool done_v2 = false, done_open = false;
659
+ bool done_v2 = false, done_open = false, done_pass4 = false;
660
661
usec_t pass1_ut = 0, pass2_ut = 0, pass3_ut = 0, pass4_ut = 0, finish_ut = 0;
662
@@ -721,28 +721,36 @@ static Pvoid_t get_page_list(
721
query_gaps = list_has_time_gaps(ctx, metric, JudyL_page_array, wanted_start_time_s, wanted_end_time_s,
722
&pages_total, &pages_found_pass4, pages_to_load_from_disk, &pages_overlapping,
723
optimal_end_time_s, true, common_status);
724
+ done_pass4 = true;
725
726
we_are_done:
727
finish_ut = now_monotonic_usec();
727
- time_delta(finish_ut, pass4_ut);
728
- time_delta(finish_ut, pass3_ut);
729
- time_delta(finish_ut, pass2_ut);
730
- time_delta(finish_ut, pass1_ut);
731
- __atomic_add_fetch(&rrdeng_cache_efficiency_stats.prep_time_in_main_cache_lookup, pass1_ut, __ATOMIC_RELAXED);
732
- __atomic_add_fetch(&rrdeng_cache_efficiency_stats.prep_time_in_open_cache_lookup, pass2_ut, __ATOMIC_RELAXED);
733
- __atomic_add_fetch(&rrdeng_cache_efficiency_stats.prep_time_in_journal_v2_lookup, pass3_ut, __ATOMIC_RELAXED);
734
- __atomic_add_fetch(&rrdeng_cache_efficiency_stats.prep_time_in_pass4_lookup, pass4_ut, __ATOMIC_RELAXED);
735
-
736
- __atomic_add_fetch(&rrdeng_cache_efficiency_stats.queries, 1, __ATOMIC_RELAXED);
728
+ time_delta(finish_ut, pass4_ut); // do not change the order
729
+ time_delta(finish_ut, pass3_ut); // do not change the order
730
+ time_delta(finish_ut, pass2_ut); // do not change the order
731
+ time_delta(finish_ut, pass1_ut); // do not change the order
732
+
733
+ time_and_count_add(&rrdeng_cache_efficiency_stats.prep_time_in_main_cache_lookup, pass1_ut);
734
+
735
+ if(done_open) {
736
+ time_and_count_add(&rrdeng_cache_efficiency_stats.prep_time_in_open_cache_lookup, pass2_ut);
737
+ __atomic_add_fetch(&rrdeng_cache_efficiency_stats.pages_meta_source_open_cache, pages_found_in_open_cache, __ATOMIC_RELAXED);
738
+ }
739
+
740
+ if(done_v2) {
741
+ time_and_count_add(&rrdeng_cache_efficiency_stats.prep_time_in_journal_v2_lookup, pass3_ut);
742
+ __atomic_add_fetch(&rrdeng_cache_efficiency_stats.pages_meta_source_journal_v2, pages_found_in_journals_v2, __ATOMIC_RELAXED);
743
+ }
744
+
745
+ if(done_pass4) {
746
+ time_and_count_add(&rrdeng_cache_efficiency_stats.prep_time_in_pass4_lookup, pass4_ut);
747
+ __atomic_add_fetch(&rrdeng_cache_efficiency_stats.pages_data_source_main_cache_at_pass4, pages_found_pass4, __ATOMIC_RELAXED);
748
+ }
749
+
750
__atomic_add_fetch(&rrdeng_cache_efficiency_stats.queries_planned_with_gaps, (query_gaps) ? 1 : 0, __ATOMIC_RELAXED);
738
- __atomic_add_fetch(&rrdeng_cache_efficiency_stats.queries_open, done_open ? 1 : 0, __ATOMIC_RELAXED);
739
- __atomic_add_fetch(&rrdeng_cache_efficiency_stats.queries_journal_v2, done_v2 ? 1 : 0, __ATOMIC_RELAXED);
751
__atomic_add_fetch(&rrdeng_cache_efficiency_stats.pages_total, pages_total, __ATOMIC_RELAXED);
752
__atomic_add_fetch(&rrdeng_cache_efficiency_stats.pages_meta_source_main_cache, pages_found_in_main_cache, __ATOMIC_RELAXED);
742
- __atomic_add_fetch(&rrdeng_cache_efficiency_stats.pages_meta_source_open_cache, pages_found_in_open_cache, __ATOMIC_RELAXED);
743
- __atomic_add_fetch(&rrdeng_cache_efficiency_stats.pages_meta_source_journal_v2, pages_found_in_journals_v2, __ATOMIC_RELAXED);
753
__atomic_add_fetch(&rrdeng_cache_efficiency_stats.pages_data_source_main_cache, pages_found_in_main_cache, __ATOMIC_RELAXED);
745
- __atomic_add_fetch(&rrdeng_cache_efficiency_stats.pages_data_source_main_cache_at_pass4, pages_found_pass4, __ATOMIC_RELAXED);
754
__atomic_add_fetch(&rrdeng_cache_efficiency_stats.pages_to_load_from_disk, *pages_to_load_from_disk, __ATOMIC_RELAXED);
755
__atomic_add_fetch(&rrdeng_cache_efficiency_stats.pages_overlapping_skipped, pages_overlapping, __ATOMIC_RELAXED);
756
@@ -754,7 +762,7 @@ inline void rrdeng_prep_wait(PDC *pdc) {
762
usec_t started_ut = now_monotonic_usec();
763
completion_wait_for(&pdc->prep_completion);
764
pdc->prep_done = true;
757
- __atomic_add_fetch(&rrdeng_cache_efficiency_stats.query_time_wait_for_prep, now_monotonic_usec() - started_ut, __ATOMIC_RELAXED);
765
+ time_and_count_add(&rrdeng_cache_efficiency_stats.query_time_wait_for_prep, now_monotonic_usec() - started_ut);
766
}
767
}
768
@@ -781,13 +789,18 @@ void rrdeng_prep_query(struct page_details_control *pdc, bool worker) {
789
if (pdc->pages_to_load_from_disk && pdc->page_list_JudyL) {
790
pdc_acquire(pdc); // we get 1 for the 1st worker in the chain: do_read_page_list_work()
791
usec_t start_ut = now_monotonic_usec();
784
- if(likely(pdc->priority == STORAGE_PRIORITY_SYNCHRONOUS))
792
+ if(likely(pdc->priority == STORAGE_PRIORITY_SYNCHRONOUS)) {
793
pdc_route_synchronously(pdc->ctx, pdc);
786
- else if(likely(pdc->priority == STORAGE_PRIORITY_SYNCHRONOUS_FIRST))
794
+ time_and_count_add(&rrdeng_cache_efficiency_stats.prep_time_to_route_sync, now_monotonic_usec() - start_ut);
795
+ }
796
+ else if(likely(pdc->priority == STORAGE_PRIORITY_SYNCHRONOUS_FIRST)) {
797
pdc_route_synchronously_first(pdc->ctx, pdc);
788
- else
798
+ time_and_count_add(&rrdeng_cache_efficiency_stats.prep_time_to_route_syncfirst, now_monotonic_usec() - start_ut);
799
+ }
800
+ else {
801
pdc_route_asynchronously(pdc->ctx, pdc);
790
- __atomic_add_fetch(&rrdeng_cache_efficiency_stats.prep_time_to_route, now_monotonic_usec() - start_ut, __ATOMIC_RELAXED);
802
+ time_and_count_add(&rrdeng_cache_efficiency_stats.prep_time_to_route_async, now_monotonic_usec() - start_ut);
803
+ }
804
}
805
else
806
completion_mark_complete(&pdc->page_completion);
@@ -973,15 +986,15 @@ struct pgc_page *pg_cache_lookup_next(
986
987
if(waited) {
988
if(preloaded)
976
- __atomic_add_fetch(&rrdeng_cache_efficiency_stats.query_time_to_slow_preload_next_page, now_monotonic_usec() - start_ut, __ATOMIC_RELAXED);
989
+ time_and_count_add(&rrdeng_cache_efficiency_stats.query_time_to_slow_preload_next_page, now_monotonic_usec() - start_ut);
990
else
978
- __atomic_add_fetch(&rrdeng_cache_efficiency_stats.query_time_to_slow_disk_next_page, now_monotonic_usec() - start_ut, __ATOMIC_RELAXED);
991
+ time_and_count_add(&rrdeng_cache_efficiency_stats.query_time_to_slow_disk_next_page, now_monotonic_usec() - start_ut);
992
}
993
else {
994
if(preloaded)
982
- __atomic_add_fetch(&rrdeng_cache_efficiency_stats.query_time_to_fast_preload_next_page, now_monotonic_usec() - start_ut, __ATOMIC_RELAXED);
995
+ time_and_count_add(&rrdeng_cache_efficiency_stats.query_time_to_fast_preload_next_page, now_monotonic_usec() - start_ut);
996
else
984
- __atomic_add_fetch(&rrdeng_cache_efficiency_stats.query_time_to_fast_disk_next_page, now_monotonic_usec() - start_ut, __ATOMIC_RELAXED);
997
+ time_and_count_add(&rrdeng_cache_efficiency_stats.query_time_to_fast_disk_next_page, now_monotonic_usec() - start_ut);
998
}
999
1000
return page;
src/database/engine/pdc.c
+55
-30
@@ -55,7 +55,7 @@ void pdc_init(void) {
55
0,
56
0,
57
NULL,
58
- NULL, NULL, false, false
58
+ NULL, NULL, false, false, true
59
);
60
61
pulse_aral_register(pdc_globals.pdc.ar, "pdc");
@@ -85,7 +85,7 @@ void page_details_init(void) {
85
0,
86
0,
87
NULL,
88
- NULL, NULL, false, false
88
+ NULL, NULL, false, false, true
89
);
90
pulse_aral_register(pdc_globals.pd.ar, "pd");
91
}
@@ -114,7 +114,7 @@ void epdl_init(void) {
114
0,
115
0,
116
NULL,
117
- NULL, NULL, false, false
117
+ NULL, NULL, false, false, true
118
);
119
pulse_aral_register(pdc_globals.epdl.ar, "epdl");
120
}
@@ -143,7 +143,7 @@ void deol_init(void) {
143
0,
144
0,
145
NULL,
146
- NULL, NULL, false, false
146
+ NULL, NULL, false, false, true
147
);
148
149
pulse_aral_register(pdc_globals.deol.ar, "deol");
@@ -455,50 +455,75 @@ void epdl_cmd_dequeued(void *epdl_ptr) {
455
epdl->cmd = NULL;
456
}
457
458
-static struct rrdeng_cmd *epdl_get_cmd(void *epdl_ptr) {
458
+static inline struct rrdeng_cmd *epdl_get_cmd(void *epdl_ptr) {
459
EPDL *epdl = epdl_ptr;
460
return epdl->cmd;
461
}
462
463
-static bool epdl_pending_add(EPDL *epdl) {
464
- bool added_new;
463
+static EPDL_EXTENT *epdl_find_extent_base(EPDL *epdl) {
464
+ EPDL_EXTENT *e = NULL;
465
+ rw_spinlock_read_lock(&epdl->datafile->extent_epdl.spinlock);
466
+ Pvoid_t *PValue = JudyLGet(epdl->datafile->extent_epdl.epdl_per_extent, epdl->extent_offset, PJE0);
467
+ internal_fatal(PValue == PJERR, "DBENGINE: corrupted pending extent judy");
468
+ if(PValue)
469
+ e = *PValue;
470
+ rw_spinlock_read_unlock(&epdl->datafile->extent_epdl.spinlock);
471
+
472
+ if(!e) {
473
+ EPDL_EXTENT *e_to_free = NULL;
474
+ e = callocz(1, sizeof(*e));
475
+
476
+ rw_spinlock_write_lock(&epdl->datafile->extent_epdl.spinlock);
477
+ Pvoid_t *PValue = JudyLIns(&epdl->datafile->extent_epdl.epdl_per_extent, epdl->extent_offset, PJE0);
478
+ internal_fatal(!PValue || PValue == PJERR, "DBENGINE: corrupted pending extent judy");
479
+ if(!*PValue) {
480
+ *PValue = e;
481
+ spinlock_init(&e->spinlock);
482
+ }
483
+ else {
484
+ e_to_free = e;
485
+ e = *PValue;
486
+ }
487
+ rw_spinlock_write_unlock(&epdl->datafile->extent_epdl.spinlock);
488
466
- spinlock_lock(&epdl->datafile->extent_queries.spinlock);
467
- Pvoid_t *PValue = JudyLIns(&epdl->datafile->extent_queries.pending_epdl_by_extent_offset_judyL, epdl->extent_offset, PJE0);
468
- internal_fatal(!PValue || PValue == PJERR, "DBENGINE: corrupted pending extent judy");
489
+ freez(e_to_free);
490
+ }
491
470
- EPDL *base = *PValue;
492
+ return e;
493
+}
494
472
- if(!base) {
473
- added_new = true;
474
- epdl->head_to_datafile_extent_queries_pending_for_extent = true;
475
- }
476
- else {
495
+static bool epdl_pending_add(EPDL *epdl) {
496
+ EPDL_EXTENT *e = epdl_find_extent_base(epdl);
497
+ spinlock_lock(&e->spinlock);
498
+
499
+ bool added_new;
500
+ if(unlikely(e->base)) {
501
added_new = false;
502
epdl->head_to_datafile_extent_queries_pending_for_extent = false;
503
+
504
__atomic_add_fetch(&rrdeng_cache_efficiency_stats.pages_load_extent_merged, 1, __ATOMIC_RELAXED);
505
481
- if(base->pdc->priority > epdl->pdc->priority)
482
- rrdeng_req_cmd(epdl_get_cmd, base, epdl->pdc->priority);
506
+// if(e->base->pdc->priority > epdl->pdc->priority) {
507
+// e->base->pdc->priority = epdl->pdc->priority;
508
+// rrdeng_req_cmd(epdl_get_cmd, e->base, epdl->pdc->priority);
509
+// }
510
+ }
511
+ else {
512
+ added_new = true;
513
+ epdl->head_to_datafile_extent_queries_pending_for_extent = true;
514
}
515
485
- DOUBLE_LINKED_LIST_APPEND_ITEM_UNSAFE(base, epdl, query.prev, query.next);
486
- *PValue = base;
487
-
488
- spinlock_unlock(&epdl->datafile->extent_queries.spinlock);
516
+ DOUBLE_LINKED_LIST_APPEND_ITEM_UNSAFE(e->base, epdl, query.prev, query.next);
517
+ spinlock_unlock(&e->spinlock);
518
519
return added_new;
520
}
521
522
static void epdl_pending_del(EPDL *epdl) {
494
- spinlock_lock(&epdl->datafile->extent_queries.spinlock);
495
- if(epdl->head_to_datafile_extent_queries_pending_for_extent) {
496
- epdl->head_to_datafile_extent_queries_pending_for_extent = false;
497
- int rc = JudyLDel(&epdl->datafile->extent_queries.pending_epdl_by_extent_offset_judyL, epdl->extent_offset, PJE0);
498
- (void) rc;
499
- internal_fatal(!rc, "DBENGINE: epdl not found in pending list");
500
- }
501
- spinlock_unlock(&epdl->datafile->extent_queries.spinlock);
523
+ EPDL_EXTENT *e = epdl_find_extent_base(epdl);
524
+ spinlock_lock(&e->spinlock);
525
+ e->base = NULL;
526
+ spinlock_unlock(&e->spinlock);
527
}
528
529
void pdc_to_epdl_router(struct rrdengine_instance *ctx, PDC *pdc, execute_extent_page_details_list_t exec_first_extent_list, execute_extent_page_details_list_t exec_rest_extent_list)
src/database/engine/rrdengine.c
+26
-10
@@ -142,7 +142,7 @@ static void work_request_init(void) {
142
0,
143
0,
144
NULL,
145
- NULL, NULL, false, false
145
+ NULL, NULL, false, false, true
146
);
147
148
pulse_aral_register(rrdeng_main.work_cmd.ar, "workers");
@@ -263,7 +263,7 @@ void page_descriptors_init(void) {
263
0,
264
0,
265
NULL,
266
- NULL, NULL, false, false);
266
+ NULL, NULL, false, false, true);
267
268
pulse_aral_register(rrdeng_main.descriptors.ar, "descriptors");
269
}
@@ -288,7 +288,7 @@ static void extent_io_descriptor_init(void) {
288
0,
289
0,
290
NULL,
291
- NULL, NULL, false, false
291
+ NULL, NULL, false, false, true
292
);
293
294
pulse_aral_register(rrdeng_main.xt_io_descr.ar, "extent io");
@@ -314,7 +314,7 @@ void rrdeng_query_handle_init(void) {
314
0,
315
0,
316
NULL,
317
- NULL, NULL, false, false);
317
+ NULL, NULL, false, false, true);
318
319
pulse_aral_register(rrdeng_main.handles.ar, "query handles");
320
}
@@ -359,7 +359,7 @@ static void wal_cleanup1(void) {
359
if(!spinlock_trylock(&wal_globals.protected.spinlock))
360
return;
361
362
- if(wal_globals.protected.available_items && wal_globals.protected.available > storage_tiers) {
362
+ if(wal_globals.protected.available_items && wal_globals.protected.available > nd_profile.storage_tiers) {
363
wal = wal_globals.protected.available_items;
364
DOUBLE_LINKED_LIST_REMOVE_ITEM_UNSAFE(wal_globals.protected.available_items, wal, cache.prev, cache.next);
365
wal_globals.protected.available--;
@@ -436,7 +436,7 @@ static void rrdeng_cmd_queue_init(void) {
436
0,
437
0,
438
NULL,
439
- NULL, NULL, false, false);
439
+ NULL, NULL, false, false, true);
440
441
pulse_aral_register(rrdeng_main.cmd_queue.ar, "opcodes");
442
}
@@ -1289,6 +1289,22 @@ void datafile_delete(struct rrdengine_instance *ctx, struct rrdengine_datafile *
1289
netdata_log_info("DBENGINE: deleted data file \"%s\".", path);
1290
deleted_bytes += datafile_bytes;
1291
}
1292
+
1293
+ {
1294
+ rw_spinlock_write_lock(&datafile->extent_epdl.spinlock);
1295
+ bool first = true;
1296
+ Word_t idx = 0;
1297
+ Pvoid_t *PValue;
1298
+ while ((PValue = JudyLFirstThenNext(datafile->extent_epdl.epdl_per_extent, &idx, &first))) {
1299
+ EPDL_EXTENT *e = *PValue;
1300
+ internal_fatal(e->base, "The should not be any EPDLs ");
1301
+ freez(e);
1302
+ *PValue = NULL;
1303
+ }
1304
+ JudyLFreeArray(&datafile->extent_epdl.epdl_per_extent, PJE0);
1305
+ rw_spinlock_write_unlock(&datafile->extent_epdl.spinlock);
1306
+ }
1307
+
1308
freez(journal_file);
1309
freez(datafile);
1310
@@ -1694,7 +1710,7 @@ static void retention_timer_cb(uv_timer_t *handle) {
1710
uv_stop(handle->loop);
1711
uv_update_time(handle->loop);
1712
1697
- for (size_t tier = 0; tier < storage_tiers; tier++) {
1713
+ for (size_t tier = 0; tier < nd_profile.storage_tiers; tier++) {
1714
STORAGE_ENGINE *eng = localhost->db[tier].eng;
1715
if (!eng || eng->seb != STORAGE_ENGINE_BACKEND_DBENGINE)
1716
continue;
@@ -1831,7 +1847,7 @@ void calculate_tier_disk_space_percentage(void)
1847
return;
1848
1849
uint64_t total_diskspace = 0;
1834
- for(size_t tier = 0; tier < storage_tiers ;tier++) {
1850
+ for(size_t tier = 0; tier < nd_profile.storage_tiers;tier++) {
1851
STORAGE_ENGINE *eng = localhost->db[tier].eng;
1852
if (!eng || eng->seb != STORAGE_ENGINE_BACKEND_DBENGINE) {
1853
tier_space[tier] = 0;
@@ -1845,7 +1861,7 @@ void calculate_tier_disk_space_percentage(void)
1861
}
1862
1863
if (total_diskspace) {
1848
- for (size_t tier = 0; tier < storage_tiers; tier++) {
1864
+ for (size_t tier = 0; tier < nd_profile.storage_tiers; tier++) {
1865
multidb_ctx[tier]->config.disk_percentage = (100 * tier_space[tier] / total_diskspace);
1866
}
1867
}
@@ -1861,7 +1877,7 @@ void dbengine_retention_statistics(void)
1877
1878
calculate_tier_disk_space_percentage();
1879
1864
- for (size_t tier = 0; tier < storage_tiers; tier++) {
1880
+ for (size_t tier = 0; tier < nd_profile.storage_tiers; tier++) {
1881
STORAGE_ENGINE *eng = localhost->db[tier].eng;
1882
if (!eng || eng->seb != STORAGE_ENGINE_BACKEND_DBENGINE)
1883
continue;
src/database/engine/rrdengine.h
+39
-71
@@ -327,57 +327,38 @@ void wal_release(WAL *wal);
327
* They only describe operations since DB engine instance load time.
328
*/
329
struct rrdengine_statistics {
330
- CACHE_LINE_PADDING();
331
- rrdeng_stats_t before_decompress_bytes;
332
- CACHE_LINE_PADDING();
333
- rrdeng_stats_t after_decompress_bytes;
334
- CACHE_LINE_PADDING();
335
- rrdeng_stats_t before_compress_bytes;
336
- CACHE_LINE_PADDING();
337
- rrdeng_stats_t after_compress_bytes;
338
-
339
- CACHE_LINE_PADDING();
340
- rrdeng_stats_t io_write_bytes;
341
- CACHE_LINE_PADDING();
342
- rrdeng_stats_t io_write_requests;
343
- CACHE_LINE_PADDING();
344
- rrdeng_stats_t io_read_bytes;
345
- CACHE_LINE_PADDING();
346
- rrdeng_stats_t io_read_requests;
347
-
348
- CACHE_LINE_PADDING();
349
- rrdeng_stats_t datafile_creations;
350
- CACHE_LINE_PADDING();
351
- rrdeng_stats_t datafile_deletions;
352
- CACHE_LINE_PADDING();
353
- rrdeng_stats_t journalfile_creations;
354
- CACHE_LINE_PADDING();
355
- rrdeng_stats_t journalfile_deletions;
356
-
357
- CACHE_LINE_PADDING();
358
- rrdeng_stats_t io_errors;
359
- CACHE_LINE_PADDING();
360
- rrdeng_stats_t fs_errors;
330
+ PAD64(rrdeng_stats_t) before_decompress_bytes;
331
+ PAD64(rrdeng_stats_t) after_decompress_bytes;
332
+ PAD64(rrdeng_stats_t) before_compress_bytes;
333
+ PAD64(rrdeng_stats_t) after_compress_bytes;
334
+
335
+ PAD64(rrdeng_stats_t) io_write_bytes;
336
+ PAD64(rrdeng_stats_t) io_write_requests;
337
+ PAD64(rrdeng_stats_t) io_read_bytes;
338
+ PAD64(rrdeng_stats_t) io_read_requests;
339
+
340
+ PAD64(rrdeng_stats_t) datafile_creations;
341
+ PAD64(rrdeng_stats_t) datafile_deletions;
342
+ PAD64(rrdeng_stats_t) journalfile_creations;
343
+ PAD64(rrdeng_stats_t) journalfile_deletions;
344
+
345
+ PAD64(rrdeng_stats_t) io_errors;
346
+ PAD64(rrdeng_stats_t) fs_errors;
347
};
348
349
struct rrdeng_global_stats {
364
- CACHE_LINE_PADDING();
350
/* I/O errors global counter */
366
- rrdeng_stats_t global_io_errors;
351
+ PAD64(rrdeng_stats_t) global_io_errors;
352
368
- CACHE_LINE_PADDING();
353
/* File-System errors global counter */
370
- rrdeng_stats_t global_fs_errors;
354
+ PAD64(rrdeng_stats_t) global_fs_errors;
355
372
- CACHE_LINE_PADDING();
356
/* number of File-Descriptors that have been reserved by dbengine */
374
- rrdeng_stats_t rrdeng_reserved_file_descriptors;
357
+ PAD64(rrdeng_stats_t) rrdeng_reserved_file_descriptors;
358
376
- CACHE_LINE_PADDING();
359
/* inability to flush global counters */
378
- rrdeng_stats_t global_pg_cache_over_half_dirty_events;
379
- CACHE_LINE_PADDING();
380
- rrdeng_stats_t global_flushing_pressure_page_deletions; /* number of deleted pages */
360
+ PAD64(rrdeng_stats_t) global_pg_cache_over_half_dirty_events;
361
+ PAD64(rrdeng_stats_t) global_flushing_pressure_page_deletions; /* number of deleted pages */
362
};
363
364
extern struct rrdeng_global_stats global_stats;
@@ -413,36 +394,23 @@ struct rrdengine_instance {
394
} njfv2idx;
395
396
struct {
416
- CACHE_LINE_PADDING();
417
- unsigned last_fileno; // newest index of datafile and journalfile
418
- CACHE_LINE_PADDING();
419
- unsigned last_flush_fileno; // newest index of datafile received data
420
-
421
- CACHE_LINE_PADDING();
422
- size_t collectors_running;
423
- CACHE_LINE_PADDING();
424
- size_t collectors_running_duplicate;
425
- CACHE_LINE_PADDING();
426
- size_t inflight_queries; // the number of queries currently running
427
- CACHE_LINE_PADDING();
428
- uint64_t current_disk_space; // the current disk space size used
429
-
430
- CACHE_LINE_PADDING();
431
- uint64_t transaction_id; // the transaction id of the next extent flushing
432
-
433
- CACHE_LINE_PADDING();
434
- bool migration_to_v2_running;
435
- CACHE_LINE_PADDING();
436
- bool now_deleting_files;
437
- CACHE_LINE_PADDING();
438
- unsigned extents_currently_being_flushed; // non-zero until we commit data to disk (both datafile and journal file)
439
-
440
- CACHE_LINE_PADDING();
441
- time_t first_time_s;
442
- CACHE_LINE_PADDING();
443
- uint64_t metrics;
444
- CACHE_LINE_PADDING();
445
- uint64_t samples;
397
+ PAD64(unsigned) last_fileno; // newest index of datafile and journalfile
398
+ PAD64(unsigned) last_flush_fileno; // newest index of datafile received data
399
+
400
+ PAD64(size_t) collectors_running;
401
+ PAD64(size_t) collectors_running_duplicate;
402
+ PAD64(size_t) inflight_queries; // the number of queries currently running
403
+ PAD64(uint64_t) current_disk_space; // the current disk space size used
404
+
405
+ PAD64(uint64_t) transaction_id; // the transaction id of the next extent flushing
406
+
407
+ PAD64(bool) migration_to_v2_running;
408
+ PAD64(bool) now_deleting_files;
409
+ PAD64(unsigned) extents_currently_being_flushed; // non-zero until we commit data to disk (both datafile and journal file)
410
+
411
+ PAD64(time_t) first_time_s;
412
+ PAD64(uint64_t) metrics;
413
+ PAD64(uint64_t) samples;
414
} atomic;
415
416
struct {
src/database/engine/rrdengineapi.c
+6
-6
@@ -761,8 +761,8 @@ void rrdeng_load_metric_init(STORAGE_METRIC_HANDLE *smh,
761
762
handle->dt_s = db_update_every_s;
763
if (!handle->dt_s) {
764
- handle->dt_s = default_rrd_update_every;
765
- mrg_metric_set_update_every_s_if_zero(main_mrg, metric, default_rrd_update_every);
764
+ handle->dt_s = nd_profile.update_every;
765
+ mrg_metric_set_update_every_s_if_zero(main_mrg, metric, nd_profile.update_every);
766
}
767
768
seqh->handle = (STORAGE_QUERY_HANDLE *) handle;
@@ -773,7 +773,7 @@ void rrdeng_load_metric_init(STORAGE_METRIC_HANDLE *smh,
773
774
pg_cache_preload(handle);
775
776
- __atomic_add_fetch(&rrdeng_cache_efficiency_stats.query_time_init, now_monotonic_usec() - started_ut, __ATOMIC_RELAXED);
776
+ time_and_count_add(&rrdeng_cache_efficiency_stats.query_time_init, now_monotonic_usec() - started_ut);
777
}
778
else {
779
handle->start_time_s = start_time_s;
@@ -1065,7 +1065,7 @@ static void rrdeng_populate_mrg(struct rrdengine_instance *ctx) {
1065
datafiles++;
1066
uv_rwlock_rdunlock(&ctx->datafiles.rwlock);
1067
1068
- ssize_t cpus = (ssize_t)netdata_conf_cpus() / (ssize_t)storage_tiers;
1068
+ ssize_t cpus = (ssize_t)netdata_conf_cpus() / (ssize_t)nd_profile.storage_tiers;
1069
if(cpus > (ssize_t)datafiles)
1070
cpus = (ssize_t)datafiles;
1071
@@ -1293,7 +1293,7 @@ static void populate_v2_statistics(struct rrdengine_datafile *datafile, RRDENG_S
1293
if(likely(points > 1))
1294
update_every_s = (time_t) ((end_time_s - start_time_s) / (points - 1));
1295
else {
1296
- update_every_s = (time_t) (default_rrd_update_every * get_tier_grouping(datafile->ctx->config.tier));
1296
+ update_every_s = (time_t) (nd_profile.update_every * get_tier_grouping(datafile->ctx->config.tier));
1297
stats->single_point_pages++;
1298
}
1299
@@ -1375,7 +1375,7 @@ RRDENG_SIZE_STATS rrdeng_size_statistics(struct rrdengine_instance *ctx) {
1375
// stats.sizeof_metric_in_index = 40;
1376
// stats.sizeof_page_in_index = 24;
1377
1378
- stats.default_granularity_secs = (size_t)default_rrd_update_every * get_tier_grouping(ctx->config.tier);
1378
+ stats.default_granularity_secs = (size_t)nd_profile.update_every * get_tier_grouping(ctx->config.tier);
1379
1380
return stats;
1381
}
src/database/engine/rrdengineapi.h
+66
-57
@@ -128,83 +128,92 @@ typedef struct rrdengine_size_statistics {
128
double average_page_size_bytes;
129
} RRDENG_SIZE_STATS;
130
131
+struct time_and_count {
132
+ size_t count;
133
+ usec_t usec;
134
+};
135
+
136
+static inline void time_and_count_add(struct time_and_count *tc, usec_t dt) {
137
+ __atomic_add_fetch(&tc->count, 1, __ATOMIC_RELAXED);
138
+ __atomic_add_fetch(&tc->usec, dt, __ATOMIC_RELAXED);
139
+}
140
+
141
struct rrdeng_cache_efficiency_stats {
132
- size_t queries;
133
- size_t queries_planned_with_gaps;
134
- size_t queries_executed_with_gaps;
135
- size_t queries_open;
136
- size_t queries_journal_v2;
142
+ PAD64(size_t) queries_planned_with_gaps;
143
+ PAD64(size_t) queries_executed_with_gaps;
144
138
- size_t currently_running_queries;
145
+ PAD64(size_t) currently_running_queries;
146
147
// query planner output of the queries
141
- size_t pages_total;
142
- size_t pages_to_load_from_disk;
143
- size_t extents_loaded_from_disk;
148
+ PAD64(size_t) pages_total;
149
+ PAD64(size_t) pages_to_load_from_disk;
150
+ PAD64(size_t) extents_loaded_from_disk;
151
152
// pages metadata sources
146
- size_t pages_meta_source_main_cache;
147
- size_t pages_meta_source_open_cache;
148
- size_t pages_meta_source_journal_v2;
153
+ PAD64(size_t) pages_meta_source_main_cache;
154
+ PAD64(size_t) pages_meta_source_open_cache;
155
+ PAD64(size_t) pages_meta_source_journal_v2;
156
157
// preloading
151
- size_t page_next_wait_failed;
152
- size_t page_next_wait_loaded;
153
- size_t page_next_nowait_failed;
154
- size_t page_next_nowait_loaded;
158
+ PAD64(size_t) page_next_wait_failed;
159
+ PAD64(size_t) page_next_wait_loaded;
160
+ PAD64(size_t) page_next_nowait_failed;
161
+ PAD64(size_t) page_next_nowait_loaded;
162
163
// pages data sources
157
- size_t pages_data_source_main_cache;
158
- size_t pages_data_source_main_cache_at_pass4;
159
- size_t pages_data_source_disk;
160
- size_t pages_data_source_extent_cache; // loaded by a cached extent
164
+ PAD64(size_t) pages_data_source_main_cache;
165
+ PAD64(size_t) pages_data_source_main_cache_at_pass4;
166
+ PAD64(size_t) pages_data_source_disk;
167
+ PAD64(size_t) pages_data_source_extent_cache; // loaded by a cached extent
168
169
// cache hits at different points
163
- size_t pages_load_ok_loaded_but_cache_hit_while_inserting; // found in cache while inserting it (conflict)
170
+ PAD64(size_t) pages_load_ok_loaded_but_cache_hit_while_inserting; // found in cache while inserting it (conflict)
171
172
// loading
166
- size_t pages_load_extent_merged;
167
- size_t pages_load_ok_uncompressed;
168
- size_t pages_load_ok_compressed;
169
- size_t pages_load_fail_invalid_page_in_extent;
170
- size_t pages_load_fail_cant_mmap_extent;
171
- size_t pages_load_fail_datafile_not_available;
172
- size_t pages_load_fail_unroutable;
173
- size_t pages_load_fail_not_found;
174
- size_t pages_load_fail_invalid_extent;
175
- size_t pages_load_fail_cancelled;
176
-
177
- // timings for query preparation
178
- size_t prep_time_to_route;
179
- size_t prep_time_in_main_cache_lookup;
180
- size_t prep_time_in_open_cache_lookup;
181
- size_t prep_time_in_journal_v2_lookup;
182
- size_t prep_time_in_pass4_lookup;
173
+ PAD64(size_t) pages_load_extent_merged;
174
+ PAD64(size_t) pages_load_ok_uncompressed;
175
+ PAD64(size_t) pages_load_ok_compressed;
176
+ PAD64(size_t) pages_load_fail_invalid_page_in_extent;
177
+ PAD64(size_t) pages_load_fail_cant_mmap_extent;
178
+ PAD64(size_t) pages_load_fail_datafile_not_available;
179
+ PAD64(size_t) pages_load_fail_unroutable;
180
+ PAD64(size_t) pages_load_fail_not_found;
181
+ PAD64(size_t) pages_load_fail_invalid_extent;
182
+ PAD64(size_t) pages_load_fail_cancelled;
183
+
184
+ // count of queries and times spent in them
185
+ PAD64(struct time_and_count) prep_time_to_route_sync;
186
+ PAD64(struct time_and_count) prep_time_to_route_syncfirst;
187
+ PAD64(struct time_and_count) prep_time_to_route_async;
188
+ PAD64(struct time_and_count) prep_time_in_main_cache_lookup;
189
+ PAD64(struct time_and_count) prep_time_in_open_cache_lookup;
190
+ PAD64(struct time_and_count) prep_time_in_journal_v2_lookup;
191
+ PAD64(struct time_and_count) prep_time_in_pass4_lookup;
192
193
// timings the query thread experiences
185
- size_t query_time_init;
186
- size_t query_time_wait_for_prep;
187
- size_t query_time_to_slow_disk_next_page;
188
- size_t query_time_to_fast_disk_next_page;
189
- size_t query_time_to_slow_preload_next_page;
190
- size_t query_time_to_fast_preload_next_page;
194
+ PAD64(struct time_and_count) query_time_init;
195
+ PAD64(struct time_and_count) query_time_wait_for_prep;
196
+ PAD64(struct time_and_count) query_time_to_slow_disk_next_page;
197
+ PAD64(struct time_and_count) query_time_to_fast_disk_next_page;
198
+ PAD64(struct time_and_count) query_time_to_slow_preload_next_page;
199
+ PAD64(struct time_and_count) query_time_to_fast_preload_next_page;
200
201
// query issues
193
- size_t pages_zero_time_skipped;
194
- size_t pages_past_time_skipped;
195
- size_t pages_overlapping_skipped;
196
- size_t pages_invalid_size_skipped;
197
- size_t pages_invalid_update_every_fixed;
198
- size_t pages_invalid_entries_fixed;
202
+ PAD64(size_t) pages_zero_time_skipped;
203
+ PAD64(size_t) pages_past_time_skipped;
204
+ PAD64(size_t) pages_overlapping_skipped;
205
+ PAD64(size_t) pages_invalid_size_skipped;
206
+ PAD64(size_t) pages_invalid_update_every_fixed;
207
+ PAD64(size_t) pages_invalid_entries_fixed;
208
209
// database events
201
- size_t journal_v2_mapped;
202
- size_t journal_v2_unmapped;
203
- size_t datafile_creation_started;
204
- size_t datafile_deletion_started;
205
- size_t datafile_deletion_spin;
206
- size_t journal_v2_indexing_started;
207
- size_t metrics_retention_started;
210
+ PAD64(size_t) journal_v2_mapped;
211
+ PAD64(size_t) journal_v2_unmapped;
212
+ PAD64(size_t) datafile_creation_started;
213
+ PAD64(size_t) datafile_deletion_started;
214
+ PAD64(size_t) datafile_deletion_spin;
215
+ PAD64(size_t) journal_v2_indexing_started;
216
+ PAD64(size_t) metrics_retention_started;
217
};
218
219
typedef enum rrdeng_mem {
src/database/ram/rrddim_mem.c
+4
-4
@@ -54,7 +54,7 @@ rrddim_metric_get_or_create(RRDDIM *rd, STORAGE_INSTANCE *si __maybe_unused) {
54
netdata_rwlock_wrlock(&rrddim_JudyHS_rwlock);
55
JudyAllocThreadPulseReset();
56
Pvoid_t *PValue = JudyHSIns(&rrddim_JudyHS_array, &rd->metric_uuid, sizeof(nd_uuid_t), PJE0);
57
- int64_t mem = JudyAllocThreadPulseGetAndReset();
57
+ int64_t judy_mem = JudyAllocThreadPulseGetAndReset();
58
mh = *PValue;
59
if(!mh) {
60
mh = callocz(1, sizeof(struct mem_metric_handle));
@@ -62,7 +62,7 @@ rrddim_metric_get_or_create(RRDDIM *rd, STORAGE_INSTANCE *si __maybe_unused) {
62
mh->refcount = 1;
63
update_metric_handle_from_rrddim(mh, rd);
64
*PValue = mh;
65
- pulse_db_rrd_memory_change(mem + (int64_t)sizeof(struct mem_metric_handle));
65
+ pulse_db_rrd_memory_change(judy_mem + (int64_t)sizeof(struct mem_metric_handle));
66
}
67
else {
68
if(__atomic_add_fetch(&mh->refcount, 1, __ATOMIC_RELAXED) <= 0)
@@ -111,11 +111,11 @@ void rrddim_metric_release(STORAGE_METRIC_HANDLE *smh __maybe_unused) {
111
netdata_rwlock_wrlock(&rrddim_JudyHS_rwlock);
112
JudyAllocThreadPulseReset();
113
JudyHSDel(&rrddim_JudyHS_array, &rd->metric_uuid, sizeof(nd_uuid_t), PJE0);
114
- int64_t mem = JudyAllocThreadPulseGetAndReset();
114
+ int64_t judy_mem = JudyAllocThreadPulseGetAndReset();
115
netdata_rwlock_wrunlock(&rrddim_JudyHS_rwlock);
116
117
freez(mh);
118
- pulse_db_rrd_memory_change(mem - (int64_t)sizeof(struct mem_metric_handle));
118
+ pulse_db_rrd_memory_change(judy_mem - (int64_t)sizeof(struct mem_metric_handle));
119
}
120
}
121
}
src/database/rrd.h
+50
-33
@@ -11,7 +11,7 @@ extern "C" {
11
#include "rrd-database-mode.h"
12
#include "streaming/stream-traffic-types.h"
13
#include "streaming/stream-sender-commit.h"
14
-#include "rrdhost-state-id.h"
14
+#include "streaming/stream-replication-tracking.h"
15
#include "health/health-alert-log.h"
16
#include "rrdhost-system-info.h"
17
@@ -115,11 +115,13 @@ typedef enum __attribute__ ((__packed__)) rrdset_flags {
115
RRDSET_FLAG_RECEIVER_REPLICATION_IN_PROGRESS = (1 << 21), // the receiving side has replication in progress
116
RRDSET_FLAG_RECEIVER_REPLICATION_FINISHED = (1 << 22), // the receiving side has completed replication
117
118
- RRDSET_FLAG_UPSTREAM_SEND_VARIABLES = (1 << 23), // a custom variable has been updated and needs to be exposed to parent
118
+ RRDSET_FLAG_BACKFILLED_HIGH_TIERS = (1 << 23), // we have backfilled this chart
119
120
- RRDSET_FLAG_COLLECTION_FINISHED = (1 << 24), // when set, data collection is not available for this chart
120
+ RRDSET_FLAG_UPSTREAM_SEND_VARIABLES = (1 << 24), // a custom variable has been updated and needs to be exposed to parent
121
122
- RRDSET_FLAG_HAS_RRDCALC_LINKED = (1 << 25), // this chart has at least one rrdcal linked
122
+ RRDSET_FLAG_COLLECTION_FINISHED = (1 << 25), // when set, data collection is not available for this chart
123
+
124
+ RRDSET_FLAG_HAS_RRDCALC_LINKED = (1 << 26), // this chart has at least one rrdcal linked
125
} RRDSET_FLAGS;
126
127
#include "daemon/common.h"
@@ -178,7 +180,7 @@ typedef enum __attribute__ ((__packed__)) {
180
RRD_BACKFILL_NEW
181
} RRD_BACKFILL;
182
181
-#define UPDATE_EVERY 1
183
+#define UPDATE_EVERY_MIN 1
184
#define UPDATE_EVERY_MAX 3600
185
186
#define RRD_DEFAULT_HISTORY_ENTRIES 3600
@@ -256,10 +258,11 @@ typedef enum __attribute__ ((__packed__)) rrddim_flags {
258
// this is 8 bit
259
} RRDDIM_FLAGS;
260
259
-#define rrddim_flag_get(rd) __atomic_load_n(&((rd)->flags), __ATOMIC_ACQUIRE)
260
-#define rrddim_flag_check(rd, flag) (__atomic_load_n(&((rd)->flags), __ATOMIC_ACQUIRE) & (flag))
261
-#define rrddim_flag_set(rd, flag) __atomic_or_fetch(&((rd)->flags), (flag), __ATOMIC_RELEASE)
262
-#define rrddim_flag_clear(rd, flag) __atomic_and_fetch(&((rd)->flags), ~(flag), __ATOMIC_RELEASE)
261
+#define rrddim_flag_get(rd) atomic_flags_get(&((rd)->flags))
262
+#define rrddim_flag_check(rd, flag) atomic_flags_check(&((rd)->flags), flag)
263
+#define rrddim_flag_set(rd, flag) atomic_flags_set(&((rd)->flags), flag)
264
+#define rrddim_flag_clear(rd, flag) atomic_flags_clear(&((rd)->flags), flag)
265
+#define rrddim_flag_set_and_clear(rd, set, clear) atomic_flags_set_and_clear(&((rd)->flags), set, clear)
266
267
// ----------------------------------------------------------------------------
268
// engine-specific iterator state for dimension data collection
@@ -490,7 +493,7 @@ static inline time_t storage_engine_global_first_time_s(STORAGE_ENGINE_BACKEND s
493
return rrdeng_global_first_time_s(si);
494
#endif
495
493
- return now_realtime_sec() - (time_t)(default_rrd_history_entries * default_rrd_update_every);
496
+ return now_realtime_sec() - (time_t)(default_rrd_history_entries * nd_profile.update_every);
497
}
498
499
size_t rrdeng_currently_collected_metrics(STORAGE_INSTANCE *si);
@@ -679,10 +682,11 @@ STORAGE_ENGINE* storage_engine_find(const char* name);
682
// ----------------------------------------------------------------------------
683
// RRDSET - this is a chart
684
682
-#define rrdset_flag_get(st) __atomic_load_n(&((st)->flags), __ATOMIC_ACQUIRE)
683
-#define rrdset_flag_check(st, flag) (__atomic_load_n(&((st)->flags), __ATOMIC_ACQUIRE) & (flag))
684
-#define rrdset_flag_set(st, flag) __atomic_or_fetch(&((st)->flags), flag, __ATOMIC_RELEASE)
685
-#define rrdset_flag_clear(st, flag) __atomic_and_fetch(&((st)->flags), ~(flag), __ATOMIC_RELEASE)
685
+#define rrdset_flag_get(st) atomic_flags_get(&((st)->flags))
686
+#define rrdset_flag_check(st, flag) atomic_flags_check(&((st)->flags), flag)
687
+#define rrdset_flag_set(st, flag) atomic_flags_set(&((st)->flags), flag)
688
+#define rrdset_flag_clear(st, flag) atomic_flags_clear(&((st)->flags), flag)
689
+#define rrdset_flag_set_and_clear(st, set, clear) atomic_flags_set_and_clear(&((st)->flags), set, clear)
690
691
#define rrdset_is_replicating(st) (rrdset_flag_check(st, RRDSET_FLAG_SENDER_REPLICATION_IN_PROGRESS|RRDSET_FLAG_RECEIVER_REPLICATION_IN_PROGRESS) \
692
&& !rrdset_flag_check(st, RRDSET_FLAG_SENDER_REPLICATION_FINISHED|RRDSET_FLAG_RECEIVER_REPLICATION_FINISHED))
@@ -772,9 +776,17 @@ struct rrdset {
776
uint32_t sent_version;
777
uint32_t chart_slot;
778
uint32_t dim_last_slot_used;
775
-
779
+#ifdef REPLICATION_TRACKING
780
+ REPLAY_WHO who;
781
+#endif
782
time_t resync_time_s; // the timestamp up to which we should resync clock upstream
783
} snd;
784
+
785
+ struct {
786
+#ifdef REPLICATION_TRACKING
787
+ REPLAY_WHO who;
788
+#endif
789
+ } rcv;
790
} stream;
791
792
// ------------------------------------------------------------------------
@@ -953,30 +965,31 @@ typedef enum __attribute__ ((__packed__)) rrdhost_flags {
965
// Health
966
RRDHOST_FLAG_PENDING_HEALTH_INITIALIZATION = (1 << 17), // contains charts and dims with uninitialized variables
967
RRDHOST_FLAG_INITIALIZED_HEALTH = (1 << 18), // the host has initialized health structures
956
- RRDHOST_FLAG_HEALTH_RUNNING_NOW = (1 << 19), // health is currently executing health check on this host
968
969
// Exporting
959
- RRDHOST_FLAG_EXPORTING_SEND = (1 << 20), // send it to external databases
960
- RRDHOST_FLAG_EXPORTING_DONT_SEND = (1 << 21), // don't send it to external databases
970
+ RRDHOST_FLAG_EXPORTING_SEND = (1 << 19), // send it to external databases
971
+ RRDHOST_FLAG_EXPORTING_DONT_SEND = (1 << 20), // don't send it to external databases
972
973
// ACLK
963
- RRDHOST_FLAG_ACLK_STREAM_CONTEXTS = (1 << 22), // when set, we should send ACLK stream context updates
964
- RRDHOST_FLAG_ACLK_STREAM_ALERTS = (1 << 23), // Host should stream alerts
974
+ RRDHOST_FLAG_ACLK_STREAM_CONTEXTS = (1 << 21), // when set, we should send ACLK stream context updates
975
+ RRDHOST_FLAG_ACLK_STREAM_ALERTS = (1 << 22), // Host should stream alerts
976
977
// Metadata
967
- RRDHOST_FLAG_METADATA_UPDATE = (1 << 24), // metadata needs to be stored in the database
968
- RRDHOST_FLAG_METADATA_LABELS = (1 << 25), // metadata needs to be stored in the database
969
- RRDHOST_FLAG_METADATA_INFO = (1 << 26), // metadata needs to be stored in the database
970
- RRDHOST_FLAG_PENDING_CONTEXT_LOAD = (1 << 27), // Context needs to be loaded
978
+ RRDHOST_FLAG_METADATA_UPDATE = (1 << 23), // metadata needs to be stored in the database
979
+ RRDHOST_FLAG_METADATA_LABELS = (1 << 24), // metadata needs to be stored in the database
980
+ RRDHOST_FLAG_METADATA_INFO = (1 << 25), // metadata needs to be stored in the database
981
+ RRDHOST_FLAG_PENDING_CONTEXT_LOAD = (1 << 26), // Context needs to be loaded
982
972
- RRDHOST_FLAG_METADATA_CLAIMID = (1 << 28), // metadata needs to be stored in the database
983
+ RRDHOST_FLAG_METADATA_CLAIMID = (1 << 27), // metadata needs to be stored in the database
984
974
- RRDHOST_FLAG_GLOBAL_FUNCTIONS_UPDATED = (1 << 29), // set when the host has updated global functions
985
+ RRDHOST_FLAG_GLOBAL_FUNCTIONS_UPDATED = (1 << 28), // set when the host has updated global functions
986
} RRDHOST_FLAGS;
987
977
-#define rrdhost_flag_check(host, flag) (__atomic_load_n(&((host)->flags), __ATOMIC_SEQ_CST) & (flag))
978
-#define rrdhost_flag_set(host, flag) __atomic_or_fetch(&((host)->flags), flag, __ATOMIC_SEQ_CST)
979
-#define rrdhost_flag_clear(host, flag) __atomic_and_fetch(&((host)->flags), ~(flag), __ATOMIC_SEQ_CST)
988
+#define rrdhost_flag_get(host) atomic_flags_get(&((host)->flags))
989
+#define rrdhost_flag_check(host, flag) atomic_flags_check(&((host)->flags), flag)
990
+#define rrdhost_flag_set(host, flag) atomic_flags_set(&((host)->flags), flag)
991
+#define rrdhost_flag_clear(host, flag) atomic_flags_clear(&((host)->flags), flag)
992
+#define rrdhost_flag_set_and_clear(host, set, clear) atomic_flags_set_and_clear(&((host)->flags), set, clear)
993
994
#ifdef NETDATA_INTERNAL_CHECKS
995
#define rrdset_debug(st, fmt, args...) do { if(unlikely(debug_flags & D_RRD_STATS && rrdset_flag_check(st, RRDSET_FLAG_DEBUG))) \
@@ -1030,8 +1043,7 @@ struct rrdhost {
1043
STRING *program_name; // the program name that collects metrics for this host
1044
STRING *program_version; // the program version that collects metrics for this host
1045
1033
- REFCOUNT state_refcount;
1034
- RRDHOST_STATE state_id; // every time data collection (stream receiver) (dis)connects,
1046
+ OBJECT_STATE state_id; // every time data collection (stream receiver) (dis)connects,
1047
// this gets incremented - it is used to detect stale functions,
1048
// stale backfilling requests, etc.
1049
@@ -1088,7 +1100,9 @@ struct rrdhost {
1100
uint32_t connections; // the number of times this sender has connected
1101
1102
struct {
1091
- size_t charts; // the number of charts currently being replicated to a parent
1103
+ uint32_t counter_in; // counts the number of replication statements we have received
1104
+ uint32_t counter_out; // counts the number of replication statements we have sent
1105
+ uint32_t charts; // the number of charts currently being replicated to a parent
1106
} replication;
1107
} status;
1108
@@ -1123,7 +1137,10 @@ struct rrdhost {
1137
STREAM_HANDSHAKE exit_reason; // the last receiver exit reason
1138
1139
struct {
1126
- size_t charts; // the number of charts currently being replicated from a child
1140
+ uint32_t counter_in; // counts the number of replication statements we have received
1141
+ uint32_t counter_out; // counts the number of replication statements we have sent
1142
+ uint32_t backfill_pending; // the number of replication requests pending on us
1143
+ uint32_t charts; // the number of charts currently being replicated from a child
1144
NETDATA_DOUBLE percent; // the % of replication completion
1145
} replication;
1146
} status;
src/database/rrddim.c
+11
-11
@@ -62,7 +62,7 @@ static void rrddim_insert_callback(const DICTIONARY_ITEM *item __maybe_unused, v
62
size_t entries = st->db.entries;
63
if(!entries) entries = 5;
64
65
- rd->db.data = netdata_mmap(NULL, entries * sizeof(storage_number), MAP_PRIVATE, 1, false, NULL);
65
+ rd->db.data = netdata_mmap(NULL, entries * sizeof(storage_number), MAP_PRIVATE, 1, false, true, NULL);
66
if(rd->db.data) {
67
rd->db.memsize = entries * sizeof(storage_number);
68
pulse_db_rrd_memory_add(rd->db.memsize);
@@ -90,7 +90,7 @@ static void rrddim_insert_callback(const DICTIONARY_ITEM *item __maybe_unused, v
90
// initialize the db tiers
91
{
92
size_t initialized = 0;
93
- for(size_t tier = 0; tier < storage_tiers ; tier++) {
93
+ for(size_t tier = 0; tier < nd_profile.storage_tiers; tier++) {
94
STORAGE_ENGINE *eng = host->db[tier].eng;
95
rd->tiers[tier].seb = eng->seb;
96
rd->tiers[tier].tier_grouping = host->db[tier].tier_grouping;
@@ -112,7 +112,7 @@ static void rrddim_insert_callback(const DICTIONARY_ITEM *item __maybe_unused, v
112
// initialize data collection for all tiers
113
{
114
size_t initialized = 0;
115
- for (size_t tier = 0; tier < storage_tiers; tier++) {
115
+ for (size_t tier = 0; tier < nd_profile.storage_tiers; tier++) {
116
if (rd->tiers[tier].smh) {
117
rd->tiers[tier].sch =
118
storage_metric_store_init(rd->tiers[tier].seb, rd->tiers[tier].smh, st->rrdhost->db[tier].tier_grouping * st->update_every, rd->rrdset->smg[tier]);
@@ -174,7 +174,7 @@ bool rrddim_finalize_collection_and_check_retention(RRDDIM *rd) {
174
175
size_t tiers_available = 0, tiers_said_no_retention = 0;
176
177
- for(size_t tier = 0; tier < storage_tiers ;tier++) {
177
+ for(size_t tier = 0; tier < nd_profile.storage_tiers;tier++) {
178
spinlock_lock(&rd->tiers[tier].spinlock);
179
180
if(rd->tiers[tier].sch) {
@@ -212,7 +212,7 @@ static void rrddim_delete_callback(const DICTIONARY_ITEM *item __maybe_unused, v
212
metaqueue_delete_dimension_uuid(&rd->metric_uuid);
213
}
214
215
- for(size_t tier = 0; tier < storage_tiers ;tier++) {
215
+ for(size_t tier = 0; tier < nd_profile.storage_tiers;tier++) {
216
spinlock_lock(&rd->tiers[tier].spinlock);
217
if(rd->tiers[tier].smh) {
218
STORAGE_ENGINE *eng = host->db[tier].eng;
@@ -249,7 +249,7 @@ static bool rrddim_conflict_callback(const DICTIONARY_ITEM *item __maybe_unused,
249
rc += rrddim_set_multiplier(st, rd, ctr->multiplier);
250
rc += rrddim_set_divisor(st, rd, ctr->divisor);
251
252
- for(size_t tier = 0; tier < storage_tiers ;tier++) {
252
+ for(size_t tier = 0; tier < nd_profile.storage_tiers;tier++) {
253
if (!rd->tiers[tier].sch)
254
rd->tiers[tier].sch =
255
storage_metric_store_init(rd->tiers[tier].seb, rd->tiers[tier].smh, st->rrdhost->db[tier].tier_grouping * st->update_every, rd->rrdset->smg[tier]);
@@ -283,7 +283,7 @@ static void rrddim_react_callback(const DICTIONARY_ITEM *item __maybe_unused, vo
283
}
284
285
size_t rrddim_size(void) {
286
- return sizeof(RRDDIM) + storage_tiers * sizeof(struct rrddim_tier);
286
+ return sizeof(RRDDIM) + nd_profile.storage_tiers * sizeof(struct rrddim_tier);
287
}
288
289
void rrddim_index_init(RRDSET *st) {
@@ -406,7 +406,7 @@ inline int rrddim_set_divisor(RRDSET *st, RRDDIM *rd, int32_t divisor) {
406
// ----------------------------------------------------------------------------
407
408
time_t rrddim_last_entry_s_of_tier(RRDDIM *rd, size_t tier) {
409
- if(unlikely(tier > storage_tiers || !rd->tiers[tier].smh))
409
+ if(unlikely(tier > nd_profile.storage_tiers || !rd->tiers[tier].smh))
410
return 0;
411
412
return storage_engine_latest_time_s(rd->tiers[tier].seb, rd->tiers[tier].smh);
@@ -416,7 +416,7 @@ time_t rrddim_last_entry_s_of_tier(RRDDIM *rd, size_t tier) {
416
time_t rrddim_last_entry_s(RRDDIM *rd) {
417
time_t latest_time_s = rrddim_last_entry_s_of_tier(rd, 0);
418
419
- for(size_t tier = 1; tier < storage_tiers ;tier++) {
419
+ for(size_t tier = 1; tier < nd_profile.storage_tiers;tier++) {
420
if(unlikely(!rd->tiers[tier].smh)) continue;
421
422
time_t t = rrddim_last_entry_s_of_tier(rd, tier);
@@ -428,7 +428,7 @@ time_t rrddim_last_entry_s(RRDDIM *rd) {
428
}
429
430
time_t rrddim_first_entry_s_of_tier(RRDDIM *rd, size_t tier) {
431
- if(unlikely(tier > storage_tiers || !rd->tiers[tier].smh))
431
+ if(unlikely(tier > nd_profile.storage_tiers || !rd->tiers[tier].smh))
432
return 0;
433
434
return storage_engine_oldest_time_s(rd->tiers[tier].seb, rd->tiers[tier].smh);
@@ -437,7 +437,7 @@ time_t rrddim_first_entry_s_of_tier(RRDDIM *rd, size_t tier) {
437
time_t rrddim_first_entry_s(RRDDIM *rd) {
438
time_t oldest_time_s = 0;
439
440
- for(size_t tier = 0; tier < storage_tiers ;tier++) {
440
+ for(size_t tier = 0; tier < nd_profile.storage_tiers;tier++) {
441
time_t t = rrddim_first_entry_s_of_tier(rd, tier);
442
if(t != 0 && (oldest_time_s == 0 || t < oldest_time_s))
443
oldest_time_s = t;
src/database/rrdfunctions-internals.h
+1
-1
@@ -29,7 +29,7 @@ struct rrd_host_function {
29
rrd_function_execute_cb_t execute_cb;
30
void *execute_cb_data;
31
32
- RRDHOST_STATE rrdhost_state_id;
32
+ OBJECT_STATE_ID rrdhost_state_id;
33
struct rrd_collector *collector;
34
};
35
src/database/rrdfunctions.c
+6
-6
@@ -20,7 +20,7 @@ static void rrd_functions_insert_callback(const DICTIONARY_ITEM *item __maybe_un
20
21
rrd_collector_started();
22
rdcf->collector = rrd_collector_acquire_current_thread();
23
- rdcf->rrdhost_state_id = rrdhost_state_id(host);
23
+ rdcf->rrdhost_state_id = object_state_id(&host->state_id);
24
25
if(!rdcf->priority)
26
rdcf->priority = RRDFUNCTIONS_PRIORITY_DEFAULT;
@@ -58,14 +58,14 @@ static bool rrd_functions_conflict_callback(const DICTIONARY_ITEM *item __maybe_
58
changed = true;
59
}
60
61
- if(rdcf->rrdhost_state_id != rrdhost_state_id(host)) {
61
+ if(rdcf->rrdhost_state_id != object_state_id(&host->state_id)) {
62
nd_log(NDLS_DAEMON, NDLP_DEBUG,
63
"FUNCTIONS: function '%s' of host '%s' changed state id from %u to %u",
64
dictionary_acquired_item_name(item), rrdhost_hostname(host),
65
rdcf->rrdhost_state_id,
66
- rrdhost_state_id(host));
66
+ object_state_id(&host->state_id));
67
68
- rdcf->rrdhost_state_id = rrdhost_state_id(host);
68
+ rdcf->rrdhost_state_id = object_state_id(&host->state_id);
69
changed = true;
70
}
71
@@ -272,7 +272,7 @@ int rrd_functions_find_by_name(RRDHOST *host, BUFFER *wb, const char *name, size
272
strncpyz(buffer, name, sizeof(buffer) - 1);
273
char *s = NULL;
274
275
- RRDHOST_STATE state_id = rrdhost_state_id(host);
275
+ OBJECT_STATE_ID state_id = object_state_id(&host->state_id);
276
277
bool found = false;
278
*item = NULL;
@@ -341,7 +341,7 @@ bool rrd_function_available(RRDHOST *host, const char *function) {
341
const DICTIONARY_ITEM *item = dictionary_get_and_acquire_item(host->functions, function);
342
if(item) {
343
struct rrd_host_function *rdcf = dictionary_acquired_item_value(item);
344
- if(rrd_collector_running(rdcf->collector) && rdcf->rrdhost_state_id == rrdhost_state_id(host))
344
+ if(rrd_collector_running(rdcf->collector) && rdcf->rrdhost_state_id == object_state_id(&host->state_id))
345
ret = true;
346
347
dictionary_acquired_item_release(host->functions, item);
src/database/rrdhost-state-id.c
deleted
-73
@@ -1,73 +0,0 @@
1
-// SPDX-License-Identifier: GPL-3.0-or-later
2
-
3
-#include "rrdhost-state-id.h"
4
-#include "rrd.h"
5
-
6
-RRDHOST_STATE rrdhost_state_id(struct rrdhost *host) {
7
- return __atomic_load_n(&host->state_id, __ATOMIC_RELAXED);
8
-}
9
-
10
-bool rrdhost_state_connected(RRDHOST *host) {
11
- __atomic_add_fetch(&host->state_id, 1, __ATOMIC_RELAXED);
12
-
13
- int32_t expected = __atomic_load_n(&host->state_refcount, __ATOMIC_RELAXED);
14
- int32_t desired;
15
-
16
- do {
17
- if(expected >= 0) {
18
- internal_fatal(true, "Cannot get the node connected");
19
- return false;
20
- }
21
-
22
- desired = 0;
23
-
24
- } while(!__atomic_compare_exchange_n(
25
- &host->state_refcount, &expected, desired, false, __ATOMIC_RELAXED, __ATOMIC_RELAXED));
26
-
27
- return true;
28
-}
29
-
30
-bool rrdhost_state_disconnected(RRDHOST *host) {
31
- __atomic_add_fetch(&host->state_id, 1, __ATOMIC_RELAXED);
32
-
33
- int32_t expected = __atomic_load_n(&host->state_refcount, __ATOMIC_RELAXED);
34
- int32_t desired;
35
-
36
- do {
37
- if(expected < 0) {
38
- internal_fatal(true, "Cannot get the node disconnected");
39
- return false;
40
- }
41
-
42
- desired = -1;
43
-
44
- } while(!__atomic_compare_exchange_n(
45
- &host->state_refcount, &expected, desired, false, __ATOMIC_RELAXED, __ATOMIC_RELAXED));
46
-
47
- return true;
48
-}
49
-
50
-bool rrdhost_state_acquire(RRDHOST *host, RRDHOST_STATE wanted_state_id) {
51
- int32_t expected = __atomic_load_n(&host->state_refcount, __ATOMIC_RELAXED);
52
- int32_t desired;
53
-
54
- do {
55
- if(expected < 0)
56
- return false;
57
-
58
- desired = expected + 1;
59
-
60
- } while(!__atomic_compare_exchange_n(
61
- &host->state_refcount, &expected, desired, false, __ATOMIC_RELAXED, __ATOMIC_RELAXED));
62
-
63
- if(rrdhost_state_id(host) != wanted_state_id) {
64
- rrdhost_state_release(host);
65
- return false;
66
- }
67
-
68
- return true;
69
-}
70
-
71
-void rrdhost_state_release(RRDHOST *host) {
72
- __atomic_sub_fetch(&host->state_refcount, 1, __ATOMIC_RELAXED);
73
-}
src/database/rrdhost-state-id.h
deleted
-19
@@ -1,19 +0,0 @@
1
-// SPDX-License-Identifier: GPL-3.0-or-later
2
-
3
-#ifndef NETDATA_RRDHOST_STATE_ID_H
4
-#define NETDATA_RRDHOST_STATE_ID_H
5
-
6
-#include "libnetdata/libnetdata.h"
7
-
8
-typedef uint32_t RRDHOST_STATE;
9
-
10
-struct rrdhost;
11
-RRDHOST_STATE rrdhost_state_id(struct rrdhost *host);
12
-
13
-bool rrdhost_state_connected(struct rrdhost *host);
14
-bool rrdhost_state_disconnected(struct rrdhost *host);
15
-
16
-bool rrdhost_state_acquire(struct rrdhost *host, RRDHOST_STATE wanted_state_id);
17
-void rrdhost_state_release(struct rrdhost *host);
18
-
19
-#endif //NETDATA_RRDHOST_STATE_ID_H
src/database/rrdhost-system-info.c
+1
@@ -4,6 +4,7 @@
4
#include "rrdhost-system-info.h"
5
#include "daemon/common.h"
6
#include "aclk/schema-wrappers/node_info.h"
7
+#include "daemon/win_system-info.h"
8
9
// coverity[ +tainted_string_sanitize_content : arg-0 ]
10
static inline void coverity_remove_taint(char *s __maybe_unused) { }
src/database/rrdhost.c
+18
-17
@@ -327,7 +327,7 @@ static RRDHOST *rrdhost_create(
327
}
328
329
RRDHOST *host = callocz(1, sizeof(RRDHOST));
330
- host->state_refcount = -1;
330
+ host->state_id = OBJECT_STATE_INIT_DEACTIVATED;
331
332
__atomic_add_fetch(&netdata_buffers_statistics.rrdhost_allocations_size, sizeof(RRDHOST), __ATOMIC_RELAXED);
333
@@ -394,7 +394,7 @@ static RRDHOST *rrdhost_create(
394
return NULL;
395
}
396
else {
397
- for(size_t tier = 0; tier < storage_tiers ; tier++) {
397
+ for(size_t tier = 0; tier < nd_profile.storage_tiers; tier++) {
398
host->db[tier].mode = RRD_MEMORY_MODE_DBENGINE;
399
host->db[tier].eng = storage_engine_get(host->db[tier].mode);
400
host->db[tier].si = (STORAGE_INSTANCE *)multidb_ctx[tier];
@@ -413,7 +413,7 @@ static RRDHOST *rrdhost_create(
413
414
#ifdef ENABLE_DBENGINE
415
// the first tier is reserved for the non-dbengine modes
416
- for(size_t tier = 1; tier < storage_tiers ; tier++) {
416
+ for(size_t tier = 1; tier < nd_profile.storage_tiers; tier++) {
417
host->db[tier].mode = RRD_MEMORY_MODE_DBENGINE;
418
host->db[tier].eng = storage_engine_get(host->db[tier].mode);
419
host->db[tier].si = (STORAGE_INSTANCE *) multidb_ctx[tier];
@@ -750,13 +750,14 @@ RRDHOST *rrdhost_find_or_create(
750
751
bool rrdhost_should_be_removed(RRDHOST *host, RRDHOST *protected_host, time_t now_s) {
752
if(host != protected_host
753
- && host != localhost
754
- && rrdhost_receiver_replicating_charts(host) == 0
755
- && rrdhost_sender_replicating_charts(host) == 0
756
- && rrdhost_flag_check(host, RRDHOST_FLAG_ORPHAN)
757
- && !rrdhost_flag_check(host, RRDHOST_FLAG_PENDING_CONTEXT_LOAD | RRDHOST_FLAG_HEALTH_RUNNING_NOW | RRDHOST_FLAG_COLLECTOR_ONLINE)
758
- && host->stream.rcv.status.last_disconnected
759
- && host->stream.rcv.status.last_disconnected + rrdhost_free_orphan_time_s < now_s)
753
+ && host != localhost
754
+ && rrdhost_receiver_replicating_charts(host) == 0
755
+ && rrdhost_sender_replicating_charts(host) == 0
756
+ && rrdhost_flag_check(host, RRDHOST_FLAG_ORPHAN)
757
+ && !rrdhost_flag_check(host, RRDHOST_FLAG_PENDING_CONTEXT_LOAD | RRDHOST_FLAG_COLLECTOR_ONLINE)
758
+ && health_evloop_current_iteration() - rrdhost_health_evloop_last_iteration(host) > 10
759
+ && host->stream.rcv.status.last_disconnected
760
+ && host->stream.rcv.status.last_disconnected + rrdhost_free_orphan_time_s < now_s)
761
return true;
762
763
return false;
@@ -801,15 +802,15 @@ int rrd_init(const char *hostname, struct rrdhost_system_info *system_info, bool
802
netdata_conf_dbengine_init(hostname);
803
}
804
else
804
- storage_tiers = 1;
805
+ nd_profile.storage_tiers = 1;
806
807
if (!dbengine_enabled) {
807
- if (storage_tiers > 1) {
808
+ if (nd_profile.storage_tiers > 1) {
809
nd_log(NDLS_DAEMON, NDLP_WARNING,
810
"dbengine is not enabled, but %zu tiers have been requested. Resetting tiers to 1",
810
- storage_tiers);
811
+ nd_profile.storage_tiers);
812
812
- storage_tiers = 1;
813
+ nd_profile.storage_tiers = 1;
814
}
815
816
if (default_rrd_memory_mode == RRD_MEMORY_MODE_DBENGINE) {
@@ -835,8 +836,8 @@ int rrd_init(const char *hostname, struct rrdhost_system_info *system_info, bool
836
, netdata_configured_utc_offset
837
, program_name
838
, NETDATA_VERSION
838
- , default_rrd_update_every
839
- , default_rrd_history_entries
839
+ ,
840
+ nd_profile.update_every, default_rrd_history_entries
841
, default_rrd_memory_mode
842
, health_plugin_enabled()
843
, stream_send.enabled
@@ -855,7 +856,7 @@ int rrd_init(const char *hostname, struct rrdhost_system_info *system_info, bool
856
return 1;
857
858
rrdhost_flag_set(localhost, RRDHOST_FLAG_COLLECTOR_ONLINE);
858
- rrdhost_state_connected(localhost);
859
+ object_state_activate(&localhost->state_id);
860
861
ml_host_start(localhost);
862
dyncfg_host_init(localhost);
src/database/rrdlabels.c
+2
-2
@@ -70,7 +70,7 @@ static inline void STATS_PLUS_MEMORY(struct dictionary_stats *stats, int64_t jud
70
__atomic_fetch_add(&stats->memory.index, judy_mem, __ATOMIC_RELAXED);
71
72
if(item_size)
73
- __atomic_fetch_add(&stats->memory.dict, (long)item_size, __ATOMIC_RELAXED);
73
+ __atomic_fetch_add(&stats->memory.dict, (ssize_t)item_size, __ATOMIC_RELAXED);
74
75
if(value_size)
76
__atomic_fetch_add(&stats->memory.values, (long)value_size, __ATOMIC_RELAXED);
@@ -81,7 +81,7 @@ static inline void STATS_MINUS_MEMORY(struct dictionary_stats *stats, int64_t ju
81
__atomic_fetch_add(&stats->memory.index, judy_mem, __ATOMIC_RELAXED);
82
83
if(item_size)
84
- __atomic_fetch_sub(&stats->memory.dict, (long)item_size, __ATOMIC_RELAXED);
84
+ __atomic_fetch_sub(&stats->memory.dict, (ssize_t)item_size, __ATOMIC_RELAXED);
85
86
if(value_size)
87
__atomic_fetch_sub(&stats->memory.values, (long)value_size, __ATOMIC_RELAXED);
src/database/rrdset.c
+6
-6
@@ -271,7 +271,7 @@ static void rrdset_insert_callback(const DICTIONARY_ITEM *item __maybe_unused, v
271
272
// initialize the db tiers
273
{
274
- for(size_t tier = 0; tier < storage_tiers ; tier++) {
274
+ for(size_t tier = 0; tier < nd_profile.storage_tiers; tier++) {
275
STORAGE_ENGINE *eng = st->rrdhost->db[tier].eng;
276
if(!eng) continue;
277
@@ -318,7 +318,7 @@ void rrdset_finalize_collection(RRDSET *st, bool dimensions_too) {
318
rrddim_foreach_done(rd);
319
}
320
321
- for(size_t tier = 0; tier < storage_tiers ; tier++) {
321
+ for(size_t tier = 0; tier < nd_profile.storage_tiers; tier++) {
322
STORAGE_ENGINE *eng = st->rrdhost->db[tier].eng;
323
if(!eng) continue;
324
@@ -672,7 +672,7 @@ time_t rrdset_first_entry_s(RRDSET *st) {
672
}
673
674
time_t rrdset_first_entry_s_of_tier(RRDSET *st, size_t tier) {
675
- if(unlikely(tier > storage_tiers))
675
+ if(unlikely(tier > nd_profile.storage_tiers))
676
return 0;
677
678
RRDDIM *rd;
@@ -843,7 +843,7 @@ void rrdset_reset(RRDSET *st) {
843
rd->collector.counter = 0;
844
845
if(!rrddim_flag_check(rd, RRDDIM_FLAG_ARCHIVED)) {
846
- for(size_t tier = 0; tier < storage_tiers ;tier++)
846
+ for(size_t tier = 0; tier < nd_profile.storage_tiers;tier++)
847
storage_engine_store_flush(rd->tiers[tier].sch);
848
}
849
}
@@ -1273,7 +1273,7 @@ void rrddim_store_metric(RRDDIM *rd, usec_t point_end_time_ut, NETDATA_DOUBLE n,
1273
.flags = flags
1274
};
1275
1276
- for(size_t tier = 1; tier < storage_tiers ;tier++) {
1276
+ for(size_t tier = 1; tier < nd_profile.storage_tiers;tier++) {
1277
if(unlikely(!rd->tiers[tier].smh)) continue;
1278
1279
struct rrddim_tier *t = &rd->tiers[tier];
@@ -2020,7 +2020,7 @@ time_t rrdset_set_update_every_s(RRDSET *st, time_t update_every_s) {
2020
// switch update every to the storage engine
2021
RRDDIM *rd;
2022
rrddim_foreach_read(rd, st) {
2023
- for (size_t tier = 0; tier < storage_tiers; tier++) {
2023
+ for (size_t tier = 0; tier < nd_profile.storage_tiers; tier++) {
2024
if (rd->tiers[tier].sch)
2025
storage_engine_store_change_collection_frequency(
2026
rd->tiers[tier].sch,
src/database/sqlite/sqlite_aclk_node.c
+62
-16
@@ -105,13 +105,19 @@ void aclk_check_node_info_and_collectors(void)
105
return;
106
107
size_t context_loading = 0;
108
- size_t replicating = 0;
108
+ size_t replicating_rcv = 0;
109
+ size_t replicating_snd = 0;
110
size_t context_pp = 0;
111
112
STRING *context_loading_host = NULL;
112
- STRING *replicating_host = NULL;
113
+ STRING *replicating_rcv_host = NULL;
114
+ STRING *replicating_snd_host = NULL;
115
STRING *context_pp_host = NULL;
114
-
116
+
117
+#ifdef REPLICATION_TRACKING
118
+ struct replay_who_counters replay_counters = { 0 };
119
+#endif
120
+
121
time_t now = now_realtime_sec();
122
dfe_start_reentrant(rrdhost_root_index, host)
123
{
@@ -129,13 +135,29 @@ void aclk_check_node_info_and_collectors(void)
135
if (!wc->node_info_send_time && !wc->node_collectors_send)
136
continue;
137
138
+ bool replicating = false;
139
+
140
if (unlikely(rrdhost_receiver_replicating_charts(host))) {
133
- internal_error(true, "ACLK SYNC: Host %s is still replicating", rrdhost_hostname(host));
134
- replicating++;
135
- replicating_host = host->hostname;
136
- continue;
141
+ internal_error(true, "ACLK SYNC: Host %s is still replicating in", rrdhost_hostname(host));
142
+ replicating_rcv++;
143
+ replicating_rcv_host = host->hostname;
144
+ replicating = true;
145
+ }
146
+
147
+ if (unlikely(rrdhost_sender_replicating_charts(host))) {
148
+ internal_error(true, "ACLK SYNC: Host %s is still replicating out", rrdhost_hostname(host));
149
+ replicating_snd++;
150
+ replicating_snd_host = host->hostname;
151
+ replicating = true;
152
}
153
154
+#ifdef REPLICATION_TRACKING
155
+ replication_tracking_counters(host, &replay_counters);
156
+#endif
157
+
158
+ if(replicating)
159
+ continue;
160
+
161
bool pp_queue_empty = !(host->rrdctx.pp_queue && dictionary_entries(host->rrdctx.pp_queue));
162
163
if (!pp_queue_empty && (wc->node_info_send_time || wc->node_collectors_send)) {
@@ -158,31 +180,55 @@ void aclk_check_node_info_and_collectors(void)
180
}
181
dfe_done(host);
182
161
- if (context_loading || replicating || context_pp) {
183
+ if (context_loading || replicating_rcv || replicating_snd || context_pp) {
184
+#ifdef REPLICATION_TRACKING
185
+ char replay_counters_txt[1024];
186
+ snprintfz(replay_counters_txt, sizeof(replay_counters_txt),
187
+ " - REPLAY WHO RCV { %zu unknown, %zu me, %zu them, %zu finished } - "
188
+ "REPLAY WHO SND { %zu unknown, %zu me, %zu them, %zu finished }",
189
+ replay_counters.rcv[REPLAY_WHO_UNKNOWN], replay_counters.rcv[REPLAY_WHO_ME], replay_counters.rcv[REPLAY_WHO_THEM], replay_counters.rcv[REPLAY_WHO_FINISHED],
190
+ replay_counters.snd[REPLAY_WHO_UNKNOWN], replay_counters.snd[REPLAY_WHO_ME], replay_counters.snd[REPLAY_WHO_THEM], replay_counters.snd[REPLAY_WHO_FINISHED]
191
+ );
192
+#else
193
+ char *replay_counters_txt = "";
194
+#endif
195
+
196
const char *context_loading_pre = "", *context_loading_body = "", *context_loading_post = "";
197
if(context_loading == 1) {
198
context_loading_pre = " (host '";
199
context_loading_body = string2str(context_loading_host);
200
context_loading_post = "')";
201
}
168
- const char *replicating_pre = "", *replicating_body = "", *replicating_post = "";
169
- if(replicating == 1) {
170
- replicating_pre = " (host '";
171
- replicating_body = string2str(replicating_host);
172
- replicating_post = "')";
202
+
203
+ const char *replicating_rcv_pre = "", *replicating_rcv_body = "", *replicating_rcv_post = "";
204
+ if(replicating_rcv == 1) {
205
+ replicating_rcv_pre = " (host '";
206
+ replicating_rcv_body = string2str(replicating_rcv_host);
207
+ replicating_rcv_post = "')";
208
+ }
209
+
210
+ const char *replicating_snd_pre = "", *replicating_snd_body = "", *replicating_snd_post = "";
211
+ if(replicating_snd == 1) {
212
+ replicating_snd_pre = " (host '";
213
+ replicating_snd_body = string2str(replicating_snd_host);
214
+ replicating_snd_post = "')";
215
}
216
+
217
const char *context_pp_pre = "", *context_pp_body = "", *context_pp_post = "";
218
if(context_pp == 1) {
219
context_pp_pre = " (host '";
220
context_pp_body = string2str(context_pp_host);
221
context_pp_post = "')";
222
}
223
+
224
nd_log_limit_static_thread_var(erl, 10, 100 * USEC_PER_MS);
225
nd_log_limit(&erl, NDLS_DAEMON, NDLP_INFO,
182
- "%zu nodes loading contexts%s%s%s, %zu replicating data%s%s%s, %zu pending context post processing%s%s%s",
226
+ "NODES INFO: %zu nodes loading contexts%s%s%s, %zu receiving replication%s%s%s, %zu sending replication%s%s%s, %zu pending context post processing%s%s%s%s",
227
context_loading, context_loading_pre, context_loading_body, context_loading_post,
184
- replicating, replicating_pre, replicating_body, replicating_post,
185
- context_pp, context_pp_pre, context_pp_body, context_pp_post
228
+ replicating_rcv, replicating_rcv_pre, replicating_rcv_body, replicating_rcv_post,
229
+ replicating_snd, replicating_snd_pre, replicating_snd_body, replicating_snd_post,
230
+ context_pp, context_pp_pre, context_pp_body, context_pp_post,
231
+ replay_counters_txt
232
);
233
}
234
}
src/database/sqlite/sqlite_metadata.c
+1
-1
@@ -1093,7 +1093,7 @@ static bool dimension_can_be_deleted(nd_uuid_t *dim_uuid __maybe_unused, sqlite3
1093
#ifdef ENABLE_DBENGINE
1094
if(dbengine_enabled) {
1095
bool no_retention = true;
1096
- for (size_t tier = 0; tier < storage_tiers; tier++) {
1096
+ for (size_t tier = 0; tier < nd_profile.storage_tiers; tier++) {
1097
if (!multidb_ctx[tier])
1098
continue;
1099
time_t first_time_t = 0, last_time_t = 0;
src/exporting/send_data.c
+1
-1
@@ -289,7 +289,7 @@ void simple_connector_worker(void *instance_p)
289
290
if (exporting_tls_is_enabled(instance->config.type, options) && sock != -1) {
291
if (netdata_ssl_exporting_ctx) {
292
- if (sock_delnonblock(sock) < 0)
292
+ if (sock_setnonblock(sock, false) != 0)
293
netdata_log_error("Exporting cannot remove the non-blocking flag from socket %d", sock);
294
295
if(netdata_ssl_open(&connector_specific_data->ssl, netdata_ssl_exporting_ctx, sock)) {
src/health/health-alert-log.h
+1
@@ -22,6 +22,7 @@ typedef struct health {
22
bool enabled; // 1 when this host has health enabled
23
bool use_summary_for_notifications; // whether to use the summary field as a subject for notifications
24
int32_t pending_transitions; // pending alert transitions to store
25
+ uint64_t evloop_iteration; // the last health iteration that evaluated this host
26
} HEALTH;
27
28
#endif //NETDATA_HEALTH_ALERT_LOG_H
src/health/health.h
+4
@@ -99,4 +99,8 @@ int alert_variable_lookup_trace(RRDHOST *host, RRDSET *st, const char *variable,
99
typedef void (*prototype_metadata_cb_t)(void *data, STRING *type, STRING *component, STRING *classification, STRING *recipient);
100
void health_prototype_metadata_foreach(void *data, prototype_metadata_cb_t cb);
101
102
+uint64_t health_evloop_current_iteration(void);
103
+void rrdhost_set_health_evloop_iteration(RRDHOST *host);
104
+uint64_t rrdhost_health_evloop_last_iteration(RRDHOST *host);
105
+
106
#endif //NETDATA_HEALTH_H
src/health/health_event_loop.c
+24
-9
@@ -19,6 +19,21 @@
19
#error WORKER_UTILIZATION_MAX_JOB_TYPES has to be at least 10
20
#endif
21
22
+static uint64_t health_evloop_iteration = 0;
23
+
24
+uint64_t health_evloop_current_iteration(void) {
25
+ return __atomic_load_n(&health_evloop_iteration, __ATOMIC_RELAXED);
26
+}
27
+
28
+uint64_t rrdhost_health_evloop_last_iteration(RRDHOST *host) {
29
+ return __atomic_load_n(&host->health.evloop_iteration, __ATOMIC_RELAXED);
30
+}
31
+
32
+void rrdhost_set_health_evloop_iteration(RRDHOST *host) {
33
+ __atomic_store_n(&host->health.evloop_iteration,
34
+ health_evloop_current_iteration(), __ATOMIC_RELAXED);
35
+}
36
+
37
// ----------------------------------------------------------------------------
38
// health main thread and friends
39
@@ -90,18 +105,20 @@ static inline int rrdcalc_isrunnable(RRDCALC *rc, time_t now, time_t *next_run)
105
return 1;
106
}
107
93
-static void health_sleep(time_t next_run, unsigned int loop __maybe_unused) {
108
+static void health_sleep(time_t next_run, uint64_t loop) {
109
time_t now = now_realtime_sec();
110
if(now < next_run) {
111
worker_is_idle();
97
- netdata_log_debug(D_HEALTH, "Health monitoring iteration no %u done. Next iteration in %d secs", loop, (int) (next_run - now));
112
+ netdata_log_debug(D_HEALTH, "Health monitoring iteration no %llu done. Next iteration in %d secs",
113
+ (unsigned long long)loop, (int) (next_run - now));
114
while (now < next_run && service_running(SERVICE_HEALTH)) {
115
sleep_usec(USEC_PER_SEC);
116
now = now_realtime_sec();
117
}
118
}
119
else {
104
- netdata_log_debug(D_HEALTH, "Health monitoring iteration no %u done. Next iteration now", loop);
120
+ netdata_log_debug(D_HEALTH, "Health monitoring iteration no %llu done. Next iteration now",
121
+ (unsigned long long)loop);
122
}
123
}
124
@@ -215,6 +232,8 @@ static void health_event_loop_for_host(RRDHOST *host, bool apply_hibernation_del
232
if(unlikely(!rrdhost_should_run_health(host)))
233
return;
234
235
+ rrdhost_set_health_evloop_iteration(host);
236
+
237
//#define rrdhost_pending_alert_transitions(host) (__atomic_load_n(&((host)->aclk_config.alert_transition.pending), __ATOMIC_RELAXED))
238
239
if (unlikely(__atomic_load_n(&host->health.pending_transitions, __ATOMIC_RELAXED))) {
@@ -625,7 +644,6 @@ static void health_event_loop_for_host(RRDHOST *host, bool apply_hibernation_del
644
}
645
646
static void health_event_loop(void) {
628
- unsigned int loop = 0;
647
648
while(service_running(SERVICE_HEALTH)) {
649
if(!stream_control_health_should_be_running()) {
@@ -634,9 +652,6 @@ static void health_event_loop(void) {
652
continue;
653
}
654
637
- loop++;
638
- netdata_log_debug(D_HEALTH, "Health monitoring iteration no %u started", loop);
639
-
655
time_t now = now_realtime_sec();
656
bool apply_hibernation_delay = false;
657
time_t next_run = now + health_globals.config.run_at_least_every_seconds;
@@ -661,14 +676,14 @@ static void health_event_loop(void) {
676
}
677
678
worker_is_busy(WORKER_HEALTH_JOB_RRD_LOCK);
679
+ uint64_t loop = __atomic_add_fetch(&health_evloop_iteration, 1, __ATOMIC_RELAXED);
680
+
681
RRDHOST *host;
682
dfe_start_reentrant(rrdhost_root_index, host) {
683
if(unlikely(!service_running(SERVICE_HEALTH)))
684
break;
685
669
- rrdhost_flag_set(host, RRDHOST_FLAG_HEALTH_RUNNING_NOW);
686
health_event_loop_for_host(host, apply_hibernation_delay, now, &next_run);
671
- rrdhost_flag_clear(host, RRDHOST_FLAG_HEALTH_RUNNING_NOW);
687
}
688
dfe_done(host);
689
src/health/health_log.c
+11
-5
@@ -66,7 +66,10 @@ void health_log_alert_transition_with_trace(RRDHOST *host, ALARM_ENTRY *ae, int
66
break;
67
68
case RRDCALC_STATUS_CLEAR:
69
- priority = NDLP_INFO;
69
+ if(ae->old_status == RRDCALC_STATUS_UNINITIALIZED)
70
+ priority = NDLP_DEBUG;
71
+ else
72
+ priority = NDLP_INFO;
73
break;
74
75
case RRDCALC_STATUS_WARNING:
@@ -81,10 +84,13 @@ void health_log_alert_transition_with_trace(RRDHOST *host, ALARM_ENTRY *ae, int
84
}
85
86
netdata_logger(NDLS_HEALTH, priority, file, function, line,
84
- "ALERT '%s' of instance '%s' on node '%s', transitioned from %s to %s",
85
- string2str(ae->name), string2str(ae->chart), string2str(host->hostname),
86
- rrdcalc_status2string(ae->old_status), rrdcalc_status2string(ae->new_status)
87
- );
87
+ "ALERT '%s' of instance '%s' on node '%s', transitioned from %s to %s.\n"
88
+ "%s value got from %f %s, to %f %s.",
89
+ string2str(ae->name), string2str(ae->chart), string2str(host->hostname),
90
+ rrdcalc_status2string(ae->old_status), rrdcalc_status2string(ae->new_status),
91
+ string2str(ae->name),
92
+ ae->old_value, string2str(ae->units),
93
+ ae->new_value, string2str(ae->units));
94
}
95
96
// ----------------------------------------------------------------------------
src/libnetdata/aral/aral.c
+92
-56
@@ -56,27 +56,29 @@ typedef struct aral_page {
56
struct {
57
SPINLOCK spinlock;
58
ARAL_FREE *list;
59
- } free;
59
+ } available;
60
+
61
+ struct {
62
+ SPINLOCK spinlock;
63
+ ARAL_FREE *list;
64
+ } incoming;
65
66
} ARAL_PAGE;
67
68
typedef enum {
69
ARAL_LOCKLESS = (1 << 0),
70
ARAL_ALLOCATED_STATS = (1 << 1),
71
+ ARAL_DONT_DUMP = (1 << 2),
72
} ARAL_OPTIONS;
73
74
struct aral_ops {
75
struct {
70
- CACHE_LINE_PADDING();
71
- size_t allocators; // the number of threads currently trying to allocate memory
72
- CACHE_LINE_PADDING();
73
- size_t deallocators; // the number of threads currently trying to deallocate memory
74
- CACHE_LINE_PADDING();
75
- bool last_allocated_or_deallocated; // stability detector, true when was last allocated
76
+ PAD64(size_t) allocators; // the number of threads currently trying to allocate memory
77
+ PAD64(size_t) deallocators; // the number of threads currently trying to deallocate memory
78
+ PAD64(bool) last_allocated_or_deallocated; // stability detector, true when was last allocated
79
} atomic;
80
81
struct {
79
- CACHE_LINE_PADDING();
82
SPINLOCK spinlock;
83
size_t allocating_elements; // currently allocating elements
84
size_t allocation_size; // current / next allocation size
@@ -201,14 +203,24 @@ static inline void aral_unlock_with_trace(ARAL *ar, const char *func) {
203
#define aral_lock(ar) aral_lock_with_trace(ar, __FUNCTION__)
204
#define aral_unlock(ar) aral_unlock_with_trace(ar, __FUNCTION__)
205
204
-static inline void aral_page_free_lock(ARAL *ar, ARAL_PAGE *page) {
206
+static inline void aral_page_available_lock(ARAL *ar, ARAL_PAGE *page) {
207
+ if(likely(!(ar->config.options & ARAL_LOCKLESS)))
208
+ spinlock_lock(&page->available.spinlock);
209
+}
210
+
211
+static inline void aral_page_available_unlock(ARAL *ar, ARAL_PAGE *page) {
212
if(likely(!(ar->config.options & ARAL_LOCKLESS)))
206
- spinlock_lock(&page->free.spinlock);
213
+ spinlock_unlock(&page->available.spinlock);
214
}
215
209
-static inline void aral_page_free_unlock(ARAL *ar, ARAL_PAGE *page) {
216
+static inline void aral_page_incoming_lock(ARAL *ar, ARAL_PAGE *page) {
217
if(likely(!(ar->config.options & ARAL_LOCKLESS)))
211
- spinlock_unlock(&page->free.spinlock);
218
+ spinlock_lock(&page->incoming.spinlock);
219
+}
220
+
221
+static inline void aral_page_incoming_unlock(ARAL *ar, ARAL_PAGE *page) {
222
+ if(likely(!(ar->config.options & ARAL_LOCKLESS)))
223
+ spinlock_unlock(&page->incoming.spinlock);
224
}
225
226
static inline bool aral_adders_trylock(ARAL *ar, bool marked) {
@@ -490,7 +502,7 @@ static ARAL_PAGE *aral_create_page___no_lock_needed(ARAL *ar, size_t size TRACE_
502
page->filename = strdupz(filename);
503
page->mapped = true;
504
493
- page->data = netdata_mmap(page->filename, size, MAP_SHARED, 0, false, NULL);
505
+ page->data = netdata_mmap(page->filename, size, MAP_SHARED, 0, false, ar->config.options & ARAL_DONT_DUMP, NULL);
506
if (unlikely(!page->data))
507
fatal("ARAL: '%s' cannot allocate aral buffer of size %zu on filename '%s'",
508
ar->config.name, size, page->filename);
@@ -512,7 +524,7 @@ static ARAL_PAGE *aral_create_page___no_lock_needed(ARAL *ar, size_t size TRACE_
524
525
if (size >= ARAL_MMAP_PAGES_ABOVE) {
526
bool mapped;
515
- uint8_t *ptr = netdata_mmap(NULL, size, MAP_PRIVATE, 1, false, NULL);
527
+ uint8_t *ptr = netdata_mmap(NULL, size, MAP_PRIVATE, 1, false, ar->config.options & ARAL_DONT_DUMP, NULL);
528
if (ptr) {
529
mapped = true;
530
stats = &ar->stats->mmap;
@@ -539,7 +551,8 @@ static ARAL_PAGE *aral_create_page___no_lock_needed(ARAL *ar, size_t size TRACE_
551
}
552
#endif
553
542
- spinlock_init(&page->free.spinlock);
554
+ spinlock_init(&page->available.spinlock);
555
+ spinlock_init(&page->incoming.spinlock);
556
page->size = size;
557
page->max_elements = aral_elements_in_page_size(ar, page->size);
558
page->aral_lock.free_elements = page->max_elements;
@@ -560,7 +573,7 @@ static ARAL_PAGE *aral_create_page___no_lock_needed(ARAL *ar, size_t size TRACE_
573
574
fr->size = page->max_elements * ar->config.element_size;
575
fr->next = NULL;
563
- page->free.list = fr;
576
+ page->available.list = fr;
577
578
aral_free_validate_internal_check(ar, fr);
579
@@ -741,37 +754,31 @@ static inline ARAL_PAGE *aral_get_first_page_with_a_free_slot(ARAL *ar, bool mar
754
return page;
755
}
756
744
-void *aral_callocz_internal(ARAL *ar, bool marked TRACE_ALLOCATIONS_FUNCTION_DEFINITION_PARAMS) {
745
- void *r = aral_mallocz_internal(ar, marked TRACE_ALLOCATIONS_FUNCTION_CALL_PARAMS);
746
- memset(r, 0, ar->config.requested_element_size);
747
- return r;
748
-}
749
-
750
-void *aral_mallocz_internal(ARAL *ar, bool marked TRACE_ALLOCATIONS_FUNCTION_DEFINITION_PARAMS) {
751
-#if defined(FSANITIZE_ADDRESS)
752
- return mallocz(ar->config.requested_element_size);
753
-#endif
754
-
755
- // reserve a slot on a free page
756
- ARAL_PAGE *page = aral_get_first_page_with_a_free_slot(ar, marked TRACE_ALLOCATIONS_FUNCTION_CALL_PARAMS);
757
- // the page returned has reserved a slot for us
757
+static void *aral_get_free_slot___no_lock_required(ARAL *ar, ARAL_PAGE *page, bool marked) {
758
+ aral_page_available_lock(ar, page);
759
759
- aral_page_free_lock(ar, page);
760
+ if(!page->available.list) {
761
+ aral_page_incoming_lock(ar, page);
762
+ page->available.list = page->incoming.list;
763
+ page->incoming.list = NULL;
764
+ aral_page_incoming_unlock(ar, page);
765
+ }
766
761
- internal_fatal(!page->free.list,
762
- "ARAL: '%s' free item to use, cannot be NULL.", ar->config.name);
767
+ ARAL_FREE *found_fr;
768
+ found_fr = page->available.list;
769
764
- internal_fatal(page->free.list->size < ar->config.element_size,
765
- "ARAL: '%s' free item size %zu, cannot be smaller than %zu",
766
- ar->config.name, page->free.list->size, ar->config.element_size);
770
+ internal_fatal(!found_fr,
771
+ "ARAL: '%s' incoming free list, cannot be NULL.", ar->config.name);
772
768
- ARAL_FREE *found_fr = page->free.list;
773
+ internal_fatal(found_fr->size < ar->config.element_size,
774
+ "ARAL: '%s' free element size %zu, cannot be smaller than %zu",
775
+ ar->config.name, page->available.list->size, ar->config.element_size);
776
777
// check if the remaining size (after we use this slot) is not enough for another element
778
if(unlikely(found_fr->size - ar->config.element_size < ar->config.element_size)) {
779
// we can use the entire free space entry
780
774
- page->free.list = found_fr->next;
781
+ page->available.list = found_fr->next;
782
}
783
else {
784
// we can split the free space entry
@@ -783,12 +790,12 @@ void *aral_mallocz_internal(ARAL *ar, bool marked TRACE_ALLOCATIONS_FUNCTION_DEF
790
791
// link the free slot first in the page
792
fr->next = found_fr->next;
786
- page->free.list = fr;
793
+ page->available.list = fr;
794
795
aral_free_validate_internal_check(ar, fr);
796
}
797
791
- aral_page_free_unlock(ar, page);
798
+ aral_page_available_unlock(ar, page);
799
800
// put the page pointer after the element
801
aral_set_page_pointer_after_element___do_NOT_have_aral_lock(ar, page, found_fr, marked);
@@ -798,9 +805,44 @@ void *aral_mallocz_internal(ARAL *ar, bool marked TRACE_ALLOCATIONS_FUNCTION_DEF
805
else
806
__atomic_add_fetch(&ar->stats->malloc.used_bytes, ar->config.requested_element_size, __ATOMIC_RELAXED);
807
801
- internal_fatal((uintptr_t)found_fr % SYSTEM_REQUIRED_ALIGNMENT != 0, "Pointer is not aligned properly");
808
+ return found_fr;
809
+}
810
+
811
+static inline void aral_add_free_slot___no_lock_required(ARAL *ar, ARAL_PAGE *page, void *ptr) {
812
+ ARAL_FREE *fr = (ARAL_FREE *)ptr;
813
+ fr->size = ar->config.element_size;
814
+
815
+ aral_page_incoming_lock(ar, page);
816
+ fr->next = page->incoming.list;
817
+ page->incoming.list = fr;
818
+ aral_page_incoming_unlock(ar, page);
819
+
820
+ if(unlikely(ar->config.mmap.enabled || page->mapped))
821
+ __atomic_sub_fetch(&ar->stats->mmap.used_bytes, ar->config.requested_element_size, __ATOMIC_RELAXED);
822
+ else
823
+ __atomic_sub_fetch(&ar->stats->malloc.used_bytes, ar->config.requested_element_size, __ATOMIC_RELAXED);
824
+}
825
+
826
+void *aral_callocz_internal(ARAL *ar, bool marked TRACE_ALLOCATIONS_FUNCTION_DEFINITION_PARAMS) {
827
+ void *r = aral_mallocz_internal(ar, marked TRACE_ALLOCATIONS_FUNCTION_CALL_PARAMS);
828
+ memset(r, 0, ar->config.requested_element_size);
829
+ return r;
830
+}
831
+
832
+void *aral_mallocz_internal(ARAL *ar, bool marked TRACE_ALLOCATIONS_FUNCTION_DEFINITION_PARAMS) {
833
+#if defined(FSANITIZE_ADDRESS)
834
+ return mallocz(ar->config.requested_element_size);
835
+#endif
836
+
837
+ // reserve a slot on a free page
838
+ ARAL_PAGE *page = aral_get_first_page_with_a_free_slot(ar, marked TRACE_ALLOCATIONS_FUNCTION_CALL_PARAMS);
839
+ // the page returned has reserved a slot for us
840
+
841
+ void *data = aral_get_free_slot___no_lock_required(ar, page, marked);
842
+
843
+ internal_fatal((uintptr_t)data % SYSTEM_REQUIRED_ALIGNMENT != 0, "Pointer is not aligned properly");
844
803
- return (void *)found_fr;
845
+ return data;
846
}
847
848
// returns true if it moved the page to the unmarked list
@@ -869,13 +911,7 @@ void aral_freez_internal(ARAL *ar, void *ptr TRACE_ALLOCATIONS_FUNCTION_DEFINITI
911
__atomic_sub_fetch(&ar->stats->malloc.used_bytes, ar->config.requested_element_size, __ATOMIC_RELAXED);
912
913
// make this element available
872
- ARAL_FREE *fr = (ARAL_FREE *)ptr;
873
- fr->size = ar->config.element_size;
874
-
875
- aral_page_free_lock(ar, page);
876
- fr->next = page->free.list;
877
- page->free.list = fr;
878
- aral_page_free_unlock(ar, page);
914
+ aral_add_free_slot___no_lock_required(ar, page, ptr);
915
916
aral_lock(ar);
917
@@ -989,9 +1025,10 @@ size_t aral_actual_element_size(ARAL *ar) {
1025
}
1026
1027
ARAL *aral_create(const char *name, size_t element_size, size_t initial_page_elements, size_t max_page_size,
992
- struct aral_statistics *stats, const char *filename, const char **cache_dir, bool mmap, bool lockless) {
1028
+ struct aral_statistics *stats, const char *filename, const char **cache_dir,
1029
+ bool mmap, bool lockless, bool dont_dump) {
1030
ARAL *ar = callocz(1, sizeof(ARAL));
994
- ar->config.options = ((lockless) ? ARAL_LOCKLESS : 0);
1031
+ ar->config.options = ((lockless) ? ARAL_LOCKLESS : 0) | ((dont_dump) ? ARAL_DONT_DUMP : 0);
1032
ar->config.requested_element_size = element_size;
1033
ar->config.initial_page_elements = initial_page_elements;
1034
ar->config.requested_max_page_size = max_page_size;
@@ -1148,7 +1185,7 @@ ARAL *aral_by_size_acquire(size_t size) {
1185
aral_by_size_globals.array[size].refcount++;
1186
1187
internal_fatal(
1151
- aral_requested_element_size(ar) != size, "DICTIONARY: aral has size %zu but we want %zu",
1188
+ aral_requested_element_size(ar) != size, "ARAL BY SIZE: aral has size %zu but we want %zu",
1189
aral_requested_element_size(ar), size);
1190
}
1191
@@ -1160,7 +1197,7 @@ ARAL *aral_by_size_acquire(size_t size) {
1197
0,
1198
0,
1199
&aral_by_size_globals.shared_statistics,
1163
- NULL, NULL, false, false);
1200
+ NULL, NULL, false, false, false);
1201
1202
if(size <= ARAL_BY_SIZE_MAX_SIZE) {
1203
aral_by_size_globals.array[size].ar = ar;
@@ -1347,7 +1384,7 @@ int aral_stress_test(size_t threads, size_t elements, size_t seconds) {
1384
16384,
1385
NULL,
1386
"aral-stress-test",
1350
- NULL, false, false),
1387
+ NULL, false, false, false),
1388
.elements = elements,
1389
.errors = 0,
1390
};
@@ -1424,8 +1461,7 @@ int aral_unittest(size_t elements) {
1461
NULL,
1462
"aral-test",
1463
&cache_dir,
1427
- false,
1428
- false),
1464
+ false, false, false),
1465
.elements = elements,
1466
.errors = 0,
1467
};
src/libnetdata/aral/aral.h
+8
-13
@@ -9,22 +9,16 @@
9
typedef struct aral ARAL;
10
11
struct aral_page_type_stats {
12
- CACHE_LINE_PADDING();
13
- size_t allocations;
14
- CACHE_LINE_PADDING();
15
- size_t allocated_bytes;
16
- CACHE_LINE_PADDING();
17
- size_t used_bytes;
18
- CACHE_LINE_PADDING();
19
- size_t padding_bytes;
12
+ PAD64(size_t) allocations;
13
+ PAD64(size_t) allocated_bytes;
14
+ PAD64(size_t) used_bytes;
15
+ PAD64(size_t) padding_bytes;
16
};
17
18
struct aral_statistics {
19
struct {
24
- CACHE_LINE_PADDING();
25
- size_t allocations;
26
- CACHE_LINE_PADDING();
27
- size_t allocated_bytes;
20
+ PAD64(size_t) allocations;
21
+ PAD64(size_t) allocated_bytes;
22
} structures;
23
24
struct aral_page_type_stats malloc;
@@ -36,7 +30,8 @@ struct aral_statistics {
30
const char *aral_name(ARAL *ar);
31
32
ARAL *aral_create(const char *name, size_t element_size, size_t initial_page_elements, size_t max_page_size,
39
- struct aral_statistics *stats, const char *filename, const char **cache_dir, bool mmap, bool lockless);
33
+ struct aral_statistics *stats, const char *filename, const char **cache_dir,
34
+ bool mmap, bool lockless, bool dont_dump);
35
36
// --------------------------------------------------------------------------------------------------------------------
37
src/libnetdata/atomics/atomic_flags.h
new
+28
@@ -0,0 +1,28 @@
1
+// SPDX-License-Identifier: GPL-3.0-or-later
2
+
3
+#ifndef NETDATA_ATOMIC_FLAGS_H
4
+#define NETDATA_ATOMIC_FLAGS_H
5
+
6
+#define atomic_flags_get(pvalue) \
7
+ __atomic_load_n((pvalue), __ATOMIC_ACQUIRE)
8
+
9
+#define atomic_flags_check(pvalue, flag) \
10
+ (__atomic_load_n((pvalue), __ATOMIC_ACQUIRE) & (flag))
11
+
12
+#define atomic_flags_set(pvalue, flag) \
13
+ __atomic_or_fetch((pvalue), (flag), __ATOMIC_RELEASE)
14
+
15
+#define atomic_flags_clear(pvalue, flag) \
16
+ __atomic_and_fetch((pvalue), ~(flag), __ATOMIC_RELEASE)
17
+
18
+// returns the old flags (before applying the changes)
19
+#define atomic_flags_set_and_clear(pvalue, set, clear) ({ \
20
+ __typeof__(*pvalue) __old = *(pvalue), __new; \
21
+ do { \
22
+ __new = (__old | (set)) & ~(clear); \
23
+ } while(!__atomic_compare_exchange_n((pvalue), &__old, __new, \
24
+ false, __ATOMIC_RELEASE, __ATOMIC_RELAXED)); \
25
+ __old; \
26
+})
27
+
28
+#endif //NETDATA_ATOMIC_FLAGS_H
src/libnetdata/atomics/atomics.h
new
+8
@@ -0,0 +1,8 @@
1
+// SPDX-License-Identifier: GPL-3.0-or-later
2
+
3
+#ifndef NETDATA_ATOMICS_H
4
+#define NETDATA_ATOMICS_H
5
+
6
+#include "atomic_flags.h"
7
+
8
+#endif //NETDATA_ATOMICS_H
src/libnetdata/common.h
+1
-1
@@ -396,7 +396,7 @@ typedef uint32_t uid_t;
396
397
#define CONCAT_INDIRECT(a, b) a##b
398
#define CONCAT(a, b) CONCAT_INDIRECT(a, b)
399
-#define CACHE_LINE_PADDING() uint8_t CONCAT(padding, __COUNTER__)[64 - sizeof(size_t)];
399
+#define PAD64(type) uint8_t CONCAT(padding, __COUNTER__)[64 - sizeof(type)]; type
400
401
// --------------------------------------------------------------------------------------------------------------------
402
src/libnetdata/dictionary/dictionary-internals.h
+1
-1
@@ -47,7 +47,7 @@ typedef enum __attribute__ ((__packed__)) item_flags {
47
#define item_shared_flag_set(item, flag) __atomic_or_fetch(&((item)->shared->flags), flag, __ATOMIC_RELAXED)
48
#define item_shared_flag_clear(item, flag) __atomic_and_fetch(&((item)->shared->flags), ~(flag), __ATOMIC_RELAXED)
49
50
-#define REFCOUNT_DELETING (-100)
50
+#define REFCOUNT_DICONNECTED (-100)
51
52
#define ITEM_FLAGS_TYPE uint8_t
53
#define KEY_LEN_TYPE uint32_t
src/libnetdata/dictionary/dictionary-refcount.h
+2
-2
@@ -189,7 +189,7 @@ static inline int item_is_not_referenced_and_can_be_removed_advanced(DICTIONARY
189
// if we can set refcount to REFCOUNT_DELETING, we can delete this item
190
191
size_t spins = 0;
192
- REFCOUNT refcount, desired = REFCOUNT_DELETING;
192
+ REFCOUNT refcount, desired = REFCOUNT_DICONNECTED;
193
194
int ret = RC_ITEM_OK;
195
@@ -234,7 +234,7 @@ static inline bool item_shared_release_and_check_if_it_can_be_freed(DICTIONARY *
234
// if we can set refcount to REFCOUNT_DELETING, we can delete this item
235
236
REFCOUNT links = __atomic_sub_fetch(&item->shared->links, 1, __ATOMIC_RELEASE);
237
- if(links == 0 && __atomic_compare_exchange_n(&item->shared->links, &links, REFCOUNT_DELETING, false, __ATOMIC_ACQUIRE, __ATOMIC_RELAXED)) {
237
+ if(links == 0 && __atomic_compare_exchange_n(&item->shared->links, &links, REFCOUNT_DICONNECTED, false, __ATOMIC_ACQUIRE, __ATOMIC_RELAXED)) {
238
239
// we can delete it
240
return true;
src/libnetdata/dictionary/dictionary.h
+26
-52
@@ -66,75 +66,49 @@ struct dictionary_stats {
66
const char *name; // the name of the category
67
68
struct {
69
- CACHE_LINE_PADDING();
70
- size_t active; // the number of active dictionaries
71
- CACHE_LINE_PADDING();
72
- size_t deleted; // the number of dictionaries queued for destruction
69
+ PAD64(size_t) active; // the number of active dictionaries
70
+ PAD64(size_t) deleted; // the number of dictionaries queued for destruction
71
} dictionaries;
72
73
struct {
76
- CACHE_LINE_PADDING();
77
- long entries; // active items in the dictionary
78
- CACHE_LINE_PADDING();
79
- long pending_deletion; // pending deletion items in the dictionary
80
- CACHE_LINE_PADDING();
81
- long referenced; // referenced items in the dictionary
74
+ PAD64(long) entries; // active items in the dictionary
75
+ PAD64(long) pending_deletion; // pending deletion items in the dictionary
76
+ PAD64(long) referenced; // referenced items in the dictionary
77
} items;
78
79
struct {
85
- CACHE_LINE_PADDING();
86
- size_t creations; // dictionary creations
87
- CACHE_LINE_PADDING();
88
- size_t destructions; // dictionary destructions
89
- CACHE_LINE_PADDING();
90
- size_t flushes; // dictionary flushes
91
- CACHE_LINE_PADDING();
92
- size_t traversals; // dictionary foreach
93
- CACHE_LINE_PADDING();
94
- size_t walkthroughs; // dictionary walkthrough
95
- CACHE_LINE_PADDING();
96
- size_t garbage_collections; // dictionary garbage collections
97
- CACHE_LINE_PADDING();
98
- size_t searches; // item searches
99
- CACHE_LINE_PADDING();
100
- size_t inserts; // item inserts
101
- CACHE_LINE_PADDING();
102
- size_t resets; // item resets
103
- CACHE_LINE_PADDING();
104
- size_t deletes; // item deletes
80
+ PAD64(size_t) creations; // dictionary creations
81
+ PAD64(size_t) destructions; // dictionary destructions
82
+ PAD64(size_t) flushes; // dictionary flushes
83
+ PAD64(size_t) traversals; // dictionary foreach
84
+ PAD64(size_t) walkthroughs; // dictionary walkthrough
85
+ PAD64(size_t) garbage_collections; // dictionary garbage collections
86
+ PAD64(size_t) searches; // item searches
87
+ PAD64(size_t) inserts; // item inserts
88
+ PAD64(size_t) resets; // item resets
89
+ PAD64(size_t) deletes; // item deletes
90
} ops;
91
92
struct {
108
- CACHE_LINE_PADDING();
109
- size_t inserts; // number of times the insert callback is called
110
- CACHE_LINE_PADDING();
111
- size_t conflicts; // number of times the conflict callback is called
112
- CACHE_LINE_PADDING();
113
- size_t reacts; // number of times the react callback is called
114
- CACHE_LINE_PADDING();
115
- size_t deletes; // number of times the delete callback is called
93
+ PAD64(size_t) inserts; // number of times the insert callback is called
94
+ PAD64(size_t) conflicts; // number of times the conflict callback is called
95
+ PAD64(size_t) reacts; // number of times the react callback is called
96
+ PAD64(size_t) deletes; // number of times the delete callback is called
97
} callbacks;
98
99
// memory
100
struct {
120
- CACHE_LINE_PADDING();
121
- ssize_t index; // bytes of keys indexed (indication of the index size)
122
- CACHE_LINE_PADDING();
123
- ssize_t values; // bytes of caller structures
124
- CACHE_LINE_PADDING();
125
- ssize_t dict; // bytes of the structures dictionary needs
101
+ PAD64(ssize_t) index; // bytes of keys indexed (indication of the index size)
102
+ PAD64(ssize_t) values; // bytes of caller structures
103
+ PAD64(ssize_t) dict; // bytes of the structures dictionary needs
104
} memory;
105
106
// spin locks
107
struct {
130
- CACHE_LINE_PADDING();
131
- size_t use_spins; // number of times a reference to item had to spin to acquire it or ignore it
132
- CACHE_LINE_PADDING();
133
- size_t search_spins; // number of times a successful search result had to be thrown away
134
- CACHE_LINE_PADDING();
135
- size_t insert_spins; // number of times an insertion to the hash table had to be repeated
136
- CACHE_LINE_PADDING();
137
- size_t delete_spins; // number of times a deletion had to spin to get a decision
108
+ PAD64(size_t) use_spins; // number of times a reference to item had to spin to acquire it or ignore it
109
+ PAD64(size_t) search_spins; // number of times a successful search result had to be thrown away
110
+ PAD64(size_t) insert_spins; // number of times an insertion to the hash table had to be repeated
111
+ PAD64(size_t) delete_spins; // number of times a deletion had to spin to get a decision
112
} spin_locks;
113
};
114
src/libnetdata/http/http_defs.c
-1
@@ -6,7 +6,6 @@ ENUM_STR_MAP_DEFINE(HTTP_REQUEST_MODE) =
6
{
7
{ .name = "OPTIONS", .id = HTTP_REQUEST_MODE_OPTIONS },
8
{ .name = "GET", .id = HTTP_REQUEST_MODE_GET },
9
- { .name = "FILECOPY", .id = HTTP_REQUEST_MODE_FILECOPY },
9
{ .name = "POST", .id = HTTP_REQUEST_MODE_POST },
10
{ .name = "PUT", .id = HTTP_REQUEST_MODE_PUT },
11
{ .name = "DELETE", .id = HTTP_REQUEST_MODE_DELETE },
src/libnetdata/libjudy/judy-malloc.c
+1
-1
@@ -30,7 +30,7 @@ __attribute__((constructor)) void aral_judy_init(void) {
30
0,
31
0,
32
&judy_sizes_aral_statistics,
33
- NULL, NULL, false, false);
33
+ NULL, NULL, false, false, false);
34
}
35
}
36
src/libnetdata/libjudy/judyl-typed.h
+14
-17
@@ -5,7 +5,8 @@
5
6
#include <Judy.h>
7
8
-#define DEFINE_JUDYL_TYPED(NAME, TYPE) \
8
+// Advanced macro for types requiring conversion
9
+#define DEFINE_JUDYL_TYPED_ADVANCED(NAME, TYPE, PACK_MACRO, UNPACK_MACRO) \
10
_Static_assert(sizeof(TYPE) <= sizeof(Word_t), \
11
#NAME "_type_must_have_same_size_as_Word_t"); \
12
typedef struct { \
@@ -19,45 +20,37 @@
20
static inline bool NAME##_SET(NAME##_JudyLSet *set, Word_t index, TYPE value) { \
21
Pvoid_t *pValue = JudyLIns(&set->judyl, index, PJE0); \
22
if (pValue == PJERR) return false; \
22
- *pValue = (void *)(uintptr_t)value; \
23
+ *pValue = (void *)PACK_MACRO(value); \
24
return true; \
25
} \
26
\
27
static inline TYPE NAME##_GET(NAME##_JudyLSet *set, Word_t index) { \
28
Pvoid_t *pValue = JudyLGet(set->judyl, index, PJE0); \
28
- return (pValue != NULL) ? (TYPE)(uintptr_t)(*pValue) : (TYPE)0; \
29
- } \
30
- \
31
- static inline TYPE *NAME##_GETPTR(NAME##_JudyLSet *set, Word_t index) { \
32
- Pvoid_t *pValue = JudyLGet(set->judyl, index, PJE0); \
33
- return (TYPE *)pValue; \
29
+ return (pValue != NULL) ? (TYPE)UNPACK_MACRO(*pValue) : (TYPE){0}; \
30
} \
31
\
32
static inline bool NAME##_DEL(NAME##_JudyLSet *set, Word_t index) { \
37
- int Rc; \
38
- PPvoid_t ppJudy = &set->judyl; \
39
- Rc = JudyLDel(ppJudy, index, PJE0); \
40
- return Rc == 1; \
33
+ return JudyLDel(&set->judyl, index, PJE0) == 1; \
34
} \
35
\
36
static inline TYPE NAME##_FIRST(NAME##_JudyLSet *set, Word_t *index) { \
37
Pvoid_t *pValue = JudyLFirst(set->judyl, index, PJE0); \
45
- return (pValue != NULL) ? (TYPE)(uintptr_t)(*pValue) : (TYPE)0; \
38
+ return (pValue != NULL) ? (TYPE)UNPACK_MACRO(*pValue) : (TYPE){0}; \
39
} \
40
\
41
static inline TYPE NAME##_NEXT(NAME##_JudyLSet *set, Word_t *index) { \
42
Pvoid_t *pValue = JudyLNext(set->judyl, index, PJE0); \
50
- return (pValue != NULL) ? (TYPE)(uintptr_t)(*pValue) : (TYPE)0; \
43
+ return (pValue != NULL) ? (TYPE)UNPACK_MACRO(*pValue) : (TYPE){0}; \
44
} \
45
\
46
static inline TYPE NAME##_LAST(NAME##_JudyLSet *set, Word_t *index) { \
47
Pvoid_t *pValue = JudyLLast(set->judyl, index, PJE0); \
55
- return (pValue != NULL) ? (TYPE)(uintptr_t)(*pValue) : (TYPE)0; \
48
+ return (pValue != NULL) ? (TYPE)UNPACK_MACRO(*pValue) : (TYPE){0}; \
49
} \
50
\
51
static inline TYPE NAME##_PREV(NAME##_JudyLSet *set, Word_t *index) { \
52
Pvoid_t *pValue = JudyLPrev(set->judyl, index, PJE0); \
60
- return (pValue != NULL) ? (TYPE)(uintptr_t)(*pValue) : (TYPE)0; \
53
+ return (pValue != NULL) ? (TYPE)UNPACK_MACRO(*pValue) : (TYPE){0}; \
54
} \
55
\
56
static inline void NAME##_FREE(NAME##_JudyLSet *set, void (*callback)(Word_t, TYPE)) { \
@@ -67,12 +60,16 @@
60
for (pValue = JudyLFirst(set->judyl, &index, PJE0); \
61
pValue != NULL; \
62
pValue = JudyLNext(set->judyl, &index, PJE0)) { \
70
- callback(index, (TYPE)(uintptr_t)(*pValue)); \
63
+ callback(index, (TYPE)UNPACK_MACRO(*pValue)); \
64
} \
65
} \
66
JudyLFreeArray(&set->judyl, PJE0); \
67
}
68
69
+// Basic macro for types with no conversion
70
+#define JUDYL_TYPED_NO_CONVERSION(value) (uintptr_t)(value)
71
72
+#define DEFINE_JUDYL_TYPED(NAME, TYPE) \
73
+ DEFINE_JUDYL_TYPED_ADVANCED(NAME, TYPE, JUDYL_TYPED_NO_CONVERSION, JUDYL_TYPED_NO_CONVERSION)
74
75
#endif //NETDATA_JUDYL_TYPED_H
src/libnetdata/libnetdata.c
+11
-19
@@ -449,19 +449,11 @@ void posix_memfree(void *ptr) {
449
#endif
450
451
void mallocz_release_as_much_memory_to_the_system(void) {
452
-#if defined(HAVE_C_MALLOPT) || defined(HAVE_C_MALLOC_TRIM)
452
+#if defined(HAVE_C_MALLOC_TRIM)
453
static SPINLOCK spinlock = SPINLOCK_INITIALIZER;
454
spinlock_lock(&spinlock);
455
456
-#ifdef HAVE_C_MALLOPT
457
- // the default is 128KiB
458
- size_t trim_threshold = 65ULL * 1024;
459
- mallopt(M_TRIM_THRESHOLD, (int)trim_threshold);
460
-#endif
461
-
462
-#ifdef HAVE_C_MALLOC_TRIM
456
malloc_trim(0);
464
-#endif
457
458
spinlock_unlock(&spinlock);
459
#endif
@@ -527,7 +519,7 @@ inline int madvise_sequential(void *mem, size_t len) {
519
int ret = madvise(mem, len, MADV_SEQUENTIAL);
520
521
if (ret != 0 && logger-- > 0)
530
- netdata_log_error("madvise(MADV_SEQUENTIAL) failed.");
522
+ netdata_log_error("madvise(MADV_SEQUENTIAL) of size %zu, failed.", len);
523
return ret;
524
}
525
@@ -536,7 +528,7 @@ inline int madvise_random(void *mem, size_t len) {
528
int ret = madvise(mem, len, MADV_RANDOM);
529
530
if (ret != 0 && logger-- > 0)
539
- netdata_log_error("madvise(MADV_RANDOM) failed.");
531
+ netdata_log_error("madvise(MADV_RANDOM) of size %zu, failed.", len);
532
return ret;
533
}
534
@@ -545,7 +537,7 @@ inline int madvise_dontfork(void *mem, size_t len) {
537
int ret = madvise(mem, len, MADV_DONTFORK);
538
539
if (ret != 0 && logger-- > 0)
548
- netdata_log_error("madvise(MADV_DONTFORK) failed.");
540
+ netdata_log_error("madvise(MADV_DONTFORK) of size %zu, failed.", len);
541
return ret;
542
}
543
@@ -554,7 +546,7 @@ inline int madvise_willneed(void *mem, size_t len) {
546
int ret = madvise(mem, len, MADV_WILLNEED);
547
548
if (ret != 0 && logger-- > 0)
557
- netdata_log_error("madvise(MADV_WILLNEED) failed.");
549
+ netdata_log_error("madvise(MADV_WILLNEED) of size %zu, failed.", len);
550
return ret;
551
}
552
@@ -563,7 +555,7 @@ inline int madvise_dontneed(void *mem, size_t len) {
555
int ret = madvise(mem, len, MADV_DONTNEED);
556
557
if (ret != 0 && logger-- > 0)
566
- netdata_log_error("madvise(MADV_DONTNEED) failed.");
558
+ netdata_log_error("madvise(MADV_DONTNEED) of size %zu, failed.", len);
559
return ret;
560
}
561
@@ -573,7 +565,7 @@ inline int madvise_dontdump(void *mem __maybe_unused, size_t len __maybe_unused)
565
int ret = madvise(mem, len, MADV_DONTDUMP);
566
567
if (ret != 0 && logger-- > 0)
576
- netdata_log_error("madvise(MADV_DONTDUMP) failed.");
568
+ netdata_log_error("madvise(MADV_DONTDUMP) of size %zu, failed.", len);
569
return ret;
570
#else
571
return 0;
@@ -586,14 +578,14 @@ inline int madvise_mergeable(void *mem __maybe_unused, size_t len __maybe_unused
578
int ret = madvise(mem, len, MADV_MERGEABLE);
579
580
if (ret != 0 && logger-- > 0)
589
- netdata_log_error("madvise(MADV_MERGEABLE) failed.");
581
+ netdata_log_error("madvise(MADV_MERGEABLE) of size %zu, failed.", len);
582
return ret;
583
#else
584
return 0;
585
#endif
586
}
587
596
-void *netdata_mmap(const char *filename, size_t size, int flags, int ksm, bool read_only, int *open_fd)
588
+void *netdata_mmap(const char *filename, size_t size, int flags, int ksm, bool read_only, bool dont_dump, int *open_fd)
589
{
590
// netdata_log_info("netdata_mmap('%s', %zu", filename, size);
591
@@ -651,8 +643,8 @@ void *netdata_mmap(const char *filename, size_t size, int flags, int ksm, bool r
643
}
644
645
// madvise_sequential(mem, size);
654
- madvise_dontfork(mem, size);
655
- madvise_dontdump(mem, size);
646
+ // madvise_dontfork(mem, size); // aral is initialized before we daemonize
647
+ if(dont_dump) madvise_dontdump(mem, size);
648
// if(flags & MAP_SHARED) madvise_willneed(mem, size);
649
if(ksm) madvise_mergeable(mem, size);
650
}
src/libnetdata/libnetdata.h
+4
-3
@@ -9,15 +9,15 @@ extern "C" {
9
10
#include "common.h"
11
12
-#define JUDYHS_INDEX_SIZE_ESTIMATE(key_bytes) (((key_bytes) + sizeof(Word_t) - 1) / sizeof(Word_t) * 4)
13
-
12
// NETDATA_TRACE_ALLOCATIONS does not work under musl libc, so don't enable it
13
//#if defined(NETDATA_INTERNAL_CHECKS) && !defined(NETDATA_TRACE_ALLOCATIONS)
14
//#define NETDATA_TRACE_ALLOCATIONS 1
15
//#endif
16
17
+#include "atomics/atomics.h"
18
#include "libjudy/judy-malloc.h"
19
20
+#include "object-state/object-state.h"
21
#include "storage-point.h"
22
#include "paths/paths.h"
23
@@ -67,7 +67,7 @@ void posix_memfree(void *ptr);
67
void json_escape_string(char *dst, const char *src, size_t size);
68
void json_fix_string(char *s);
69
70
-void *netdata_mmap(const char *filename, size_t size, int flags, int ksm, bool read_only, int *open_fd);
70
+void *netdata_mmap(const char *filename, size_t size, int flags, int ksm, bool read_only, bool dont_dump, int *open_fd);
71
int netdata_munmap(void *ptr, size_t size);
72
int memory_file_save(const char *filename, void *mem, size_t size);
73
@@ -125,6 +125,7 @@ extern const char *netdata_configured_host_prefix;
125
#include "locks/spinlock.h"
126
#include "locks/rw-spinlock.h"
127
#include "completion/completion.h"
128
+#include "waiting-queue/waiting-queue.h"
129
#include "clocks/clocks.h"
130
#include "simple_pattern/simple_pattern.h"
131
#include "libnetdata/log/nd_log.h"
src/libnetdata/local-sockets/local-sockets.h
+2
-2
@@ -1192,14 +1192,14 @@ static inline void local_sockets_init(LS_STATE *ls) {
1192
sizeof(LOCAL_SOCKET),
1193
65536 / sizeof(LOCAL_SOCKET),
1194
65536,
1195
- NULL, NULL, NULL, false, true);
1195
+ NULL, NULL, NULL, false, true, true);
1196
1197
ls->pid_socket_aral = aral_create(
1198
"pid-sockets",
1199
sizeof(struct pid_socket),
1200
65536 / sizeof(struct pid_socket),
1201
65536,
1202
- NULL, NULL, NULL, false, true);
1202
+ NULL, NULL, NULL, false, true, true);
1203
1204
memset(&ls->stats, 0, sizeof(ls->stats));
1205
src/libnetdata/locks/rw-spinlock.c
+108
-45
@@ -5,78 +5,141 @@
5
// ----------------------------------------------------------------------------
6
// rw_spinlock implementation
7
8
-void rw_spinlock_init_with_trace(RW_SPINLOCK *rw_spinlock, const char *func) {
9
- rw_spinlock->readers = 0;
10
- spinlock_init_with_trace(&rw_spinlock->spinlock, func);
8
+void rw_spinlock_init_with_trace(RW_SPINLOCK *rw_spinlock, const char *func __maybe_unused) {
9
+ rw_spinlock->counter = 0;
10
+}
11
+
12
+bool rw_spinlock_tryread_lock_with_trace(RW_SPINLOCK *rw_spinlock, const char *func) {
13
+ size_t spins = 0;
14
+
15
+ REFCOUNT expected = rw_spinlock->counter;
16
+ while (true) {
17
+ if(expected < 0)
18
+ // writer is active
19
+ return false;
20
+
21
+ // increment reader count
22
+ if (__atomic_compare_exchange_n(
23
+ &rw_spinlock->counter,
24
+ &expected,
25
+ expected + 1,
26
+ false, // Strong CAS
27
+ __ATOMIC_ACQUIRE, // Success memory order
28
+ __ATOMIC_RELAXED // Failure memory order
29
+ ))
30
+ break;
31
+
32
+ spins++;
33
+ }
34
+
35
+ worker_spinlock_contention(func, spins);
36
+ nd_thread_rwspinlock_read_locked();
37
+ return true;
38
}
39
40
void rw_spinlock_read_lock_with_trace(RW_SPINLOCK *rw_spinlock, const char *func) {
14
- spinlock_lock_with_trace(&rw_spinlock->spinlock, func);
15
- __atomic_add_fetch(&rw_spinlock->readers, 1, __ATOMIC_RELAXED);
16
- spinlock_unlock_with_trace(&rw_spinlock->spinlock, func);
41
+ size_t spins = 0;
42
+
43
+ REFCOUNT expected = rw_spinlock->counter;
44
+
45
+ // we should not increase it if it is negative (a writer holds the lock)
46
+ if(expected < 0) expected = 0;
47
+
48
+ while (true) {
49
+ // Attempt to increment reader count
50
+ if (__atomic_compare_exchange_n(
51
+ &rw_spinlock->counter,
52
+ &expected,
53
+ expected + 1,
54
+ false, // Strong CAS
55
+ __ATOMIC_ACQUIRE, // Success memory order
56
+ __ATOMIC_RELAXED // Failure memory order
57
+ ))
58
+ break;
59
+
60
+ spins++;
61
+
62
+ if (expected < 0) {
63
+ // writer is active
64
65
+ // we should not increase it if it is negative (a writer holds the lock)
66
+ expected = 0;
67
+
68
+ // wait a bit before retrying
69
+ tinysleep();
70
+ yield_the_processor();
71
+ }
72
+ }
73
+
74
+ worker_spinlock_contention(func, spins);
75
nd_thread_rwspinlock_read_locked();
76
}
77
78
void rw_spinlock_read_unlock_with_trace(RW_SPINLOCK *rw_spinlock, const char *func __maybe_unused) {
79
#ifndef NETDATA_INTERNAL_CHECKS
23
- __atomic_sub_fetch(&rw_spinlock->readers, 1, __ATOMIC_RELAXED);
80
+ __atomic_sub_fetch(&rw_spinlock->counter, 1, __ATOMIC_RELEASE);
81
#else
25
- int32_t x = __atomic_sub_fetch(&rw_spinlock->readers, 1, __ATOMIC_RELAXED);
26
- if(x < 0)
82
+ REFCOUNT x = __atomic_sub_fetch(&rw_spinlock->counter, 1, __ATOMIC_RELEASE);
83
+ if (x < 0)
84
fatal("RW_SPINLOCK: readers is negative %d", x);
85
#endif
86
87
nd_thread_rwspinlock_read_unlocked();
88
}
89
90
+bool rw_spinlock_trywrite_lock_with_trace(RW_SPINLOCK *rw_spinlock, const char *func) {
91
+ REFCOUNT expected = 0;
92
+
93
+ // Attempt to acquire writer lock when no readers or writers are active
94
+ if (!__atomic_compare_exchange_n(
95
+ &rw_spinlock->counter,
96
+ &expected,
97
+ -1,
98
+ false, // Strong CAS
99
+ __ATOMIC_ACQUIRE, // Success memory order
100
+ __ATOMIC_RELAXED // Failure memory order
101
+ )) {
102
+ return false;
103
+ }
104
+
105
+ worker_spinlock_contention(func, 0);
106
+ nd_thread_rwspinlock_write_locked();
107
+ return true;
108
+}
109
+
110
void rw_spinlock_write_lock_with_trace(RW_SPINLOCK *rw_spinlock, const char *func) {
111
size_t spins = 0;
35
- while(true) {
36
- spinlock_lock_with_trace(&rw_spinlock->spinlock, func);
112
38
- if(__atomic_load_n(&rw_spinlock->readers, __ATOMIC_RELAXED) == 0)
113
+ while (true) {
114
+ REFCOUNT expected = 0;
115
+
116
+ // Attempt to acquire writer lock when no readers or writers are active
117
+ if (__atomic_compare_exchange_n(
118
+ &rw_spinlock->counter,
119
+ &expected,
120
+ -1,
121
+ false, // Strong CAS
122
+ __ATOMIC_ACQUIRE, // Success memory order
123
+ __ATOMIC_RELAXED // Failure memory order
124
+ )) {
125
break;
126
+ }
127
41
- // Busy wait until all readers have released their locks.
42
- spinlock_unlock_with_trace(&rw_spinlock->spinlock, func);
43
- tinysleep();
128
spins++;
129
+ tinysleep();
130
}
131
132
worker_spinlock_contention(func, spins);
133
nd_thread_rwspinlock_write_locked();
134
}
135
51
-void rw_spinlock_write_unlock_with_trace(RW_SPINLOCK *rw_spinlock, const char *func) {
52
- spinlock_unlock_with_trace(&rw_spinlock->spinlock, func);
53
- nd_thread_rwspinlock_write_unlocked();
54
-}
55
-
56
-bool rw_spinlock_tryread_lock_with_trace(RW_SPINLOCK *rw_spinlock, const char *func) {
57
- if(spinlock_trylock_with_trace(&rw_spinlock->spinlock, func)) {
58
- __atomic_add_fetch(&rw_spinlock->readers, 1, __ATOMIC_RELAXED);
59
- spinlock_unlock_with_trace(&rw_spinlock->spinlock, func);
60
- nd_thread_rwspinlock_read_locked();
61
- return true;
62
- }
63
-
64
- return false;
65
-}
66
-
67
-bool rw_spinlock_trywrite_lock_with_trace(RW_SPINLOCK *rw_spinlock, const char *func) {
68
- if(spinlock_trylock_with_trace(&rw_spinlock->spinlock, func)) {
69
- if (__atomic_load_n(&rw_spinlock->readers, __ATOMIC_RELAXED) == 0) {
70
- // No readers, we've successfully acquired the write lock
71
- nd_thread_rwspinlock_write_locked();
72
- return true;
73
- }
74
- else {
75
- // There are readers, unlock the spinlock and return false
76
- spinlock_unlock_with_trace(&rw_spinlock->spinlock, func);
77
- }
78
- }
136
+void rw_spinlock_write_unlock_with_trace(RW_SPINLOCK *rw_spinlock, const char *func __maybe_unused) {
137
+#ifdef NETDATA_INTERNAL_CHECKS
138
+ int32_t x = __atomic_load_n(&rw_spinlock->counter, __ATOMIC_RELAXED);
139
+ if (x != -1)
140
+ fatal("RW_SPINLOCK: writer unlock encountered unexpected state: %d", x);
141
+#endif
142
80
- return false;
143
+ __atomic_store_n(&rw_spinlock->counter, 0, __ATOMIC_RELEASE); // Release writer lock
144
+ nd_thread_rwspinlock_write_unlocked();
145
}
82
-
src/libnetdata/locks/rw-spinlock.h
+2
-3
@@ -7,11 +7,10 @@
7
#include "spinlock.h"
8
9
typedef struct netdata_rw_spinlock {
10
- int32_t readers;
11
- SPINLOCK spinlock;
10
+ REFCOUNT counter; // positive is readers, negative is a writer
11
} RW_SPINLOCK;
12
14
-#define RW_SPINLOCK_INITIALIZER { .readers = 0, .spinlock = SPINLOCK_INITIALIZER}
13
+#define RW_SPINLOCK_INITIALIZER { .counter = 0, }
14
15
void rw_spinlock_init_with_trace(RW_SPINLOCK *rw_spinlock, const char *func);
16
void rw_spinlock_read_lock_with_trace(RW_SPINLOCK *rw_spinlock, const char *func);
src/libnetdata/locks/spinlock.c
+14
-13
@@ -2,6 +2,8 @@
2
3
#include "libnetdata/libnetdata.h"
4
5
+#define MAX_USEC 512 // Maximum backoff limit in microseconds
6
+
7
// ----------------------------------------------------------------------------
8
// spinlock implementation
9
// https://www.youtube.com/watch?v=rmGJc9PXpuE&t=41s
@@ -14,22 +16,21 @@ void spinlock_init_with_trace(SPINLOCK *spinlock, const char *func __maybe_unuse
16
17
void spinlock_lock_with_trace(SPINLOCK *spinlock, const char *func) {
18
size_t spins = 0;
19
+ usec_t usec = 1;
20
18
- for(int i = 1;
19
- __atomic_load_n(&spinlock->locked, __ATOMIC_RELAXED) ||
20
- __atomic_test_and_set(&spinlock->locked, __ATOMIC_ACQUIRE)
21
- ; i++
22
- ) {
21
+ while (true) {
22
+ if (!__atomic_load_n(&spinlock->locked, __ATOMIC_RELAXED) &&
23
+ !__atomic_test_and_set(&spinlock->locked, __ATOMIC_ACQUIRE)) {
24
+ // Acquired the lock
25
+ break;
26
+ }
27
28
+ // Backoff strategy with exponential growth
29
spins++;
25
- if(unlikely(i % 8 == 0)) {
26
- i = 0;
27
- tinysleep();
28
- }
30
+ microsleep(usec);
31
+ usec = usec > MAX_USEC ? MAX_USEC : usec * 2;
32
}
33
31
- // we have the lock
32
-
34
#ifdef NETDATA_INTERNAL_CHECKS
35
spinlock->spins += spins;
36
spinlock->locker_pid = gettid_cached();
@@ -50,9 +51,9 @@ void spinlock_unlock_with_trace(SPINLOCK *spinlock, const char *func __maybe_unu
51
}
52
53
bool spinlock_trylock_with_trace(SPINLOCK *spinlock, const char *func __maybe_unused) {
53
- if(!__atomic_load_n(&spinlock->locked, __ATOMIC_RELAXED) &&
54
+ if (!__atomic_load_n(&spinlock->locked, __ATOMIC_RELAXED) &&
55
!__atomic_test_and_set(&spinlock->locked, __ATOMIC_ACQUIRE)) {
55
- // we got the lock
56
+ // Acquired the lock
57
nd_thread_spinlock_locked();
58
return true;
59
}
src/libnetdata/log/nd_log.c
+27
-9
@@ -409,21 +409,40 @@ void netdata_logger_with_limit(ERROR_LIMIT *erl, ND_LOG_SOURCES source, ND_LOG_F
409
erl->count = 0;
410
}
411
412
-void netdata_logger_fatal( const char *file, const char *function, const unsigned long line, const char *fmt, ... ) {
413
- int saved_errno = errno;
412
+void netdata_logger_fatal(const char *file, const char *function, const unsigned long line, const char *fmt, ... ) {
413
+ static size_t already_in_fatal = 0;
414
+
415
+ size_t recursion = __atomic_add_fetch(&already_in_fatal, 1, __ATOMIC_SEQ_CST);
416
+ if(recursion > 1) {
417
+ // exit immediately, nothing more to be done
418
+ fprintf(stderr, "RECURSIVE FATAL STATEMENTS, latest from %lu@%s() of %s, EXITING NOW!\n",
419
+ line, function, file);
420
+ fflush(stderr);
421
+ _exit(1);
422
+ }
423
424
+ int saved_errno = errno;
425
size_t saved_winerror = 0;
426
#if defined(OS_WINDOWS)
427
saved_winerror = GetLastError();
428
#endif
429
420
- ND_LOG_SOURCES source = NDLS_DAEMON;
421
- source = nd_log_validate_source(source);
430
+ // make sure the msg id does not leak
431
+ {
432
+ ND_LOG_STACK lgs[] = {
433
+ ND_LOG_FIELD_UUID(NDF_MESSAGE_ID, &netdata_fatal_msgid),
434
+ ND_LOG_FIELD_END(),
435
+ };
436
+ ND_LOG_STACK_PUSH(lgs);
437
423
- va_list args;
424
- va_start(args, fmt);
425
- nd_logger(file, function, line, source, NDLP_ALERT, true, saved_errno, saved_winerror, fmt, args);
426
- va_end(args);
438
+ ND_LOG_SOURCES source = NDLS_DAEMON;
439
+ source = nd_log_validate_source(source);
440
+
441
+ va_list args;
442
+ va_start(args, fmt);
443
+ nd_logger(file, function, line, source, NDLP_ALERT, true, saved_errno, saved_winerror, fmt, args);
444
+ va_end(args);
445
+ }
446
447
char date[LOG_DATE_LENGTH];
448
log_date(date, LOG_DATE_LENGTH, now_realtime_sec());
@@ -462,4 +481,3 @@ void netdata_logger_fatal( const char *file, const char *function, const unsigne
481
482
netdata_cleanup_and_exit(1, "FATAL", action_result, action_data);
483
}
465
-
src/libnetdata/log/systemd-journal-helpers.c
+1
-1
@@ -51,7 +51,7 @@ int journal_direct_fd(const char *path) {
51
int fd = socket(AF_UNIX, SOCK_DGRAM| DEFAULT_SOCKET_FLAGS, 0);
52
if (fd < 0) return -1;
53
54
- sock_setcloexec(fd);
54
+ sock_setcloexec(fd, true);
55
56
struct sockaddr_un addr;
57
memset(&addr, 0, sizeof(struct sockaddr_un));
src/libnetdata/object-state/object-state.c
new
+76
@@ -0,0 +1,76 @@
1
+#include "object-state.h"
2
+
3
+OBJECT_STATE_ID object_state_id(OBJECT_STATE *os) {
4
+ return __atomic_load_n(&os->state_id, __ATOMIC_ACQUIRE);
5
+}
6
+
7
+void object_state_activate(OBJECT_STATE *os) {
8
+ __atomic_add_fetch(&os->state_id, 1, __ATOMIC_RELAXED);
9
+
10
+ REFCOUNT expected = __atomic_load_n(&os->state_refcount, __ATOMIC_RELAXED);
11
+ REFCOUNT desired;
12
+
13
+ do {
14
+ if(expected != OBJECT_STATE_DEACTIVATED) {
15
+ fatal("OBJECT_STATE: attempt to activate already activated object (state refcount is %d)", expected);
16
+ return;
17
+ }
18
+
19
+ desired = 0;
20
+
21
+ } while(!__atomic_compare_exchange_n(
22
+ &os->state_refcount, &expected, desired, false, __ATOMIC_ACQUIRE, __ATOMIC_RELAXED));
23
+}
24
+
25
+void object_state_deactivate(OBJECT_STATE *os) {
26
+ __atomic_add_fetch(&os->state_id, 1, __ATOMIC_RELAXED);
27
+
28
+ REFCOUNT expected = __atomic_load_n(&os->state_refcount, __ATOMIC_RELAXED);
29
+ REFCOUNT desired;
30
+
31
+ do {
32
+ if(expected == OBJECT_STATE_DEACTIVATED) {
33
+ fatal("OBJECT_STATE: attempt to deactivate object that is already deactivated (state refcount %d)", expected);
34
+ return;
35
+ }
36
+ else if(expected < 0) {
37
+ fatal("OBJECT_STATE: attempt to deactivate object that is already deactivating (state refcount %d)", expected);
38
+ return;
39
+ }
40
+
41
+ // Current (-INT32_MAX) + holders
42
+ desired = OBJECT_STATE_DEACTIVATED + expected;
43
+
44
+ } while(!__atomic_compare_exchange_n(
45
+ &os->state_refcount, &expected, desired, false, __ATOMIC_SEQ_CST, __ATOMIC_RELAXED));
46
+
47
+ // Now wait for all holders to release
48
+ while(__atomic_load_n(&os->state_refcount, __ATOMIC_ACQUIRE) != OBJECT_STATE_DEACTIVATED)
49
+ tinysleep(); // Busy wait until all holders are gone
50
+}
51
+
52
+bool object_state_acquire(OBJECT_STATE *os, OBJECT_STATE_ID wanted_state_id) {
53
+ REFCOUNT expected = __atomic_load_n(&os->state_refcount, __ATOMIC_RELAXED);
54
+ REFCOUNT desired;
55
+
56
+ do {
57
+ // If refcount is negative, it means deactivation is in progress or complete
58
+ if(expected < 0)
59
+ return false;
60
+
61
+ desired = expected + 1;
62
+
63
+ } while(!__atomic_compare_exchange_n(
64
+ &os->state_refcount, &expected, desired, false, __ATOMIC_ACQUIRE, __ATOMIC_RELAXED));
65
+
66
+ if(object_state_id(os) != wanted_state_id) {
67
+ object_state_release(os);
68
+ return false;
69
+ }
70
+
71
+ return true;
72
+}
73
+
74
+void object_state_release(OBJECT_STATE *os) {
75
+ __atomic_sub_fetch(&os->state_refcount, 1, __ATOMIC_RELEASE);
76
+}
src/libnetdata/object-state/object-state.h
new
+34
@@ -0,0 +1,34 @@
1
+// SPDX-License-Identifier: GPL-3.0-or-later
2
+
3
+#ifndef NETDATA_OBJECT_STATE_ID_H
4
+#define NETDATA_OBJECT_STATE_ID_H
5
+
6
+#include "libnetdata/libnetdata.h"
7
+
8
+typedef uint32_t OBJECT_STATE_ID;
9
+
10
+typedef struct {
11
+ OBJECT_STATE_ID state_id;
12
+ REFCOUNT state_refcount;
13
+} OBJECT_STATE;
14
+
15
+#define OBJECT_STATE_DEACTIVATED (-INT32_MAX)
16
+
17
+#define OBJECT_STATE_INIT_ACTIVATED (OBJECT_STATE){ .state_id = 0, .state_refcount = 0, };
18
+#define OBJECT_STATE_INIT_DEACTIVATED (OBJECT_STATE){ .state_id = 0, .state_refcount = OBJECT_STATE_DEACTIVATED, };
19
+
20
+// get the current state id of the object
21
+OBJECT_STATE_ID object_state_id(OBJECT_STATE *os);
22
+
23
+// increments the object's state id
24
+// enables using the object - users may acquire and release the object
25
+void object_state_activate(OBJECT_STATE *os);
26
+
27
+// increments the object's state id
28
+// prevents users from acquiring it, and waits until all of its holders have released it
29
+void object_state_deactivate(OBJECT_STATE *os);
30
+
31
+bool object_state_acquire(OBJECT_STATE *os, OBJECT_STATE_ID wanted_state_id);
32
+void object_state_release(OBJECT_STATE *os);
33
+
34
+#endif //NETDATA_OBJECT_STATE_ID_H
src/libnetdata/required_dummies.h
+1
-1
@@ -14,7 +14,7 @@ void netdata_cleanup_and_exit(int ret, const char *action, const char *action_re
14
}
15
16
void rrdset_thread_rda_free(void){}
17
-void sender_commit_thread_buffer_free(void){}
17
+void sender_thread_buffer_free(void){}
18
void query_target_free(void){}
19
void service_exits(void){}
20
void rrd_collector_finished(void){}
src/libnetdata/socket/connect-to.c
+2
-2
@@ -26,7 +26,7 @@ static inline int connect_to_unix(const char *path, struct timeval *timeout) {
26
path);
27
}
28
29
- sock_setcloexec(fd);
29
+ sock_setcloexec(fd, true);
30
31
struct sockaddr_un addr;
32
memset(&addr, 0, sizeof(addr));
@@ -145,7 +145,7 @@ int connect_to_this_ip46(
145
"Failed to set timeout on the socket to ip '%s' port '%s'",
146
hostBfr, servBfr);
147
}
148
- sock_setcloexec(fd);
148
+ sock_setcloexec(fd, true);
149
150
errno_clear();
151
if(connect(fd, ai->ai_addr, ai->ai_addrlen) < 0) {
src/libnetdata/socket/listen-sockets.c
+51
-13
@@ -70,9 +70,13 @@ static int create_listen_socket_unix(const char *path, int listen_backlog) {
70
return -1;
71
}
72
73
- sock_setnonblock(sock);
74
- sock_setcloexec(sock);
75
- sock_enlarge_in(sock);
73
+ if(sock_setnonblock(sock, true) != 1)
74
+ nd_log(NDLS_DAEMON, NDLP_ERR,
75
+ "LISTENER: UNIX socket on path '%s' failed to set non-blocking mode.",
76
+ path);
77
+
78
+ sock_setcloexec(sock, true);
79
+ sock_enlarge_rcv_buf(sock);
80
81
struct sockaddr_un name;
82
memset(&name, 0, sizeof(struct sockaddr_un));
@@ -124,11 +128,24 @@ static int create_listen_socket4(int socktype, const char *ip, uint16_t port, in
128
129
return -1;
130
}
127
- sock_setreuse(sock, 1);
128
- sock_setreuse_port(sock, 0);
129
- sock_setnonblock(sock);
130
- sock_setcloexec(sock);
131
- sock_enlarge_in(sock);
131
+
132
+ if(sock_setreuse_addr(sock, true) != 1)
133
+ nd_log(NDLS_DAEMON, NDLP_ERR,
134
+ "LISTENER: IPv4 socket on ip '%s' port %d, socktype %d failed to enable reuse address.",
135
+ ip, port, socktype);
136
+
137
+ if(sock_setreuse_port(sock, false) == 1) // -1 means not supported
138
+ nd_log(NDLS_DAEMON, NDLP_ERR,
139
+ "LISTENER: IPv4 socket on ip '%s' port %d, socktype %d failed to disable reuse port.",
140
+ ip, port, socktype);
141
+
142
+ if(sock_setnonblock(sock, true) != 1)
143
+ nd_log(NDLS_DAEMON, NDLP_ERR,
144
+ "LISTENER: IPv4 socket on ip '%s' port %d, socktype %d failed to set non-blocking mode.",
145
+ ip, port, socktype);
146
+
147
+ sock_setcloexec(sock, true);
148
+ sock_enlarge_rcv_buf(sock);
149
150
struct sockaddr_in name;
151
memset(&name, 0, sizeof(struct sockaddr_in));
@@ -163,6 +180,10 @@ static int create_listen_socket4(int socktype, const char *ip, uint16_t port, in
180
return -1;
181
}
182
183
+ // Add TCP_DEFER_ACCEPT for TCP sockets
184
+ if(socktype == SOCK_STREAM)
185
+ sock_set_tcp_defer_accept(sock, true);
186
+
187
nd_log(NDLS_DAEMON, NDLP_DEBUG,
188
"LISTENER: Listening on IPv4 ip '%s' port %d, socktype %d",
189
ip, port, socktype);
@@ -182,11 +203,24 @@ static int create_listen_socket6(int socktype, uint32_t scope_id, const char *ip
203
204
return -1;
205
}
185
- sock_setreuse(sock, 1);
186
- sock_setreuse_port(sock, 0);
187
- sock_setnonblock(sock);
188
- sock_setcloexec(sock);
189
- sock_enlarge_in(sock);
206
+
207
+ if(sock_setreuse_addr(sock, true) != 1)
208
+ nd_log(NDLS_DAEMON, NDLP_ERR,
209
+ "LISTENER: IPv6 socket on ip '%s' port %d, socktype %d failed to set reuse address.",
210
+ ip, port, socktype);
211
+
212
+ if(sock_setreuse_port(sock, false) == 1) // -1 means not supported
213
+ nd_log(NDLS_DAEMON, NDLP_ERR,
214
+ "LISTENER: IPv6 socket on ip '%s' port %d, socktype %d failed to disable reuse port.",
215
+ ip, port, socktype);
216
+
217
+ if(sock_setnonblock(sock, true) != 1)
218
+ nd_log(NDLS_DAEMON, NDLP_ERR,
219
+ "LISTENER: IPv6 socket on ip '%s' port %d, socktype %d, failed to set non-blocking mode.",
220
+ ip, port, socktype);
221
+
222
+ sock_setcloexec(sock, true);
223
+ sock_enlarge_rcv_buf(sock);
224
225
/* IPv6 only */
226
if(setsockopt(sock, IPPROTO_IPV6, IPV6_V6ONLY, (void*)&ipv6only, sizeof(ipv6only)) != 0)
@@ -230,6 +264,10 @@ static int create_listen_socket6(int socktype, uint32_t scope_id, const char *ip
264
return -1;
265
}
266
267
+ // Add TCP_DEFER_ACCEPT for TCP sockets
268
+ if(socktype == SOCK_STREAM)
269
+ sock_set_tcp_defer_accept(sock, true);
270
+
271
nd_log(NDLS_DAEMON, NDLP_DEBUG,
272
"LISTENER: Listening on IPv6 ip '%s' port %d, socktype %d",
273
ip, port, socktype);
src/libnetdata/socket/nd-poll.c
+112
-50
@@ -2,6 +2,10 @@
2
3
#include "nd-poll.h"
4
5
+#ifndef POLLRDHUP
6
+#define POLLRDHUP 0
7
+#endif
8
+
9
#if defined(OS_LINUX)
10
#include <sys/epoll.h>
11
@@ -13,6 +17,7 @@ struct nd_poll_t {
17
struct epoll_event ev[MAX_EVENTS_PER_CALL];
18
size_t last_pos;
19
size_t used;
20
+ size_t nfds;
21
};
22
23
// Initialize the event poll context
@@ -28,53 +33,92 @@ nd_poll_t *nd_poll_create() {
33
return ndpl;
34
}
35
36
+static inline void nd_poll_replace_data_on_loaded_events(nd_poll_t *ndpl, int fd __maybe_unused, void *old_data, void *new_data) {
37
+ for(size_t i = ndpl->last_pos; i < ndpl->used; i++) {
38
+ if(ndpl->ev[i].data.ptr == old_data)
39
+ ndpl->ev[i].data.ptr = new_data;
40
+ }
41
+}
42
+
43
// Add a file descriptor to the event poll
44
bool nd_poll_add(nd_poll_t *ndpl, int fd, nd_poll_event_t events, void *data) {
45
+ internal_fatal(!data, "nd_poll() does not support NULL data pointers");
46
+
47
struct epoll_event ev = {
48
.events = (events & ND_POLL_READ ? EPOLLIN : 0) | (events & ND_POLL_WRITE ? EPOLLOUT : 0),
49
.data.ptr = data,
50
};
37
- return epoll_ctl(ndpl->epoll_fd, EPOLL_CTL_ADD, fd, &ev) == 0;
51
+ bool rc = epoll_ctl(ndpl->epoll_fd, EPOLL_CTL_ADD, fd, &ev) == 0;
52
+ if(rc) ndpl->nfds++;
53
+ internal_fatal(!rc, "epoll_ctl() failed");
54
+ return rc;
55
}
56
57
// Remove a file descriptor from the event poll
41
-bool nd_poll_del(nd_poll_t *ndpl, int fd) {
42
- return epoll_ctl(ndpl->epoll_fd, EPOLL_CTL_DEL, fd, NULL) == 0;
58
+bool nd_poll_del(nd_poll_t *ndpl, int fd, void *data) {
59
+ internal_fatal(!data, "nd_poll() does not support NULL data pointers");
60
+
61
+ ndpl->nfds--; // we can't check for success/failure here, because epoll() removes fds when they are closed
62
+ bool rc = epoll_ctl(ndpl->epoll_fd, EPOLL_CTL_DEL, fd, NULL) == 0;
63
+ internal_error(!rc, "epoll_ctl() failed (is the socket already closed)"); // this is ok if the socket is already closed
64
+
65
+ // we may have an event pending for this fd.
66
+ // but epoll() does not give us fd in the events,
67
+ // so we use the data pointer to invalidate it
68
+ nd_poll_replace_data_on_loaded_events(ndpl, fd, data, NULL);
69
+ return rc;
70
}
71
72
// Update an existing file descriptor in the event poll
73
bool nd_poll_upd(nd_poll_t *ndpl, int fd, nd_poll_event_t events, void *data) {
74
+ internal_fatal(!data, "nd_poll() does not support NULL data pointers - you should also NEVER change the pointer with an update");
75
+
76
struct epoll_event ev = {
77
.events = (events & ND_POLL_READ ? EPOLLIN : 0) | (events & ND_POLL_WRITE ? EPOLLOUT : 0),
78
.data.ptr = data,
79
};
51
- return epoll_ctl(ndpl->epoll_fd, EPOLL_CTL_MOD, fd, &ev) == 0;
80
+ bool rc = epoll_ctl(ndpl->epoll_fd, EPOLL_CTL_MOD, fd, &ev) == 0;
81
+ internal_fatal(!rc, "epoll_ctl() failed");
82
+ return rc;
83
}
84
54
-static inline bool nd_poll_get_next_event(nd_poll_t *ndpl, nd_poll_result_t *result) {
55
- for(size_t i = ndpl->last_pos; i < ndpl->used ;i++) {
56
- *result = (nd_poll_result_t){
57
- .events = 0,
58
- .data = ndpl->ev[i].data.ptr,
59
- };
85
+static inline nd_poll_event_t nd_poll_events_from_epoll_events(uint32_t events) {
86
+ nd_poll_event_t nd_poll_events = ND_POLL_NONE;
87
+
88
+ if (events & (EPOLLIN|EPOLLPRI|EPOLLRDNORM|EPOLLRDBAND))
89
+ nd_poll_events |= ND_POLL_READ;
90
61
- if (ndpl->ev[i].events & EPOLLIN)
62
- result->events |= ND_POLL_READ;
91
+ if (events & (EPOLLOUT|EPOLLWRNORM|EPOLLWRBAND))
92
+ nd_poll_events |= ND_POLL_WRITE;
93
64
- if (ndpl->ev[i].events & EPOLLOUT)
65
- result->events |= ND_POLL_WRITE;
94
+ if (events & EPOLLERR)
95
+ nd_poll_events |= ND_POLL_ERROR;
96
67
- if (ndpl->ev[i].events & EPOLLERR)
68
- result->events |= ND_POLL_ERROR;
97
+ if (events & (EPOLLHUP|EPOLLRDHUP))
98
+ nd_poll_events |= ND_POLL_HUP;
99
70
- if (ndpl->ev[i].events & EPOLLHUP)
71
- result->events |= ND_POLL_HUP;
100
+ return nd_poll_events;
101
+}
102
73
- ndpl->last_pos = i + 1;
103
+static inline bool nd_poll_get_next_event(nd_poll_t *ndpl, nd_poll_result_t *result) {
104
+ while(ndpl->last_pos < ndpl->used) {
105
+ void *data = ndpl->ev[ndpl->last_pos].data.ptr;
106
+
107
+ // Skip events that have been invalidated by nd_poll_del()
108
+ if(!data) {
109
+ ndpl->last_pos++;
110
+ continue;
111
+ }
112
+
113
+ *result = (nd_poll_result_t){
114
+ .events = nd_poll_events_from_epoll_events(ndpl->ev[ndpl->last_pos].events),
115
+ .data = data,
116
+ };
117
+
118
+ ndpl->last_pos++;
119
return true;
120
}
121
77
- ndpl->last_pos = _countof(ndpl->ev);
122
return false;
123
}
124
@@ -87,10 +131,15 @@ int nd_poll_wait(nd_poll_t *ndpl, int timeout_ms, nd_poll_result_t *result) {
131
errno_clear();
132
ndpl->last_pos = 0;
133
ndpl->used = 0;
90
- int n = epoll_wait(ndpl->epoll_fd, &ndpl->ev[0], _countof(ndpl->ev), timeout_ms);
134
+
135
+ int maxevents = ndpl->nfds / 2;
136
+ if(maxevents > (int)_countof(ndpl->ev)) maxevents = (int)_countof(ndpl->ev);
137
+ if(maxevents < 2) maxevents = 2;
138
+
139
+ int n = epoll_wait(ndpl->epoll_fd, &ndpl->ev[0], maxevents, timeout_ms);
140
141
if(unlikely(n <= 0)) {
93
- if (n == 0) {
142
+ if(n == 0) {
143
result->events = ND_POLL_TIMEOUT;
144
result->data = NULL;
145
return 0;
@@ -99,15 +148,17 @@ int nd_poll_wait(nd_poll_t *ndpl, int timeout_ms, nd_poll_result_t *result) {
148
if(errno == EINTR || errno == EAGAIN)
149
continue;
150
102
- result->events = ND_POLL_OTHER_ERROR;
151
+ result->events = ND_POLL_POLL_FAILED;
152
result->data = NULL;
153
return -1;
154
}
155
156
ndpl->used = n;
157
+ ndpl->last_pos = 0;
158
if (nd_poll_get_next_event(ndpl, result))
159
return 1;
160
161
+ internal_fatal(true, "nd_poll_get_next_event() should have 1 event!");
162
} while(true);
163
}
164
@@ -172,13 +223,18 @@ bool nd_poll_add(nd_poll_t *ndpl, int fd, nd_poll_event_t events, void *data) {
223
}
224
225
// Remove a file descriptor from the event poll
175
-bool nd_poll_del(nd_poll_t *ndpl, int fd) {
226
+bool nd_poll_del(nd_poll_t *ndpl, int fd, void *data __maybe_unused) {
227
for (nfds_t i = 0; i < ndpl->nfds; i++) {
228
if (ndpl->fds[i].fd == fd) {
229
+
230
// Remove the file descriptor by shifting the array
231
memmove(&ndpl->fds[i], &ndpl->fds[i + 1], (ndpl->nfds - i - 1) * sizeof(struct pollfd));
232
ndpl->nfds--;
233
POINTERS_DEL(&ndpl->pointers, fd);
234
+
235
+ if(i < ndpl->last_pos)
236
+ ndpl->last_pos--;
237
+
238
return true;
239
}
240
}
@@ -205,31 +261,37 @@ bool nd_poll_upd(nd_poll_t *ndpl, int fd, nd_poll_event_t events, void *data) {
261
return false;
262
}
263
208
-static inline bool nd_poll_get_next_event(nd_poll_t *ndpl, nd_poll_result_t *result) {
209
- for (nfds_t i = ndpl->last_pos; i < ndpl->nfds; i++) {
210
- if (ndpl->fds[i].revents != 0) {
264
+static inline nd_poll_event_t nd_poll_events_from_poll_revents(short int events) {
265
+ nd_poll_event_t nd_poll_events = ND_POLL_NONE;
266
212
- result->data = POINTERS_GET(&ndpl->pointers, ndpl->fds[i].fd);
267
+ if (events & (POLLIN|POLLPRI|POLLRDNORM|POLLRDBAND))
268
+ nd_poll_events |= ND_POLL_READ;
269
214
- result->events = 0;
215
- if (ndpl->fds[i].revents & (POLLIN|POLLPRI))
216
- result->events |= ND_POLL_READ;
270
+ if (events & (POLLOUT|POLLWRNORM|POLLWRBAND))
271
+ nd_poll_events |= ND_POLL_WRITE;
272
218
- if (ndpl->fds[i].revents & POLLOUT)
219
- result->events |= ND_POLL_WRITE;
273
+ if (events & POLLERR)
274
+ nd_poll_events |= ND_POLL_ERROR;
275
221
- if (ndpl->fds[i].revents & POLLERR)
222
- result->events |= ND_POLL_ERROR;
276
+ if (events & (POLLHUP|POLLRDHUP))
277
+ nd_poll_events |= ND_POLL_HUP;
278
224
- if (ndpl->fds[i].revents & POLLHUP)
225
- result->events |= ND_POLL_HUP;
279
+ if (events & (POLLNVAL))
280
+ nd_poll_events |= ND_POLL_INVALID;
281
+
282
+ return nd_poll_events;
283
+}
284
+
285
+static inline bool nd_poll_get_next_event(nd_poll_t *ndpl, nd_poll_result_t *result) {
286
+ for (nfds_t i = ndpl->last_pos; i < ndpl->nfds; i++) {
287
+ if (ndpl->fds[i].revents != 0) {
288
227
- if (ndpl->fds[i].revents & POLLNVAL)
228
- result->events |= ND_POLL_INVALID;
289
+ result->data = POINTERS_GET(&ndpl->pointers, ndpl->fds[i].fd);
290
+ result->events = nd_poll_events_from_poll_revents(ndpl->fds[i].revents);
291
+ ndpl->fds[i].revents = 0;
292
230
- ndpl->fds[i].revents = 0; // Clear the event after handling
293
ndpl->last_pos = i + 1;
232
- return true; // Return only the first triggered event
294
+ return true;
295
}
296
}
297
@@ -259,25 +321,25 @@ int nd_poll_wait(nd_poll_t *ndpl, int timeout_ms, nd_poll_result_t *result) {
321
int ret = poll(ndpl->fds, ndpl->nfds, timeout_ms);
322
323
if(unlikely(ret <= 0)) {
262
- if (ret < 0) {
263
- if(errno == EAGAIN || errno == EINTR)
264
- continue;
265
-
266
- result->events = ND_POLL_OTHER_ERROR;
267
- result->data = NULL;
268
- return -1;
269
- }
270
- else {
324
+ if(ret == 0) {
325
result->events = ND_POLL_TIMEOUT;
326
result->data = NULL;
327
return 0;
328
}
329
+
330
+ if(errno == EAGAIN || errno == EINTR)
331
+ continue;
332
+
333
+ result->events = ND_POLL_POLL_FAILED;
334
+ result->data = NULL;
335
+ return -1;
336
}
337
338
// Process the next event
339
if (nd_poll_get_next_event(ndpl, result))
340
return 1;
341
342
+ internal_fatal(true, "nd_poll_get_next_event() should have 1 event!");
343
} while (true);
344
}
345
src/libnetdata/socket/nd-poll.h
+25
-11
@@ -3,16 +3,19 @@
3
#ifndef NETDATA_ND_POLL_H
4
#define NETDATA_ND_POLL_H
5
6
-#include "libnetdata/libnetdata.h"
6
+#include "libnetdata/common.h"
7
+
8
+typedef enum __attribute__((packed)) {
9
+ ND_POLL_NONE = 0,
10
+
11
+ ND_POLL_READ = 1 << 0, // same as EPOLLIN, POLLIN
12
+ ND_POLL_WRITE = 1 << 2, // same as EPOLLOUT, POLLOUT
13
+ ND_POLL_ERROR = 1 << 3, // same as EPOLLERR, POLLERR
14
+ ND_POLL_HUP = 1 << 4, // same as EPOLLHUP, POLLHUP
15
+ ND_POLL_INVALID = 1 << 5, // same as POLLNVAL
16
8
-typedef enum {
9
- ND_POLL_READ = 1 << 0,
10
- ND_POLL_WRITE = 1 << 1,
11
- ND_POLL_ERROR = 1 << 2,
12
- ND_POLL_HUP = 1 << 3,
13
- ND_POLL_INVALID = 1 << 4,
14
- ND_POLL_TIMEOUT = 1 << 5,
15
- ND_POLL_OTHER_ERROR = 1 << 6,
17
+ ND_POLL_TIMEOUT = 1 << 6,
18
+ ND_POLL_POLL_FAILED = 1 << 7,
19
} nd_poll_event_t;
20
21
typedef struct {
@@ -23,13 +26,24 @@ typedef struct {
26
typedef struct nd_poll_t nd_poll_t;
27
28
nd_poll_t *nd_poll_create();
29
+void nd_poll_destroy(nd_poll_t *ndpl);
30
+
31
+// the events can be updated with nd_poll_upd
32
+// the data pointer SHOULD NEVER be changed and cannot be NULL
33
bool nd_poll_add(nd_poll_t *ndpl, int fd, nd_poll_event_t events, void *data);
27
-bool nd_poll_del(nd_poll_t *ndpl, int fd);
34
+
35
+// give the same data pointer used in nd_poll_add()
36
+// otherwise, you may receive back invalid events
37
+bool nd_poll_del(nd_poll_t *ndpl, int fd, void *data);
38
+
39
+// this is for updating events
40
+// the data pointer must be the same used in nd_poll_add()
41
+// to change the data pointer, delete and add the same fd again
42
bool nd_poll_upd(nd_poll_t *ndpl, int fd, nd_poll_event_t events, void *data);
43
44
// returns -1 = error, 0 = timeout, 1 = event in result
45
int nd_poll_wait(nd_poll_t *ndpl, int timeout_ms, nd_poll_result_t *result);
46
33
-void nd_poll_destroy(nd_poll_t *ndpl);
47
+#include "libnetdata/libnetdata.h"
48
49
#endif //NETDATA_ND_POLL_H
src/libnetdata/socket/poll-events.c
+151
-294
@@ -2,25 +2,31 @@
2
3
#include "libnetdata/libnetdata.h"
4
5
+static inline void poll_process_updated_events(POLLINFO *pi) {
6
+ if(pi->events != pi->events_we_wait_for) {
7
+ if(!nd_poll_upd(pi->p->ndpl, pi->fd, pi->events, pi))
8
+ nd_log(NDLS_DAEMON, NDLP_ERR, "Failed to update socket %d to nd_poll", pi->fd);
9
+ pi->events_we_wait_for = pi->events;
10
+ }
11
+}
12
+
13
// poll() based listener
14
// this should be the fastest possible listener for up to 100 sockets
15
// above 100, an epoll() interface is needed on Linux
16
9
-#define POLL_FDS_INCREASE_STEP 10
10
-
11
-inline POLLINFO *poll_add_fd(POLLJOB *p
12
- , int fd
13
- , int socktype
14
- , HTTP_ACL port_acl
15
- , uint32_t flags
16
- , const char *client_ip
17
- , const char *client_port
18
- , const char *client_host
19
- , void *(*add_callback)(POLLINFO * /*pi*/, short int * /*events*/, void * /*data*/)
20
- , void (*del_callback)(POLLINFO * /*pi*/)
21
- , int (*rcv_callback)(POLLINFO * /*pi*/, short int * /*events*/)
22
- , int (*snd_callback)(POLLINFO * /*pi*/, short int * /*events*/)
23
- , void *data
17
+POLLINFO *poll_add_fd(POLLJOB *p
18
+ , int fd
19
+ , int socktype
20
+ , HTTP_ACL port_acl
21
+ , uint32_t flags
22
+ , const char *client_ip
23
+ , const char *client_port
24
+ , const char *client_host
25
+ , poll_events_add_callback_t add_callback
26
+ , poll_events_del_callback_t del_callback
27
+ , poll_events_rcv_callback_t rcv_callback
28
+ , poll_events_snd_callback_t snd_callback
29
+ , void *data
30
) {
31
if(unlikely(fd < 0)) return NULL;
32
@@ -30,56 +36,14 @@ inline POLLINFO *poll_add_fd(POLLJOB *p
36
// return NULL;
37
//}
38
33
- if(unlikely(!p->first_free)) {
34
- size_t new_slots = p->slots + POLL_FDS_INCREASE_STEP;
35
-
36
- p->fds = reallocz(p->fds, sizeof(struct pollfd) * new_slots);
37
- p->inf = reallocz(p->inf, sizeof(POLLINFO) * new_slots);
38
-
39
- // reset all the newly added slots
40
- ssize_t i;
41
- for(i = new_slots - 1; i >= (ssize_t)p->slots ; i--) {
42
- p->fds[i].fd = -1;
43
- p->fds[i].events = 0;
44
- p->fds[i].revents = 0;
45
-
46
- p->inf[i].p = p;
47
- p->inf[i].slot = (size_t)i;
48
- p->inf[i].flags = 0;
49
- p->inf[i].socktype = -1;
50
- p->inf[i].port_acl = -1;
51
-
52
- p->inf[i].client_ip = NULL;
53
- p->inf[i].client_port = NULL;
54
- p->inf[i].client_host = NULL;
55
- p->inf[i].del_callback = p->del_callback;
56
- p->inf[i].rcv_callback = p->rcv_callback;
57
- p->inf[i].snd_callback = p->snd_callback;
58
- p->inf[i].data = NULL;
59
-
60
- // link them so that the first free will be earlier in the array
61
- // (we loop decrementing i)
62
- p->inf[i].next = p->first_free;
63
- p->first_free = &p->inf[i];
64
- }
65
-
66
- p->slots = new_slots;
67
- }
68
-
69
- POLLINFO *pi = p->first_free;
70
- p->first_free = p->first_free->next;
71
-
72
- struct pollfd *pf = &p->fds[pi->slot];
73
- pf->fd = fd;
74
- pf->events = POLLIN;
75
- pf->revents = 0;
39
+ POLLINFO *pi = callocz(1, sizeof(*pi));
40
41
pi->fd = fd;
42
+ pi->events = ND_POLL_READ;
43
pi->p = p;
44
pi->socktype = socktype;
45
pi->port_acl = port_acl;
46
pi->flags = flags;
82
- pi->next = NULL;
47
pi->client_ip = strdupz(client_ip);
48
pi->client_port = strdupz(client_port);
49
pi->client_host = strdupz(client_host);
@@ -96,81 +60,48 @@ inline POLLINFO *poll_add_fd(POLLJOB *p
60
pi->send_count = 0;
61
62
p->used++;
99
- if(unlikely(pi->slot > p->max))
100
- p->max = pi->slot;
63
102
- if(pi->flags & POLLINFO_FLAG_CLIENT_SOCKET) {
103
- pi->data = add_callback(pi, &pf->events, data);
104
- }
64
+ if(pi->flags & POLLINFO_FLAG_CLIENT_SOCKET)
65
+ pi->data = add_callback(pi, &pi->events, data);
66
106
- if(pi->flags & POLLINFO_FLAG_SERVER_SOCKET) {
107
- p->min = pi->slot;
108
- }
67
+ DOUBLE_LINKED_LIST_APPEND_ITEM_UNSAFE(p->ll, pi, prev, next);
68
+ pi->events_we_wait_for = pi->events;
69
+ if(!nd_poll_add(pi->p->ndpl, pi->fd, pi->events, pi))
70
+ nd_log(NDLS_DAEMON, NDLP_ERR, "Failed to add socket %d to nd_poll", pi->fd);
71
72
return pi;
73
}
74
113
-inline void poll_close_fd(POLLINFO *pi) {
75
+static inline void poll_close_fd(POLLINFO *pi, const char *func) {
76
POLLJOB *p = pi->p;
77
116
- struct pollfd *pf = &p->fds[pi->slot];
117
-
118
- if(unlikely(pf->fd == -1)) return;
78
+ DOUBLE_LINKED_LIST_REMOVE_ITEM_UNSAFE(p->ll, pi, prev, next);
79
+ if(!nd_poll_del(p->ndpl, pi->fd, pi))
80
+ // this is ok, if the socket is already closed
81
+ nd_log(NDLS_DAEMON, NDLP_DEBUG,
82
+ "Failed to delete socket %d from nd_poll() - called from %s() - is the socket already closed?",
83
+ pi->fd, func);
84
85
if(pi->flags & POLLINFO_FLAG_CLIENT_SOCKET) {
86
pi->del_callback(pi);
87
88
if(likely(!(pi->flags & POLLINFO_FLAG_DONT_CLOSE))) {
124
- if(close(pf->fd) == -1)
89
+ if(close(pi->fd) == -1)
90
nd_log(NDLS_DAEMON, NDLP_ERR,
91
"Failed to close() poll_events() socket %d",
127
- pf->fd);
92
+ pi->fd);
93
}
94
}
95
131
- pf->fd = -1;
132
- pf->events = 0;
133
- pf->revents = 0;
134
-
135
- pi->fd = -1;
136
- pi->socktype = -1;
137
- pi->flags = 0;
138
- pi->data = NULL;
139
-
140
- pi->del_callback = NULL;
141
- pi->rcv_callback = NULL;
142
- pi->snd_callback = NULL;
143
-
96
freez(pi->client_ip);
145
- pi->client_ip = NULL;
146
-
97
freez(pi->client_port);
148
- pi->client_port = NULL;
149
-
98
freez(pi->client_host);
151
- pi->client_host = NULL;
152
-
153
- pi->next = p->first_free;
154
- p->first_free = pi;
99
+ freez(pi);
100
101
p->used--;
157
- if(unlikely(p->max == pi->slot)) {
158
- p->max = p->min;
159
- ssize_t i;
160
- for(i = (ssize_t)pi->slot; i > (ssize_t)p->min ;i--) {
161
- if (unlikely(p->fds[i].fd != -1)) {
162
- p->max = (size_t)i;
163
- break;
164
- }
165
- }
166
- }
102
}
103
169
-void *poll_default_add_callback(POLLINFO *pi, short int *events, void *data) {
170
- (void)pi;
171
- (void)events;
172
- (void)data;
173
-
104
+void *poll_default_add_callback(POLLINFO *pi __maybe_unused, nd_poll_event_t *events __maybe_unused, void *data __maybe_unused) {
105
return NULL;
106
}
107
@@ -180,8 +111,8 @@ void poll_default_del_callback(POLLINFO *pi) {
111
"POLLFD: internal error: del_callback_default() called with data pointer - possible memory leak");
112
}
113
183
-int poll_default_rcv_callback(POLLINFO *pi, short int *events) {
184
- *events |= POLLIN;
114
+int poll_default_rcv_callback(POLLINFO *pi, nd_poll_event_t *events) {
115
+ *events |= ND_POLL_READ;
116
117
char buffer[1024 + 1];
118
@@ -208,8 +139,8 @@ int poll_default_rcv_callback(POLLINFO *pi, short int *events) {
139
return 0;
140
}
141
211
-int poll_default_snd_callback(POLLINFO *pi, short int *events) {
212
- *events &= ~POLLOUT;
142
+int poll_default_snd_callback(POLLINFO *pi, nd_poll_event_t *events) {
143
+ *events &= ~ND_POLL_WRITE;
144
145
nd_log(NDLS_DAEMON, NDLP_WARNING,
146
"POLLFD: internal error: poll_default_snd_callback(): nothing to send on socket %d",
@@ -226,16 +157,17 @@ static void poll_events_cleanup(void *pptr) {
157
POLLJOB *p = CLEANUP_FUNCTION_GET_PTR(pptr);
158
if(!p) return;
159
229
- for(size_t i = 0 ; i <= p->max ; i++) {
230
- POLLINFO *pi = &p->inf[i];
231
- poll_close_fd(pi);
160
+ while(p->ll) {
161
+ POLLINFO *pi = p->ll;
162
+ pi->flags &= ~(POLLINFO_FLAG_DONT_CLOSE);
163
+ poll_close_fd(pi, __FUNCTION__ );
164
}
165
234
- freez(p->fds);
235
- freez(p->inf);
166
+ nd_poll_destroy(p->ndpl);
167
+ p->ndpl = NULL;
168
}
169
238
-static int poll_process_error(POLLINFO *pi, struct pollfd *pf, short int revents) {
170
+static int poll_process_error(POLLINFO *pi, nd_poll_event_t revents) {
171
ND_LOG_STACK lgs[] = {
172
ND_LOG_FIELD_TXT(NDF_SRC_IP, pi->client_ip),
173
ND_LOG_FIELD_TXT(NDF_SRC_PORT, pi->client_port),
@@ -244,62 +176,50 @@ static int poll_process_error(POLLINFO *pi, struct pollfd *pf, short int revents
176
ND_LOG_STACK_PUSH(lgs);
177
178
nd_log(NDLS_DAEMON, NDLP_DEBUG,
247
- "POLLFD: LISTENER: received %s %s %s on socket at slot %zu (fd %d) client '%s' port '%s' expecting %s %s %s, having %s %s %s"
248
- , revents & POLLERR ? "POLLERR" : ""
249
- , revents & POLLHUP ? "POLLHUP" : ""
250
- , revents & POLLNVAL ? "POLLNVAL" : ""
251
- , pi->slot
179
+ "POLLFD: LISTENER: received %s %s %s on socket %d client '%s' port '%s' expecting %s %s, having %s %s"
180
+ , revents & ND_POLL_ERROR ? "ERROR" : ""
181
+ , revents & ND_POLL_HUP ? "HUP" : ""
182
+ , revents & ND_POLL_INVALID ? "INVALID" : ""
183
, pi->fd
184
, pi->client_ip ? pi->client_ip : "<undefined-ip>"
185
, pi->client_port ? pi->client_port : "<undefined-port>"
255
- , pf->events & POLLIN ? "POLLIN" : "", pf->events & POLLOUT ? "POLLOUT" : "", pf->events & POLLPRI ? "POLLPRI" : ""
256
- , revents & POLLIN ? "POLLIN" : "", revents & POLLOUT ? "POLLOUT" : "", revents & POLLPRI ? "POLLPRI" : ""
186
+ , pi->events & ND_POLL_READ ? "READ" : "", pi->events & ND_POLL_WRITE ? "WRITE" : ""
187
+ , revents & ND_POLL_READ ? "READ" : "", revents & ND_POLL_WRITE ? "WRITE" : ""
188
);
189
259
- pf->events = 0;
260
- poll_close_fd(pi);
190
+ poll_close_fd(pi, __FUNCTION__ );
191
return 1;
192
}
193
264
-static inline int poll_process_send(POLLJOB *p, POLLINFO *pi, struct pollfd *pf, time_t now) {
194
+static inline int poll_process_send(POLLINFO *pi, time_t now) {
195
pi->last_sent_t = now;
196
pi->send_count++;
197
268
- pf->events = 0;
269
-
270
- // remember the slot, in case we need to close it later
271
- // the callback may manipulate the socket list and our pf and pi pointers may be invalid after that call
272
- size_t slot = pi->slot;
198
+ pi->events = 0;
199
274
- if (unlikely(pi->snd_callback(pi, &pf->events) == -1))
275
- poll_close_fd(&p->inf[slot]);
276
-
277
- // IMPORTANT:
278
- // pf and pi may be invalid below this point, they may have been reallocated.
200
+ if (unlikely(pi->snd_callback(pi, &pi->events) == -1))
201
+ poll_close_fd(pi, __FUNCTION__ );
202
+ else
203
+ poll_process_updated_events(pi);
204
205
return 1;
206
}
207
283
-static inline int poll_process_tcp_read(POLLJOB *p, POLLINFO *pi, struct pollfd *pf, time_t now) {
208
+static inline int poll_process_tcp_read(POLLINFO *pi, time_t now) {
209
pi->last_received_t = now;
210
pi->recv_count++;
211
287
- pf->events = 0;
288
-
289
- // remember the slot, in case we need to close it later
290
- // the callback may manipulate the socket list and our pf and pi pointers may be invalid after that call
291
- size_t slot = pi->slot;
212
+ pi->events = 0;
213
293
- if (pi->rcv_callback(pi, &pf->events) == -1)
294
- poll_close_fd(&p->inf[slot]);
295
-
296
- // IMPORTANT:
297
- // pf and pi may be invalid below this point, they may have been reallocated.
214
+ if (pi->rcv_callback(pi, &pi->events) == -1)
215
+ poll_close_fd(pi, __FUNCTION__ );
216
+ else
217
+ poll_process_updated_events(pi);
218
219
return 1;
220
}
221
302
-static inline int poll_process_udp_read(POLLINFO *pi, struct pollfd *pf, time_t now __maybe_unused) {
222
+static inline int poll_process_udp_read(POLLINFO *pi, time_t now __maybe_unused) {
223
pi->last_received_t = now;
224
pi->recv_count++;
225
@@ -307,17 +227,19 @@ static inline int poll_process_udp_read(POLLINFO *pi, struct pollfd *pf, time_t
227
// but checking the access list on every UDP packet will destroy
228
// performance, especially for statsd.
229
310
- pf->events = 0;
311
- if(pi->rcv_callback(pi, &pf->events) == -1)
312
- return 0;
230
+ pi->events = 0;
231
314
- // IMPORTANT:
315
- // pf and pi may be invalid below this point, they may have been reallocated.
316
-
317
- return 1;
232
+ if(pi->rcv_callback(pi, &pi->events) == -1)
233
+ return 0;
234
+ else {
235
+ poll_process_updated_events(pi);
236
+ return 1;
237
+ }
238
}
239
320
-static int poll_process_new_tcp_connection(POLLJOB *p, POLLINFO *pi, struct pollfd *pf, time_t now) {
240
+static int poll_process_new_tcp_connection(POLLINFO *pi, time_t now) {
241
+ POLLJOB *p = pi->p;
242
+
243
pi->last_received_t = now;
244
pi->recv_count++;
245
@@ -332,7 +254,7 @@ static int poll_process_new_tcp_connection(POLLJOB *p, POLLINFO *pi, struct poll
254
#endif
255
256
int nfd = accept_socket(
335
- pf->fd, flags,
257
+ pi->fd, flags,
258
client_ip, INET6_ADDRSTRLEN, client_port,NI_MAXSERV, client_host, NI_MAXHOST,
259
p->access_list, p->allow_dns
260
);
@@ -358,6 +280,12 @@ static int poll_process_new_tcp_connection(POLLJOB *p, POLLINFO *pi, struct poll
280
"POLLFD: LISTENER: accept() failed.");
281
282
}
283
+ else if(is_socket_closed(nfd)) {
284
+ nd_log_limit_static_global_var(erl, 10, 1000);
285
+ nd_log_limit(&erl, NDLS_DAEMON, NDLP_ERR,
286
+ "POLLFD: LISTENER: received client socket %d is closed on accept(), dropping connection", nfd);
287
+ close(nfd);
288
+ }
289
else {
290
// accept ok
291
@@ -376,9 +304,6 @@ static int poll_process_new_tcp_connection(POLLJOB *p, POLLINFO *pi, struct poll
304
, NULL
305
);
306
379
- // IMPORTANT:
380
- // pf and pi may be invalid below this point, they may have been reallocated.
381
-
307
return 1;
308
}
309
@@ -386,13 +311,13 @@ static int poll_process_new_tcp_connection(POLLJOB *p, POLLINFO *pi, struct poll
311
}
312
313
void poll_events(LISTEN_SOCKETS *sockets
389
- , void *(*add_callback)(POLLINFO * /*pi*/, short int * /*events*/, void * /*data*/)
390
- , void (*del_callback)(POLLINFO * /*pi*/)
391
- , int (*rcv_callback)(POLLINFO * /*pi*/, short int * /*events*/)
392
- , int (*snd_callback)(POLLINFO * /*pi*/, short int * /*events*/)
393
- , void (*tmr_callback)(void * /*timer_data*/)
394
- , bool (*check_to_stop_callback)(void)
395
- , SIMPLE_PATTERN *access_list
314
+ , poll_events_add_callback_t add_callback
315
+ , poll_events_del_callback_t del_callback
316
+ , poll_events_rcv_callback_t rcv_callback
317
+ , poll_events_snd_callback_t snd_callback
318
+ , poll_events_tmr_callback_t tmr_callback
319
+ , bool (*check_to_stop_callback)(void)
320
+ , SIMPLE_PATTERN *access_list
321
, int allow_dns
322
, void *data
323
, time_t tcp_request_timeout_seconds
@@ -412,13 +337,9 @@ void poll_events(LISTEN_SOCKETS *sockets
337
int retval;
338
339
POLLJOB p = {
415
- .slots = 0,
340
+ .ndpl = nd_poll_create(),
341
.used = 0,
417
- .max = 0,
342
.limit = max_tcp_sockets,
419
- .fds = NULL,
420
- .inf = NULL,
421
- .first_free = NULL,
343
344
.complete_request_timeout = tcp_request_timeout_seconds,
345
.idle_timeout = tcp_idle_timeout_seconds,
@@ -496,21 +417,20 @@ void poll_events(LISTEN_SOCKETS *sockets
417
"%s listening sockets (used TCP sockets %zu, max allowed for this worker %zu)",
418
(listen_sockets_active)?"ENABLING":"DISABLING", p.used, p.limit);
419
499
- for (i = 0; i <= p.max; i++) {
500
- if(p.inf[i].flags & POLLINFO_FLAG_SERVER_SOCKET && p.inf[i].socktype == SOCK_STREAM) {
501
- p.fds[i].events = (short int) ((listen_sockets_active) ? POLLIN : 0);
420
+ for(POLLINFO *pi = p.ll; pi ; pi = pi->next) {
421
+ if((pi->flags & POLLINFO_FLAG_SERVER_SOCKET) && pi->socktype == SOCK_STREAM) {
422
+ pi->events = (short int) ((listen_sockets_active) ? ND_POLL_READ : 0);
423
+ poll_process_updated_events(pi);
424
}
425
}
426
}
427
506
- retval = poll(p.fds, p.max + 1, ND_CHECK_CANCELLABILITY_WHILE_WAITING_EVERY_MS);
428
+ nd_poll_result_t result;
429
+ retval = nd_poll_wait(p.ndpl, ND_CHECK_CANCELLABILITY_WHILE_WAITING_EVERY_MS, &result);
430
time_t now = now_boottime_sec();
431
432
if(unlikely(retval == -1)) {
510
- nd_log(NDLS_DAEMON, NDLP_ERR,
511
- "POLLFD: LISTENER: poll() failed while waiting on %zu sockets.",
512
- p.max + 1);
513
-
433
+ nd_log(NDLS_DAEMON, NDLP_ERR, "POLLFD: LISTENER: nd_poll_wait() failed.");
434
break;
435
}
436
else if(unlikely(!retval)) {
@@ -518,134 +438,71 @@ void poll_events(LISTEN_SOCKETS *sockets
438
;
439
}
440
else {
521
- POLLINFO *pi;
522
- struct pollfd *pf;
523
- size_t idx, processed = 0;
524
- short int revents;
525
-
526
- // keep fast lookup arrays per function
527
- // to avoid looping through the entire list every time
528
- size_t sends[p.max + 1], sends_max = 0;
529
- size_t reads[p.max + 1], reads_max = 0;
530
- size_t conns[p.max + 1], conns_max = 0;
531
- size_t udprd[p.max + 1], udprd_max = 0;
532
-
533
- for (i = 0; i <= p.max; i++) {
534
- pi = &p.inf[i];
535
- pf = &p.fds[i];
536
- revents = pf->revents;
537
-
538
- if(unlikely(revents == 0 || pf->fd == -1))
539
- continue;
540
-
541
- if (unlikely(revents & (POLLERR|POLLHUP|POLLNVAL))) {
542
- // something is wrong to one of our sockets
543
-
544
- pf->revents = 0;
545
- processed += poll_process_error(pi, pf, revents);
546
- }
547
- else if (likely(revents & POLLOUT)) {
548
- // a client is ready to receive data
441
+ POLLINFO *pi = result.data;
442
550
- sends[sends_max++] = i;
551
- }
552
- else if (likely(revents & (POLLIN|POLLPRI))) {
553
- if (pi->flags & POLLINFO_FLAG_CLIENT_SOCKET) {
554
- // a client sent data to us
443
+ if(result.events & (ND_POLL_HUP | ND_POLL_INVALID | ND_POLL_ERROR))
444
+ poll_process_error(pi, result.events);
445
556
- reads[reads_max++] = i;
557
- }
558
- else if (pi->flags & POLLINFO_FLAG_SERVER_SOCKET) {
559
- // something is coming to our server sockets
560
-
561
- if(pi->socktype == SOCK_DGRAM) {
562
- // UDP receive, directly on our listening socket
563
-
564
- udprd[udprd_max++] = i;
565
- }
566
- else if(pi->socktype == SOCK_STREAM) {
567
- // new TCP connection
568
-
569
- conns[conns_max++] = i;
570
- }
571
- else
572
- nd_log(NDLS_DAEMON, NDLP_ERR,
573
- "POLLFD: LISTENER: server slot %zu (fd %d) connection from %s port %s using unhandled socket type %d."
574
- , i
575
- , pi->fd
576
- , pi->client_ip ? pi->client_ip : "<undefined-ip>"
577
- , pi->client_port ? pi->client_port : "<undefined-port>"
578
- , pi->socktype
579
- );
580
- }
581
- else
446
+ else if(result.events & ND_POLL_WRITE) {
447
+ poll_process_send(pi, now);
448
+ }
449
+
450
+ else if(result.events & ND_POLL_READ) {
451
+ if (pi->flags & POLLINFO_FLAG_CLIENT_SOCKET) {
452
+ if (pi->socktype == SOCK_DGRAM)
453
+ poll_process_udp_read(pi, now);
454
+ else if (pi->socktype == SOCK_STREAM)
455
+ poll_process_tcp_read(pi, now);
456
+ else {
457
nd_log(NDLS_DAEMON, NDLP_ERR,
583
- "POLLFD: LISTENER: client slot %zu (fd %d) data from %s port %s using flags %08X is neither client nor server."
584
- , i
585
- , pi->fd
586
- , pi->client_ip ? pi->client_ip : "<undefined-ip>"
587
- , pi->client_port ? pi->client_port : "<undefined-port>"
588
- , pi->flags
589
- );
458
+ "POLLFD: LISTENER: server slot %zu (fd %d) connection from %s port %s using unhandled socket type %d.",
459
+ i,
460
+ pi->fd,
461
+ pi->client_ip ? pi->client_ip : "<undefined-ip>",
462
+ pi->client_port ? pi->client_port : "<undefined-port>",
463
+ pi->socktype);
464
+
465
+ poll_close_fd(pi, "poll_events1");
466
+ }
467
+ }
468
+ else if (pi->flags & POLLINFO_FLAG_SERVER_SOCKET) {
469
+ if(!p.limit || p.used < p.limit)
470
+ poll_process_new_tcp_connection(pi, now);
471
}
591
- else
472
+ else {
473
nd_log(NDLS_DAEMON, NDLP_ERR,
593
- "POLLFD: LISTENER: socket slot %zu (fd %d) client %s port %s unhandled event id %d."
474
+ "POLLFD: LISTENER: client slot %zu (fd %d) data from %s port %s using flags %08X is neither client nor server."
475
, i
476
, pi->fd
477
, pi->client_ip ? pi->client_ip : "<undefined-ip>"
478
, pi->client_port ? pi->client_port : "<undefined-port>"
598
- , revents
479
+ , pi->flags
480
);
600
- }
481
602
- // process sends
603
- for (idx = 0; idx < sends_max; idx++) {
604
- i = sends[idx];
605
- pi = &p.inf[i];
606
- pf = &p.fds[i];
607
- pf->revents = 0;
608
- processed += poll_process_send(&p, pi, pf, now);
609
- }
610
-
611
- // process UDP reads
612
- for (idx = 0; idx < udprd_max; idx++) {
613
- i = udprd[idx];
614
- pi = &p.inf[i];
615
- pf = &p.fds[i];
616
- pf->revents = 0;
617
- processed += poll_process_udp_read(pi, pf, now);
618
- }
619
-
620
- // process TCP reads
621
- for (idx = 0; idx < reads_max; idx++) {
622
- i = reads[idx];
623
- pi = &p.inf[i];
624
- pf = &p.fds[i];
625
- pf->revents = 0;
626
- processed += poll_process_tcp_read(&p, pi, pf, now);
627
- }
628
-
629
- if(!processed && (!p.limit || p.used < p.limit)) {
630
- // nothing processed above (rcv, snd) and we have room for another TCP connection
631
- // so, accept one TCP connection
632
- for (idx = 0; idx < conns_max; idx++) {
633
- i = conns[idx];
634
- pi = &p.inf[i];
635
- pf = &p.fds[i];
636
- pf->revents = 0;
637
- if (poll_process_new_tcp_connection(&p, pi, pf, now))
638
- break;
482
+ poll_close_fd(pi, "poll_events2");
483
}
484
}
485
+ else {
486
+ nd_log(NDLS_DAEMON, NDLP_ERR,
487
+ "POLLFD: LISTENER: socket slot %zu (fd %d) client %s port %s unhandled event id %d."
488
+ , i
489
+ , pi->fd
490
+ , pi->client_ip ? pi->client_ip : "<undefined-ip>"
491
+ , pi->client_port ? pi->client_port : "<undefined-port>"
492
+ , (int)result.events
493
+ );
494
+
495
+ poll_close_fd(pi, "poll_events3");
496
+ }
497
}
498
499
if(unlikely(p.checks_every > 0 && now - last_check > p.checks_every)) {
500
last_check = now;
501
502
// cleanup old sockets
647
- for(i = 0; i <= p.max; i++) {
648
- POLLINFO *pi = &p.inf[i];
503
+ POLLINFO *pi, *next = NULL;
504
+ for(pi = p.ll; pi ; pi = next) {
505
+ next = pi->next;
506
507
if(likely(pi->flags & POLLINFO_FLAG_CLIENT_SOCKET)) {
508
if (unlikely(pi->send_count == 0 && p.complete_request_timeout > 0 && (now - pi->connected_t) >= p.complete_request_timeout)) {
@@ -657,7 +514,7 @@ void poll_events(LISTEN_SOCKETS *sockets
514
, pi->client_port ? pi->client_port : "<undefined-port>"
515
, (size_t) p.complete_request_timeout
516
);
660
- poll_close_fd(pi);
517
+ poll_close_fd(pi, "poll_events4");
518
}
519
else if(unlikely(pi->recv_count && p.idle_timeout > 0 && now - ((pi->last_received_t > pi->last_sent_t) ? pi->last_received_t : pi->last_sent_t) >= p.idle_timeout )) {
520
nd_log(NDLS_DAEMON, NDLP_DEBUG,
@@ -668,7 +525,7 @@ void poll_events(LISTEN_SOCKETS *sockets
525
, pi->client_port ? pi->client_port : "<undefined-port>"
526
, (size_t) p.idle_timeout
527
);
671
- poll_close_fd(pi);
528
+ poll_close_fd(pi, "poll_events5");
529
}
530
}
531
}
src/libnetdata/socket/poll-events.h
+41
-42
@@ -3,15 +3,23 @@
3
#ifndef NETDATA_POLL_EVENTS_H
4
#define NETDATA_POLL_EVENTS_H
5
6
+#include "nd-poll.h"
7
+
8
#define POLLINFO_FLAG_SERVER_SOCKET 0x00000001
9
#define POLLINFO_FLAG_CLIENT_SOCKET 0x00000002
10
#define POLLINFO_FLAG_DONT_CLOSE 0x00000004
11
12
typedef struct poll POLLJOB;
13
+typedef struct pollinfo POLLINFO;
14
+
15
+typedef void *(*poll_events_add_callback_t)(POLLINFO *pi, nd_poll_event_t *events, void *data);
16
+typedef void (*poll_events_del_callback_t)(POLLINFO *pi);
17
+typedef int (*poll_events_rcv_callback_t)(POLLINFO *pi, nd_poll_event_t *events);
18
+typedef int (*poll_events_snd_callback_t)(POLLINFO *pi, nd_poll_event_t *events);
19
+typedef void (*poll_events_tmr_callback_t)(void *timer_data);
20
12
-typedef struct pollinfo {
21
+struct pollinfo {
22
POLLJOB *p; // the parent
14
- size_t slot; // the slot id
23
24
int fd; // the file descriptor
25
int socktype; // the client socket type
@@ -20,6 +28,9 @@ typedef struct pollinfo {
28
char *client_port; // Max NI_MAXSERV bytes
29
char *client_host; // Max NI_MAXHOST bytes
30
31
+ nd_poll_event_t events;
32
+ nd_poll_event_t events_we_wait_for;
33
+
34
time_t connected_t; // the time the socket connected
35
time_t last_received_t; // the time the socket last received data
36
time_t last_sent_t; // the time the socket last sent data
@@ -30,26 +41,21 @@ typedef struct pollinfo {
41
uint32_t flags; // internal flags
42
43
// callbacks for this socket
33
- void (*del_callback)(struct pollinfo *pi);
34
- int (*rcv_callback)(struct pollinfo *pi, short int *events);
35
- int (*snd_callback)(struct pollinfo *pi, short int *events);
44
+ poll_events_del_callback_t del_callback;
45
+ poll_events_rcv_callback_t rcv_callback;
46
+ poll_events_snd_callback_t snd_callback;
47
48
// the user data
49
void *data;
50
40
- // linking of free pollinfo structures
41
- // for quickly finding the next available
42
- // this is like a stack, it grows and shrinks
43
- // (with gaps - lower empty slots are preferred)
44
- struct pollinfo *next;
45
-} POLLINFO;
51
+ struct pollinfo *prev, *next;
52
+};
53
54
struct poll {
48
- size_t slots;
49
- size_t used;
50
- size_t min;
51
- size_t max;
55
+ nd_poll_t *ndpl;
56
+ POLLINFO *ll;
57
58
+ size_t used;
59
size_t limit;
60
61
time_t complete_request_timeout;
@@ -59,26 +65,20 @@ struct poll {
65
time_t timer_milliseconds;
66
void *timer_data;
67
62
- struct pollfd *fds;
63
- struct pollinfo *inf;
64
- struct pollinfo *first_free;
65
-
68
SIMPLE_PATTERN *access_list;
69
int allow_dns;
70
69
- void *(*add_callback)(POLLINFO *pi, short int *events, void *data);
70
- void (*del_callback)(POLLINFO *pi);
71
- int (*rcv_callback)(POLLINFO *pi, short int *events);
72
- int (*snd_callback)(POLLINFO *pi, short int *events);
73
- void (*tmr_callback)(void *timer_data);
71
+ poll_events_add_callback_t add_callback;
72
+ poll_events_del_callback_t del_callback;
73
+ poll_events_rcv_callback_t rcv_callback;
74
+ poll_events_snd_callback_t snd_callback;
75
+ poll_events_tmr_callback_t tmr_callback;
76
};
77
76
-#define pollinfo_from_slot(p, slot) (&((p)->inf[(slot)]))
77
-
78
-int poll_default_snd_callback(POLLINFO *pi, short int *events);
79
-int poll_default_rcv_callback(POLLINFO *pi, short int *events);
78
+int poll_default_snd_callback(POLLINFO *pi, nd_poll_event_t *events);
79
+int poll_default_rcv_callback(POLLINFO *pi, nd_poll_event_t *events);
80
void poll_default_del_callback(POLLINFO *pi);
81
-void *poll_default_add_callback(POLLINFO *pi, short int *events, void *data);
81
+void *poll_default_add_callback(POLLINFO *pi, nd_poll_event_t *events, void *data);
82
83
POLLINFO *poll_add_fd(POLLJOB *p
84
, int fd
@@ -88,22 +88,21 @@ POLLINFO *poll_add_fd(POLLJOB *p
88
, const char *client_ip
89
, const char *client_port
90
, const char *client_host
91
- , void *(*add_callback)(POLLINFO *pi, short int *events, void *data)
92
- , void (*del_callback)(POLLINFO *pi)
93
- , int (*rcv_callback)(POLLINFO *pi, short int *events)
94
- , int (*snd_callback)(POLLINFO *pi, short int *events)
95
- , void *data
91
+ , poll_events_add_callback_t add_callback
92
+ , poll_events_del_callback_t del_callback
93
+ , poll_events_rcv_callback_t rcv_callback
94
+ , poll_events_snd_callback_t snd_callback
95
+ , void *data
96
);
97
-void poll_close_fd(POLLINFO *pi);
97
98
void poll_events(LISTEN_SOCKETS *sockets
100
- , void *(*add_callback)(POLLINFO *pi, short int *events, void *data)
101
- , void (*del_callback)(POLLINFO *pi)
102
- , int (*rcv_callback)(POLLINFO *pi, short int *events)
103
- , int (*snd_callback)(POLLINFO *pi, short int *events)
104
- , void (*tmr_callback)(void *timer_data)
105
- , bool (*check_to_stop_callback)(void)
106
- , SIMPLE_PATTERN *access_list
99
+ , poll_events_add_callback_t add_callback
100
+ , poll_events_del_callback_t del_callback
101
+ , poll_events_rcv_callback_t rcv_callback
102
+ , poll_events_snd_callback_t snd_callback
103
+ , poll_events_tmr_callback_t tmr_callback
104
+ , bool (*check_to_stop_callback)(void)
105
+ , SIMPLE_PATTERN *access_list
106
, int allow_dns
107
, void *data
108
, time_t tcp_request_timeout_seconds
src/libnetdata/socket/socket.c
+212
-67
@@ -48,12 +48,6 @@ bool ip_to_hostname(const char *ip, char *dst, size_t dst_len) {
48
// --------------------------------------------------------------------------------------------------------------------
49
// various library calls
50
51
-#ifdef __gnu_linux__
52
-#define LARGE_SOCK_SIZE 33554431 // don't ask why - I found it at brubeck source - I guess it is just a large number
53
-#else
54
-#define LARGE_SOCK_SIZE 4096
55
-#endif
56
-
51
bool fd_is_socket(int fd) {
52
int type;
53
socklen_t len = sizeof(type);
@@ -63,7 +57,7 @@ bool fd_is_socket(int fd) {
57
return true;
58
}
59
66
-#ifdef POLLRDHUP
60
+#if defined(POLLRDHUP) && 0 // ktsaou: disabled because the recv() method is faster (1 syscall vs multiple by poll())
61
bool is_socket_closed(int fd) {
62
if(fd < 0)
63
return true;
@@ -114,94 +108,245 @@ bool is_socket_closed(int fd) {
108
}
109
#endif
110
117
-int sock_setnonblock(int fd) {
118
- int flags;
111
+#if defined(OS_LINUX)
112
+// Valid from: 4 KB to 64 MB (typical range)
113
+// Default is usually: 128 KB to 256 KB
114
+// Maximum is controlled by: /proc/sys/net/core/rmem_max and /proc/sys/net/core/wmem_max
115
+// Interactive applications should use: 256 KB
116
+// High-performance applications should use: 8 MB to 64 MB
117
+#define LARGE_SOCK_SIZE (32 * 1024 * 1024)
118
+
119
+#elif defined(OS_FREEBSD)
120
+// Valid from: 4 KB to 16 MB (typical range)
121
+// Default is usually: 64 KB to 256 KB
122
+// Maximum is controlled by: kern.ipc.maxsockbuf
123
+// Interactive applications should use: 128 KB to 256 KB
124
+// High-performance applications should use: 2 MB to 16 MB
125
+#define LARGE_SOCK_SIZE (8 * 1024 * 1024)
126
+
127
+#elif defined(OS_MACOS)
128
+// Valid from: 4 KB to 8 MB (typical range)
129
+// Default is usually: 128 KB
130
+// Maximum is controlled by: net.inet.tcp.sendspace and net.inet.tcp.recvspace
131
+// Interactive applications should use: 128 KB
132
+// High-performance applications should use: 1 MB to 8 MB
133
+#define LARGE_SOCK_SIZE (4 * 1024 * 1024)
134
+
135
+#elif defined(OS_WINDOWS)
136
+// Valid from: 8 KB to 16 MB (typical range)
137
+// Default is usually: 8 KB to 64 KB
138
+// Maximum is controlled by: registry keys such as TcpWindowSize
139
+// Interactive applications should use: 64 KB to 128 KB
140
+// High-performance applications should use: 1 MB to 16 MB
141
+#define LARGE_SOCK_SIZE (8 * 1024 * 1024)
142
120
- flags = fcntl(fd, F_GETFL);
121
- flags |= O_NONBLOCK;
143
+#else
144
+// Valid from: 4 KB to platform-dependent maximum
145
+// Default is usually: 64 KB to 256 KB
146
+// Interactive applications should use: 128 KB to 256 KB
147
+// High-performance applications should use: 1 MB to platform-dependent maximum
148
+#define LARGE_SOCK_SIZE (1 * 1024 * 1024)
149
+#endif
150
123
- int ret = fcntl(fd, F_SETFL, flags);
124
- if(ret < 0)
125
- nd_log(NDLS_DAEMON, NDLP_ERR,
126
- "Failed to set O_NONBLOCK on socket %d",
127
- fd);
151
+// Returns -1 for errors, current buffer size if successful
152
+int sock_enlarge_rcv_buf(int fd) {
153
+ int ret = -1;
154
+ int bs = LARGE_SOCK_SIZE;
155
+ int current_bs = 0;
156
+ socklen_t optlen = sizeof(current_bs);
157
+
158
+ // Get the current receive buffer size
159
+ if (getsockopt(fd, SOL_SOCKET, SO_RCVBUF, ¤t_bs, &optlen) == 0) {
160
+ // Set the buffer size only if it's smaller than the desired size
161
+ if (current_bs < bs) {
162
+ setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &bs, sizeof(bs));
163
+
164
+ // Re-check the buffer size after attempting to set it
165
+ if (getsockopt(fd, SOL_SOCKET, SO_RCVBUF, ¤t_bs, &optlen) == 0)
166
+ ret = current_bs;
167
+ } else {
168
+ // Current buffer size is already large enough
169
+ ret = current_bs;
170
+ }
171
+ }
172
173
return ret;
174
}
175
132
-int sock_delnonblock(int fd) {
133
- int flags;
176
+// Returns -1 for errors, current buffer size if successful
177
+int sock_enlarge_snd_buf(int fd) {
178
+ int ret = -1;
179
+ int bs = LARGE_SOCK_SIZE;
180
+ int current_bs = 0;
181
+ socklen_t optlen = sizeof(current_bs);
182
+
183
+ // Get the current send buffer size
184
+ if (getsockopt(fd, SOL_SOCKET, SO_SNDBUF, ¤t_bs, &optlen) == 0) {
185
+ // Set the buffer size only if it's smaller than the desired size
186
+ if (current_bs < bs) {
187
+ setsockopt(fd, SOL_SOCKET, SO_SNDBUF, &bs, sizeof(bs));
188
+
189
+ // Re-check the buffer size after attempting to set it
190
+ if (getsockopt(fd, SOL_SOCKET, SO_SNDBUF, ¤t_bs, &optlen) == 0)
191
+ ret = current_bs;
192
+ } else {
193
+ // Current buffer size is already large enough
194
+ ret = current_bs;
195
+ }
196
+ }
197
135
- flags = fcntl(fd, F_GETFL);
136
- flags &= ~O_NONBLOCK;
198
+ return ret;
199
+}
200
138
- int ret = fcntl(fd, F_SETFL, flags);
139
- if(ret < 0)
140
- nd_log(NDLS_DAEMON, NDLP_ERR,
141
- "Failed to remove O_NONBLOCK on socket %d",
142
- fd);
201
+// returns -1 for errors, 0 if cork is unset, 1 if cork is set
202
+int sock_setcork(int fd __maybe_unused, bool cork __maybe_unused) {
203
+ int rc = -1;
204
144
- return ret;
205
+#ifdef TCP_CORK
206
+ int tcp_cork = (cork) ? 1 : 0;
207
+ socklen_t optlen = sizeof(tcp_cork);
208
+
209
+ if(setsockopt(fd, IPPROTO_TCP, TCP_CORK, &tcp_cork, optlen) == 0) {
210
+ // setting was successful, return the intended state
211
+ rc = cork ? 1 : 0;
212
+ }
213
+ else if(getsockopt(fd, IPPROTO_TCP, TCP_CORK, &tcp_cork, &optlen) == 0) {
214
+ // return the current state since retrieval is successful
215
+ rc = tcp_cork ? 1 : 0;
216
+ }
217
+#endif
218
+
219
+ return rc;
220
}
221
147
-int sock_setreuse(int fd, int reuse) {
148
- int ret = setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &reuse, sizeof(reuse));
222
+// Returns -1 for errors, 0 if O_NONBLOCK is unset, 1 if O_NONBLOCK is set
223
+int sock_setnonblock(int fd, bool nonblock) {
224
+ int rc = -1;
225
+ int flags = fcntl(fd, F_GETFL);
226
150
- if(ret == -1)
151
- nd_log(NDLS_DAEMON, NDLP_ERR,
152
- "Failed to set SO_REUSEADDR on socket %d",
153
- fd);
227
+ if (flags < 0) {
228
+ // Failed to get current flags
229
+ return -1;
230
+ }
231
155
- return ret;
232
+ int new_flags = nonblock ? (flags | O_NONBLOCK) : (flags & ~O_NONBLOCK);
233
+
234
+ if (fcntl(fd, F_SETFL, new_flags) == 0) {
235
+ // Setting was successful, return the intended state
236
+ rc = nonblock ? 1 : 0;
237
+ } else {
238
+ // If setting failed, return the current state
239
+ flags = fcntl(fd, F_GETFL);
240
+ if (flags >= 0)
241
+ rc = (flags & O_NONBLOCK) ? 1 : 0;
242
+ }
243
+
244
+ return rc;
245
}
246
158
-void sock_setcloexec(int fd)
159
-{
160
- UNUSED(fd);
161
- int flags = fcntl(fd, F_GETFD);
162
- if (flags != -1)
163
- (void) fcntl(fd, F_SETFD, flags | FD_CLOEXEC);
247
+// Returns -1 for errors, 0 if SO_REUSEADDR is unset, 1 if SO_REUSEADDR is set
248
+int sock_setreuse_addr(int fd, bool reuse) {
249
+ int rc = -1;
250
+ int reuse_val = reuse ? 1 : 0;
251
+ socklen_t optlen = sizeof(reuse_val);
252
+
253
+ // Attempt to set SO_REUSEADDR
254
+ if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &reuse_val, optlen) == 0) {
255
+ // Setting was successful, return the intended state
256
+ rc = reuse ? 1 : 0;
257
+ } else {
258
+ // If setting failed, attempt to retrieve the current state
259
+ if (getsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &reuse_val, &optlen) == 0) {
260
+ // Return the current state
261
+ rc = reuse_val ? 1 : 0;
262
+ }
263
+ }
264
+
265
+ return rc;
266
}
267
166
-int sock_setreuse_port(int fd __maybe_unused, int reuse __maybe_unused) {
167
- int ret;
268
+// Returns -1 for errors, 0 if SO_REUSEPORT is unset, 1 if SO_REUSEPORT is set
269
+int sock_setreuse_port(int fd __maybe_unused, bool reuse __maybe_unused) {
270
+ int rc = -1;
271
272
#ifdef SO_REUSEPORT
170
- ret = setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, &reuse, sizeof(reuse));
171
- if(ret == -1 && errno != ENOPROTOOPT)
172
- nd_log(NDLS_DAEMON, NDLP_ERR,
173
- "failed to set SO_REUSEPORT on socket %d",
174
- fd);
273
+ int reuse_val = reuse ? 1 : 0;
274
+ socklen_t optlen = sizeof(reuse_val);
275
+
276
+ // Attempt to set SO_REUSEPORT
277
+ if (setsockopt(fd, SOL_SOCKET, SO_REUSEPORT, &reuse_val, optlen) == 0) {
278
+ // Setting was successful, return the intended state
279
+ rc = reuse ? 1 : 0;
280
+ } else if (errno != ENOPROTOOPT) {
281
+ // If setting failed for a reason other than unsupported option, check the current state
282
+ if (getsockopt(fd, SOL_SOCKET, SO_REUSEPORT, &reuse_val, &optlen) == 0) {
283
+ // Return the current state
284
+ rc = reuse_val ? 1 : 0;
285
+ }
286
+ }
287
#else
176
- ret = -1;
288
+ // SO_REUSEPORT is not supported
289
+ errno = ENOPROTOOPT;
290
#endif
291
179
- return ret;
292
+ return rc;
293
}
294
182
-int sock_enlarge_in(int fd) {
183
- int ret, bs = LARGE_SOCK_SIZE;
184
-
185
- ret = setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &bs, sizeof(bs));
295
+// Returns -1 for errors, 0 if FD_CLOEXEC is unset, 1 if FD_CLOEXEC is set
296
+int sock_setcloexec(int fd, bool cloexec) {
297
+ int rc = -1;
298
187
- if(ret == -1)
188
- nd_log(NDLS_DAEMON, NDLP_ERR,
189
- "Failed to set SO_RCVBUF on socket %d",
190
- fd);
299
+ // Get current file descriptor flags
300
+ int flags = fcntl(fd, F_GETFD);
301
+ if (flags == -1)
302
+ return -1; // Error retrieving flags
303
+
304
+ int new_flags = cloexec ? (flags | FD_CLOEXEC) : (flags & ~FD_CLOEXEC);
305
+
306
+ // Set the FD_CLOEXEC flag as requested
307
+ if (fcntl(fd, F_SETFD, new_flags) == 0) {
308
+ // Setting was successful, return the intended state
309
+ rc = cloexec ? 1 : 0;
310
+ } else {
311
+ // If setting failed, return the current state
312
+ flags = fcntl(fd, F_GETFD);
313
+ if (flags != -1) {
314
+ rc = (flags & FD_CLOEXEC) ? 1 : 0;
315
+ }
316
+ }
317
192
- return ret;
318
+ return rc;
319
}
320
195
-int sock_enlarge_out(int fd) {
196
- int ret, bs = LARGE_SOCK_SIZE;
197
- ret = setsockopt(fd, SOL_SOCKET, SO_SNDBUF, &bs, sizeof(bs));
198
-
199
- if(ret == -1)
200
- nd_log(NDLS_DAEMON, NDLP_ERR,
201
- "Failed to set SO_SNDBUF on socket %d",
202
- fd);
321
+// Returns -1 for errors, 0 if TCP_DEFER_ACCEPT is unset, 1 if TCP_DEFER_ACCEPT is set
322
+int sock_set_tcp_defer_accept(int fd __maybe_unused, bool defer __maybe_unused) {
323
+#ifdef TCP_DEFER_ACCEPT
324
+ // Check if the file descriptor is a socket
325
+ if (!fd_is_socket(fd))
326
+ return 0; // Not a socket
327
+
328
+ int rc = -1;
329
+ int timeout = defer ? 5 : 0; // Set timeout to 5 seconds for enabling, 0 to disable
330
+ socklen_t optlen = sizeof(timeout);
331
+
332
+ // Attempt to set TCP_DEFER_ACCEPT
333
+ if (setsockopt(fd, IPPROTO_TCP, TCP_DEFER_ACCEPT, &timeout, optlen) == 0) {
334
+ // Setting was successful, return the intended state
335
+ rc = defer ? 1 : 0;
336
+ } else if (errno != EINVAL && errno != ENOPROTOOPT) {
337
+ // If setting failed and it's not because of invalid option or unsupported protocol
338
+ // Check the current state
339
+ if (getsockopt(fd, IPPROTO_TCP, TCP_DEFER_ACCEPT, &timeout, &optlen) == 0) {
340
+ rc = timeout > 0 ? 1 : 0;
341
+ }
342
+ }
343
204
- return ret;
344
+ return rc;
345
+#else
346
+ // TCP_DEFER_ACCEPT not supported
347
+ errno = ENOPROTOOPT;
348
+ return -1;
349
+#endif
350
}
351
352
// --------------------------------------------------------------------------------------------------------------------
@@ -437,7 +582,7 @@ int accept_socket(int fd, int flags, char *client_ip, size_t ipsize, char *clien
582
if (!strcmp(client_ip, "127.0.0.1") || !strcmp(client_ip, "::1")) {
583
strncpyz(client_ip, "localhost", ipsize);
584
}
440
- sock_setcloexec(nfd);
585
+ sock_setcloexec(nfd, true);
586
587
#ifdef __FreeBSD__
588
if(((struct sockaddr *)&sadr)->sa_family == AF_LOCAL)
src/libnetdata/socket/socket.h
+23
-7
@@ -13,13 +13,29 @@ int wait_on_socket_or_cancel_with_timeout(NETDATA_SSL *ssl, int fd, int timeout_
13
bool fd_is_socket(int fd);
14
bool is_socket_closed(int fd);
15
16
-int sock_setnonblock(int fd);
17
-int sock_delnonblock(int fd);
18
-int sock_setreuse(int fd, int reuse);
19
-void sock_setcloexec(int fd);
20
-int sock_setreuse_port(int fd, int reuse);
21
-int sock_enlarge_in(int fd);
22
-int sock_enlarge_out(int fd);
16
+// Returns -1 for errors, 0 if TCP_DEFER_ACCEPT is unset, 1 if TCP_DEFER_ACCEPT is set
17
+int sock_set_tcp_defer_accept(int fd, bool defer);
18
+
19
+// Returns -1 for errors, 0 if FD_CLOEXEC is unset, 1 if FD_CLOEXEC is set
20
+int sock_setcloexec(int fd, bool cloexec);
21
+
22
+// Returns -1 for errors, 0 if O_NONBLOCK is unset, 1 if O_NONBLOCK is set
23
+int sock_setnonblock(int fd, bool nonblock);
24
+
25
+// Returns -1 for errors, 0 if SO_REUSEADDR is unset, 1 if SO_REUSEADDR is set
26
+int sock_setreuse_addr(int fd, bool reuse);
27
+
28
+// Returns -1 for errors, 0 if SO_REUSEPORT is unset, 1 if SO_REUSEPORT is set
29
+int sock_setreuse_port(int fd, bool reuse);
30
+
31
+// Returns -1 for errors, current buffer size if successful
32
+int sock_enlarge_rcv_buf(int fd);
33
+
34
+// Returns -1 for errors, current buffer size if successful
35
+int sock_enlarge_snd_buf(int fd);
36
+
37
+// returns -1 for errors, 0 if cork is unset, 1 if cork is set
38
+int sock_setcork(int fd, bool cork);
39
40
int connection_allowed(int fd, char *client_ip, char *client_host, size_t hostsize,
41
SIMPLE_PATTERN *access_list, const char *patname, int allow_dns);
src/libnetdata/spawn_server/log-forwarder.c
+3
-2
@@ -49,7 +49,7 @@ static inline void log_forwarder_wake_up_worker(LOG_FORWARDER *lf) {
49
char ch = 0;
50
ssize_t bytes_written = write(lf->pipe_fds[PIPE_WRITE], &ch, 1);
51
if (bytes_written != 1)
52
- nd_log(NDLS_COLLECTORS, NDLP_ERR, "Failed to write to notification pipe");
52
+ nd_log(NDLS_COLLECTORS, NDLP_ERR, "Log forwarder: Failed to write to notification pipe");
53
}
54
55
// --------------------------------------------------------------------------------------------------------------------
@@ -65,7 +65,8 @@ LOG_FORWARDER *log_forwarder_start(void) {
65
}
66
67
// make sure read() will not block on this pipe
68
- sock_setnonblock(lf->pipe_fds[PIPE_READ]);
68
+ if(sock_setnonblock(lf->pipe_fds[PIPE_READ], true) != 1)
69
+ nd_log(NDLS_COLLECTORS, NDLP_ERR, "Log forwarder: Failed to set non-blocking mode");
70
71
lf->running = true;
72
lf->thread = nd_thread_create("log-fw", NETDATA_THREAD_OPTION_JOINABLE, log_forwarder_thread_func, lf);
src/libnetdata/spawn_server/spawn_server_nofork.c
+4
-4
@@ -662,8 +662,8 @@ static void spawn_server_receive_request(int sock, SPAWN_SERVER *server) {
662
struct cmsghdr *cmsg = CMSG_FIRSTHDR(&msg);
663
if (cmsg == NULL || cmsg->cmsg_len != CMSG_LEN(sizeof(int) * SPAWN_SERVER_TRANSFER_FDS)) {
664
nd_log(NDLS_COLLECTORS, NDLP_ERR,
665
- "SPAWN SERVER: Received invalid control message (expected %zu bytes, received %zu bytes)",
666
- CMSG_LEN(sizeof(int) * SPAWN_SERVER_TRANSFER_FDS), cmsg?cmsg->cmsg_len:0);
665
+ "SPAWN SERVER: Received invalid control message (expected %zu bytes, received %zu bytes)",
666
+ (size_t)(CMSG_LEN(sizeof(int) * SPAWN_SERVER_TRANSFER_FDS)), (size_t)(cmsg?cmsg->cmsg_len:0));
667
close(sock);
668
return;
669
}
@@ -908,7 +908,7 @@ static int spawn_server_event_loop(SPAWN_SERVER *server) {
908
}
909
910
// do not fork this socket
911
- sock_setcloexec(sock);
911
+ sock_setcloexec(sock, true);
912
913
// receive the request and process it
914
spawn_server_receive_request(sock, server);
@@ -1091,7 +1091,7 @@ SPAWN_SERVER* spawn_server_create(SPAWN_SERVER_OPTIONS options, const char *name
1091
};
1092
os_close_all_non_std_open_fds_except(fds_to_keep, _countof(fds_to_keep), 0);
1093
nd_log_reopen_log_files_for_spawn_server(buf);
1094
- exit(spawn_server_event_loop(server));
1094
+ _exit(spawn_server_event_loop(server));
1095
}
1096
else if (pid > 0) {
1097
// the parent
src/libnetdata/storage_number/storage_number.c
+10
-10
@@ -3,7 +3,7 @@
3
#include "../libnetdata.h"
4
5
bool is_system_ieee754_double(void) {
6
- static bool logged = false;
6
+// static bool logged = false;
7
8
struct {
9
NETDATA_DOUBLE original;
@@ -51,25 +51,25 @@ bool is_system_ieee754_double(void) {
51
uint64_t *ptr = (uint64_t *)&tests[i].original;
52
53
if(*ptr != tests[i].i && (tests[i].original == tests[i].d || (isnan(tests[i].original) && isnan(tests[i].d)))) {
54
- if(!logged)
55
- netdata_log_info("IEEE754: test #%zu, value " NETDATA_DOUBLE_FORMAT_G " is represented in this system as %016llX, but it was expected as %016llX",
56
- i+1, tests[i].original, (long long unsigned int)*ptr, (long long unsigned int)tests[i].i);
54
+// if(!logged)
55
+// netdata_log_info("IEEE754: test #%zu, value " NETDATA_DOUBLE_FORMAT_G " is represented in this system as %016llX, but it was expected as %016llX",
56
+// i+1, tests[i].original, (long long unsigned int)*ptr, (long long unsigned int)tests[i].i);
57
errors++;
58
}
59
}
60
61
if(!errors && sizeof(NETDATA_DOUBLE) == sizeof(uint64_t)) {
62
- if(!logged)
63
- netdata_log_info("IEEE754: system is using IEEE754 DOUBLE PRECISION values");
62
+// if(!logged)
63
+// netdata_log_info("IEEE754: system is using IEEE754 DOUBLE PRECISION values");
64
65
- logged = true;
65
+// logged = true;
66
return true;
67
}
68
else {
69
- if(!logged)
70
- netdata_log_info("IEEE754: system is NOT compatible with IEEE754 DOUBLE PRECISION values");
69
+// if(!logged)
70
+// netdata_log_info("IEEE754: system is NOT compatible with IEEE754 DOUBLE PRECISION values");
71
72
- logged = true;
72
+// logged = true;
73
return false;
74
}
75
}
src/libnetdata/string/string.c
+19
-14
@@ -32,7 +32,8 @@ static struct string_partition {
32
size_t deletes; // the number of successful deleted from the index
33
34
long int entries; // the number of entries in the index
35
- long int memory; // the memory used, with JudyHS (accurate)
35
+ long int memory; // the memory used
36
+ long int memory_index; // JudyHS (accurate)
37
38
#ifdef NETDATA_INTERNAL_CHECKS
39
// internal statistics
@@ -63,13 +64,14 @@ static struct string_partition {
64
#define string_internal_stats_add(partition, var, val) do {;} while(0)
65
#endif
66
66
-void string_statistics(size_t *inserts, size_t *deletes, size_t *searches, size_t *entries, size_t *references, size_t *memory, size_t *duplications, size_t *releases) {
67
+void string_statistics(size_t *inserts, size_t *deletes, size_t *searches, size_t *entries, size_t *references, size_t *memory, size_t *memory_index, size_t *duplications, size_t *releases) {
68
if (inserts) *inserts = 0;
69
if (deletes) *deletes = 0;
70
if (searches) *searches = 0;
71
if (entries) *entries = 0;
72
if (references) *references = 0;
73
if (memory) *memory = 0;
74
+ if (memory_index) *memory_index = 0;
75
if (duplications) *duplications = 0;
76
if (releases) *releases = 0;
77
@@ -78,6 +80,7 @@ void string_statistics(size_t *inserts, size_t *deletes, size_t *searches, size_
80
if (deletes) *deletes += string_base[i].deletes;
81
if (entries) *entries += (size_t) string_base[i].entries;
82
if (memory) *memory += (size_t) string_base[i].memory;
83
+ if (memory_index) *memory_index += (string_base[i].memory_index > 0) ? string_base[i].memory_index : 0;
84
85
#ifdef NETDATA_INTERNAL_CHECKS
86
if (searches) *searches += string_base[i].atomic.searches;
@@ -196,7 +199,7 @@ static inline STRING *string_index_insert(const char *str, size_t length) {
199
200
rw_spinlock_write_lock(&string_base[partition].spinlock);
201
199
- int64_t mem = 0;
202
+ int64_t judy_mem = 0;
203
204
STRING **ptr;
205
{
@@ -206,7 +209,7 @@ static inline STRING *string_index_insert(const char *str, size_t length) {
209
210
Pvoid_t *Rc = JudyHSIns(&string_base[partition].JudyHSArray, (void *)str, length - 1, &J_Error);
211
209
- mem = JudyAllocThreadPulseGetAndReset();
212
+ judy_mem = JudyAllocThreadPulseGetAndReset();
213
214
if (unlikely(Rc == PJERR)) {
215
fatal(
@@ -220,7 +223,7 @@ static inline STRING *string_index_insert(const char *str, size_t length) {
223
224
if (likely(*ptr == 0)) {
225
// a new item added to the index
223
- size_t mem_size = sizeof(STRING) + length;
226
+ long mem_size = (long)sizeof(STRING) + (long)length;
227
string = mallocz(mem_size);
228
strcpy((char *)string->str, str);
229
string->length = length;
@@ -228,7 +231,8 @@ static inline STRING *string_index_insert(const char *str, size_t length) {
231
*ptr = string;
232
string_base[partition].inserts++;
233
string_base[partition].entries++;
231
- string_base[partition].memory += (long)(mem_size + mem);
234
+ string_base[partition].memory += mem_size;
235
+ string_base[partition].memory_index += judy_mem;
236
}
237
else {
238
// the item is already in the index
@@ -264,7 +268,7 @@ static inline void string_index_delete(STRING *string) {
268
#endif
269
270
bool deleted = false;
267
- int64_t mem = 0;
271
+ int64_t judy_mem = 0;
272
273
if (likely(string_base[partition].JudyHSArray)) {
274
JError_t J_Error;
@@ -273,7 +277,7 @@ static inline void string_index_delete(STRING *string) {
277
278
int ret = JudyHSDel(&string_base[partition].JudyHSArray, (void *)string->str, string->length - 1, &J_Error);
279
276
- mem = JudyAllocThreadPulseGetAndReset();
280
+ judy_mem = JudyAllocThreadPulseGetAndReset();
281
282
if (unlikely(ret == JERR)) {
283
netdata_log_error(
@@ -288,10 +292,11 @@ static inline void string_index_delete(STRING *string) {
292
if (unlikely(!deleted))
293
netdata_log_error("STRING: tried to delete '%s' that is not in the index. Ignoring it.", string->str);
294
else {
291
- size_t mem_size = sizeof(STRING) + string->length;
295
+ long mem_size = (long)sizeof(STRING) + (long)string->length;
296
string_base[partition].deletes++;
297
string_base[partition].entries--;
294
- string_base[partition].memory -= (long)(mem_size + mem);
298
+ string_base[partition].memory -= mem_size;
299
+ string_base[partition].memory_index += judy_mem;
300
freez(string);
301
}
302
@@ -683,8 +688,8 @@ int string_unittest(size_t entries) {
688
ospins = unittest_string_spins();
689
#endif
690
686
- size_t oinserts, odeletes, osearches, oentries, oreferences, omemory, oduplications, oreleases;
687
- string_statistics(&oinserts, &odeletes, &osearches, &oentries, &oreferences, &omemory, &oduplications, &oreleases);
691
+ size_t oinserts, odeletes, osearches, oentries, oreferences, omemory, omemory_index, oduplications, oreleases;
692
+ string_statistics(&oinserts, &odeletes, &osearches, &oentries, &oreferences, &omemory, &omemory_index, &oduplications, &oreleases);
693
694
time_t seconds_to_run = 5;
695
int threads_to_create = 2;
@@ -707,8 +712,8 @@ int string_unittest(size_t entries) {
712
for (int i = 0; i < threads_to_create; i++)
713
nd_thread_join(threads[i]);
714
710
- size_t inserts, deletes, searches, sentries, references, memory, duplications, releases;
711
- string_statistics(&inserts, &deletes, &searches, &sentries, &references, &memory, &duplications, &releases);
715
+ size_t inserts, deletes, searches, sentries, references, memory, memory_index, duplications, releases;
716
+ string_statistics(&inserts, &deletes, &searches, &sentries, &references, &memory, &memory_index, &duplications, &releases);
717
718
fprintf(stderr, "inserts %zu, deletes %zu, searches %zu, entries %zu, references %zu, memory %zu, duplications %zu, releases %zu\n",
719
inserts - oinserts, deletes - odeletes, searches - osearches, sentries - oentries, references - oreferences, memory - omemory, duplications - oduplications, releases - oreleases);
src/libnetdata/string/string.h
+1
-1
@@ -36,7 +36,7 @@ static inline int string_strncmp(STRING *string, const char *s, size_t n) {
36
return strncmp(string2str(string), s, n);
37
}
38
39
-void string_statistics(size_t *inserts, size_t *deletes, size_t *searches, size_t *entries, size_t *references, size_t *memory, size_t *duplications, size_t *releases);
39
+void string_statistics(size_t *inserts, size_t *deletes, size_t *searches, size_t *entries, size_t *references, size_t *memory, size_t *memory_index, size_t *duplications, size_t *releases);
40
41
int string_unittest(size_t entries);
42
src/libnetdata/template-enum.h
+16
-6
@@ -39,10 +39,11 @@
39
40
// --------------------------------------------------------------------------------------------------------------------
41
42
-#define BITMAP_STR_DEFINE_FUNCTIONS_EXTERN(type) \
43
- type type ## _2id_one(const char *str); \
44
- const char *type##_2str_one(type id); \
45
- const char *type##_2json(BUFFER *wb, const char *key, type id);
42
+#define BITMAP_STR_DEFINE_FUNCTIONS_EXTERN(type) \
43
+ type type ## _2id_one(const char *str); \
44
+ const char *type##_2str_one(type id); \
45
+ void type##_2json(BUFFER *wb, const char *key, type id); \
46
+ void type##_2buffer(BUFFER *wb, type id, const char *separator);
47
48
#define BITMAP_STR_DEFINE_FUNCTIONS(type, def, def_str) \
49
type type##_2id_one(const char *str) \
@@ -68,7 +69,7 @@
69
return def_str; \
70
} \
71
\
71
- const char *type##_2json(BUFFER *wb, const char *key, type id) \
72
+ void type##_2json(BUFFER *wb, const char *key, type id) \
73
{ \
74
buffer_json_member_add_array(wb, key); \
75
for (size_t i = 0; type ## _names[i].name; i++) { \
@@ -76,8 +77,17 @@
77
buffer_json_add_array_item_string(wb, type ## _names[i].name); \
78
} \
79
buffer_json_array_close(wb); \
80
+ } \
81
\
80
- return def_str; \
82
+ void type##_2buffer(BUFFER *wb, type id, const char *separator) \
83
+ { \
84
+ size_t added = 0; \
85
+ for (size_t i = 0; type ## _names[i].name; i++) { \
86
+ if ((id & type ## _names[i].id) == type ## _names[i].id) { \
87
+ if(added++) buffer_strcat(wb, separator); \
88
+ buffer_strcat(wb, type##_names[i].name); \
89
+ } \
90
+ } \
91
}
92
93
#endif //NETDATA_TEMPLATE_ENUM_H
src/libnetdata/threads/threads.c
+2
-2
@@ -230,7 +230,7 @@ void netdata_threads_init_for_external_plugins(size_t stacksize) {
230
// ----------------------------------------------------------------------------
231
232
void rrdset_thread_rda_free(void);
233
-void sender_commit_thread_buffer_free(void);
233
+void sender_thread_buffer_free(void);
234
void query_target_free(void);
235
void service_exits(void);
236
void rrd_collector_finished(void);
@@ -299,7 +299,7 @@ static void nd_thread_exit(void *pptr) {
299
nd_log(NDLS_DAEMON, NDLP_DEBUG, "thread with task id %d finished", nti->tid);
300
301
rrd_collector_finished();
302
- sender_commit_thread_buffer_free();
302
+ sender_thread_buffer_free();
303
rrdset_thread_rda_free();
304
query_target_free();
305
thread_cache_destroy();
src/libnetdata/uuid/uuid.h
+1
@@ -26,6 +26,7 @@ typedef struct _uuid {
26
#endif
27
28
static const ND_UUID UUID_ZERO = (ND_UUID){ { .parts = { .hig64 = 0, .low64 = 0 } }};
29
+ND_UUID_DEFINE(netdata_fatal_msgid, 0x23, 0xe9, 0x3d, 0xfc, 0xcb, 0xf6, 0x4e, 0x11, 0xaa, 0xc8, 0x58, 0xb9, 0x41, 0x0d, 0x8a, 0x82);
30
ND_UUID_DEFINE(streaming_from_child_msgid, 0xed,0x4c,0xdb, 0x8f, 0x1b, 0xeb, 0x4a, 0xd3, 0xb5, 0x7c, 0xb3, 0xca, 0xe2, 0xd1, 0x62, 0xfa);
31
ND_UUID_DEFINE(streaming_to_parent_msgid, 0x6e, 0x2e, 0x38, 0x39, 0x06, 0x76, 0x48, 0x96, 0x8b, 0x64, 0x60, 0x45, 0xdb, 0xf2, 0x8d, 0x66);
32
ND_UUID_DEFINE(health_alert_transition_msgid, 0x9c, 0xe0, 0xcb, 0x58, 0xab, 0x8b, 0x44, 0xdf, 0x82, 0xc4, 0xbf, 0x1a, 0xd9, 0xee, 0x22, 0xde);
src/libnetdata/waiting-queue/waiting-queue.c
new
+416
@@ -0,0 +1,416 @@
1
+// SPDX-License-Identifier: GPL-3.0-or-later
2
+
3
+#include "waiting-queue.h"
4
+
5
+typedef struct waiting_thread {
6
+ uv_cond_t cond; // condition variable for this thread
7
+ usec_t waiting_since_ut; // when we started waiting
8
+ Word_t order;
9
+ struct waiting_thread *prev, *next;
10
+} WAITING_THREAD;
11
+
12
+struct waiting_queue {
13
+ uv_mutex_t mutex; // protect the queue structure
14
+ Word_t last_seqno; // incrementing sequence counter
15
+ WAITING_THREAD *list;
16
+ size_t running; // number of threads currently running/waiting
17
+ SPINLOCK spinlock;
18
+};
19
+
20
+// Determine available bits based on system word size
21
+#if SIZEOF_VOID_P == 8
22
+#define PRIORITY_SHIFT 62ULL
23
+#define SEQNO_MASK ((1ULL << PRIORITY_SHIFT) - 1)
24
+#else
25
+#define PRIORITY_SHIFT 30U
26
+#define SEQNO_MASK ((1U << PRIORITY_SHIFT) - 1)
27
+#endif
28
+
29
+static inline Word_t make_key(WAITING_QUEUE_PRIORITY priority, Word_t seqno) {
30
+ return ((Word_t)priority << PRIORITY_SHIFT) | (seqno & SEQNO_MASK);
31
+}
32
+
33
+static inline WAITING_QUEUE_PRIORITY key_get_priority(Word_t key) {
34
+ return (WAITING_QUEUE_PRIORITY)(key >> PRIORITY_SHIFT);
35
+}
36
+
37
+static inline Word_t key_get_seqno(Word_t key) {
38
+ return key & SEQNO_MASK;
39
+}
40
+
41
+WAITING_QUEUE *waiting_queue_create(void) {
42
+ WAITING_QUEUE *wq = callocz(1, sizeof(WAITING_QUEUE));
43
+
44
+ int ret = uv_mutex_init(&wq->mutex);
45
+ if(ret != 0) {
46
+ freez(wq);
47
+ return NULL;
48
+ }
49
+
50
+ spinlock_init(&wq->spinlock);
51
+
52
+ wq->running = 0;
53
+ return wq;
54
+}
55
+
56
+void waiting_queue_destroy(WAITING_QUEUE *wq) {
57
+ if(!wq) return;
58
+
59
+ if(wq->running)
60
+ fatal("WAITING_QUEUE: destroying waiting queue that still has %zu threads running/waiting", wq->running);
61
+
62
+ uv_mutex_destroy(&wq->mutex);
63
+ freez(wq);
64
+}
65
+
66
+static inline void WAITERS_SET(WAITING_QUEUE *wq, WAITING_THREAD *wt) {
67
+ for(WAITING_THREAD *t = wq->list ; t ;t = t->next) {
68
+ if(wt->order < t->order) {
69
+ DOUBLE_LINKED_LIST_INSERT_ITEM_BEFORE_UNSAFE(wq->list, t, wt, prev, next);
70
+ return;
71
+ }
72
+ }
73
+ DOUBLE_LINKED_LIST_APPEND_ITEM_UNSAFE(wq->list, wt, prev, next);
74
+}
75
+
76
+static inline void WAITERS_DEL(WAITING_QUEUE *wq, WAITING_THREAD *wt) {
77
+ DOUBLE_LINKED_LIST_REMOVE_ITEM_UNSAFE(wq->list, wt, prev, next);
78
+}
79
+
80
+static inline WAITING_THREAD *WAITERS_FIRST(WAITING_QUEUE *wq) {
81
+ return wq->list;
82
+}
83
+
84
+static inline void WAITING_THREAD_init(WAITING_QUEUE *wq, WAITING_THREAD *wt, WAITING_QUEUE_PRIORITY priority) {
85
+ Word_t seqno = __atomic_add_fetch(&wq->last_seqno, 1, __ATOMIC_RELAXED);
86
+ wt->order = make_key(priority, seqno);
87
+ wt->waiting_since_ut = now_monotonic_usec();
88
+ wt->prev = wt->next = NULL;
89
+
90
+ int ret = uv_cond_init(&wt->cond);
91
+ if(ret != 0)
92
+ fatal("WAITING_QUEUE: cannot initialize condition variable");
93
+}
94
+
95
+static inline void WAITING_THREAD_cleanup(WAITING_QUEUE *wq __maybe_unused, WAITING_THREAD *wt) {
96
+ uv_cond_destroy(&wt->cond);
97
+}
98
+
99
+usec_t waiting_queue_wait(WAITING_QUEUE *wq, WAITING_QUEUE_PRIORITY priority) {
100
+ // Try fast path first - if we're the only one, just go
101
+ if(__atomic_add_fetch(&wq->running, 1, __ATOMIC_RELAXED) == 1) {
102
+ if(spinlock_trylock(&wq->spinlock))
103
+ return 0;
104
+ }
105
+
106
+ // Slow path - need to wait
107
+
108
+ WAITING_THREAD wt;
109
+ WAITING_THREAD_init(wq, &wt, priority);
110
+
111
+ uv_mutex_lock(&wq->mutex);
112
+ WAITERS_SET(wq, &wt);
113
+
114
+ // Wait for our turn
115
+ do {
116
+ if (WAITERS_FIRST(wq) == &wt && spinlock_trylock(&wq->spinlock))
117
+ break;
118
+ else
119
+ uv_cond_wait(&wt.cond, &wq->mutex);
120
+ } while(true);
121
+
122
+ WAITERS_DEL(wq, &wt);
123
+ uv_mutex_unlock(&wq->mutex);
124
+ WAITING_THREAD_cleanup(wq, &wt);
125
+
126
+ return now_monotonic_usec() - wt.waiting_since_ut;
127
+}
128
+
129
+void waiting_queue_done(WAITING_QUEUE *wq) {
130
+ spinlock_unlock(&wq->spinlock);
131
+
132
+ // Fast path if we're alone
133
+ if(__atomic_sub_fetch(&wq->running, 1, __ATOMIC_RELAXED) == 0)
134
+ return;
135
+
136
+ // Slow path - need to signal next in line
137
+ uv_mutex_lock(&wq->mutex);
138
+
139
+ // Wake up next in line if any
140
+ if(wq->list)
141
+ uv_cond_signal(&wq->list->cond);
142
+
143
+ uv_mutex_unlock(&wq->mutex);
144
+}
145
+
146
+size_t waiting_queue_waiting(WAITING_QUEUE *wq) {
147
+ return __atomic_load_n(&wq->running, __ATOMIC_RELAXED);
148
+}
149
+
150
+
151
+// --------------------------------------------------------------------------------------------------------------------
152
+
153
+// For stress test statistics
154
+typedef struct thread_stats {
155
+ WAITING_QUEUE_PRIORITY priority;
156
+ size_t executions; // how many times we got through
157
+ usec_t total_wait_time; // total time spent waiting
158
+ usec_t max_wait_time; // maximum time spent waiting
159
+} THREAD_STATS;
160
+
161
+struct thread_args {
162
+ THREAD_STATS *stats;
163
+ WAITING_QUEUE *wq;
164
+ bool with_sleep;
165
+ bool *stop_flag;
166
+};
167
+
168
+static const char *priority_to_string(WAITING_QUEUE_PRIORITY p) {
169
+ switch(p) {
170
+ case WAITING_QUEUE_PRIO_URGENT: return "URGENT";
171
+ case WAITING_QUEUE_PRIO_HIGH: return "HIGH";
172
+ case WAITING_QUEUE_PRIO_NORMAL: return "NORMAL";
173
+ case WAITING_QUEUE_PRIO_LOW: return "LOW";
174
+ default: return "UNKNOWN";
175
+ }
176
+}
177
+
178
+static int unittest_functional(void) {
179
+ int errors = 0;
180
+ fprintf(stderr, "\nTesting waiting queue...\n");
181
+
182
+ WAITING_QUEUE *wq = waiting_queue_create();
183
+
184
+ // Test 1: Fast path should work with no contention
185
+ fprintf(stderr, " Test 1: Fast path - no contention: ");
186
+ usec_t wait_time = waiting_queue_wait(wq, WAITING_QUEUE_PRIO_NORMAL);
187
+ waiting_queue_done(wq);
188
+ if(wait_time != 0) {
189
+ fprintf(stderr, "FAILED (waited %"PRIu64" usec)\n", wait_time);
190
+ errors++;
191
+ }
192
+ else
193
+ fprintf(stderr, "OK\n");
194
+
195
+ // Test 2: Priorities should be respected
196
+ fprintf(stderr, " Test 2: Priority ordering: ");
197
+ WAITING_THREAD threads[100];
198
+ for(size_t t = 0; t < _countof(threads); t++) {
199
+ __atomic_add_fetch(&wq->running, 1, __ATOMIC_RELAXED);
200
+ WAITING_THREAD_init(wq, &threads[t], os_random(WAITING_QUEUE_PRIO_MAX));
201
+ WAITERS_SET(wq, &threads[t]);
202
+ }
203
+
204
+ bool failed = false;
205
+ size_t prio_counts[WAITING_QUEUE_PRIO_MAX] = { 0 };
206
+ WAITING_QUEUE_PRIORITY last_prio = WAITING_QUEUE_PRIO_URGENT;
207
+ Word_t last_seqno = 0;
208
+ for(size_t t = 0; t < _countof(threads); t++) {
209
+ WAITING_THREAD *wt = WAITERS_FIRST(wq);
210
+ WAITERS_DEL(wq, wt);
211
+ __atomic_sub_fetch(&wq->running, 1, __ATOMIC_RELAXED);
212
+
213
+ WAITING_QUEUE_PRIORITY prio = key_get_priority(wt->order);
214
+ Word_t seqno = key_get_seqno(wt->order);
215
+
216
+ prio_counts[prio]++;
217
+ if(prio < last_prio) {
218
+ if(!failed)
219
+ fprintf(stderr, "FAILED\n");
220
+
221
+ fprintf(stderr, " > ERROR: prio %u is before prio %u\n", prio, last_prio);
222
+ errors++;
223
+ failed = true;
224
+ }
225
+ else if(prio == last_prio && seqno < last_seqno) {
226
+ if(!failed)
227
+ fprintf(stderr, "FAILED\n");
228
+
229
+ fprintf(stderr, " > ERROR: seqno %lu is before seqno %lu\n", seqno, last_seqno);
230
+ errors++;
231
+ failed = true;
232
+ }
233
+
234
+ last_seqno = seqno;
235
+ last_prio = prio;
236
+ WAITING_THREAD_cleanup(wq, wt);
237
+ }
238
+
239
+ if(!failed)
240
+ fprintf(stderr, "OK\n");
241
+
242
+ for(size_t p = 0; p < WAITING_QUEUE_PRIO_MAX ;p++)
243
+ fprintf(stderr, " > prio %zu got %zu waiters\n", p, prio_counts[p]);
244
+
245
+ // Test 3: Queue stats should be accurate
246
+ fprintf(stderr, " Test 3: Queue statistics: ");
247
+ size_t waiting = waiting_queue_waiting(wq);
248
+ if(waiting != 0) {
249
+ fprintf(stderr, "FAILED (queue shows %zu waiting)\n", waiting);
250
+ errors++;
251
+ }
252
+ else
253
+ fprintf(stderr, "OK\n");
254
+
255
+ waiting_queue_destroy(wq);
256
+ return errors;
257
+}
258
+
259
+static void *stress_thread(void *arg) {
260
+ struct thread_args *args = arg;
261
+
262
+ THREAD_STATS *stats = args->stats;
263
+ WAITING_QUEUE *wq = args->wq;
264
+ bool with_sleep = args->with_sleep;
265
+ bool *stop_flag = args->stop_flag;
266
+
267
+ while(!__atomic_load_n(stop_flag, __ATOMIC_ACQUIRE)) {
268
+ usec_t wait_time = waiting_queue_wait(wq, stats->priority);
269
+ stats->executions++;
270
+ stats->total_wait_time += wait_time;
271
+ if(wait_time > stats->max_wait_time)
272
+ stats->max_wait_time = wait_time;
273
+
274
+ if(with_sleep)
275
+ tinysleep();
276
+
277
+ waiting_queue_done(wq);
278
+ }
279
+
280
+ return NULL;
281
+}
282
+
283
+static void print_thread_stats(THREAD_STATS *stats, size_t count, usec_t duration) {
284
+ fprintf(stderr, "\n%-8s %12s %12s %12s %12s %12s\n",
285
+ "PRIORITY", "EXECUTIONS", "EXEC/SEC", "AVG WAIT", "MAX WAIT", "% WAITING");
286
+
287
+ size_t total_execs = 0;
288
+ for(size_t i = 0; i < count; i++)
289
+ total_execs += stats[i].executions;
290
+
291
+ double total_time_sec = duration / (double)USEC_PER_SEC;
292
+
293
+ for(size_t i = 0; i < count; i++) {
294
+ double execs_per_sec = stats[i].executions / total_time_sec;
295
+ double avg_wait = stats[i].executions ? (double)stats[i].total_wait_time / stats[i].executions : 0;
296
+ double percent_waiting = stats[i].total_wait_time * 100.0 / duration;
297
+
298
+ fprintf(stderr, "%-8s %12zu %12.1f %12.1f %12"PRIu64" %12.1f%%\n",
299
+ priority_to_string(stats[i].priority),
300
+ stats[i].executions,
301
+ execs_per_sec,
302
+ avg_wait,
303
+ stats[i].max_wait_time,
304
+ percent_waiting);
305
+ }
306
+}
307
+
308
+#define THREADS_PER_PRIORITY 2
309
+#define TEST_DURATION_SEC 5
310
+
311
+static int unittest_stress(void) {
312
+ int errors = 0;
313
+ fprintf(stderr, "\nStress testing waiting queue...\n");
314
+
315
+ WAITING_QUEUE *wq = waiting_queue_create();
316
+ const size_t num_priorities = 4;
317
+ const size_t total_threads = num_priorities * THREADS_PER_PRIORITY;
318
+
319
+ // Test both with and without sleep
320
+ for(int test = 0; test < 2; test++) {
321
+ bool with_sleep = (test == 1);
322
+ bool stop_flag = false;
323
+
324
+ fprintf(stderr, "\nRunning %ds stress test %s sleep:\n",
325
+ TEST_DURATION_SEC, with_sleep ? "with" : "without");
326
+
327
+ // Prepare thread stats and args
328
+ THREAD_STATS stats[total_threads];
329
+ struct thread_args thread_args[total_threads];
330
+ ND_THREAD *threads[total_threads];
331
+
332
+ fprintf(stderr, "Starting %zu threads for %ds test %s sleep...\n",
333
+ total_threads,
334
+ TEST_DURATION_SEC,
335
+ with_sleep ? "with" : "without");
336
+
337
+ // Initialize stats and create threads
338
+ size_t thread_idx = 0;
339
+ for(int prio = WAITING_QUEUE_PRIO_URGENT; prio <= WAITING_QUEUE_PRIO_LOW; prio++) {
340
+ for(int t = 0; t < THREADS_PER_PRIORITY; t++) {
341
+ stats[thread_idx] = (THREAD_STATS){
342
+ .priority = prio,
343
+ .executions = 0,
344
+ .total_wait_time = 0,
345
+ .max_wait_time = 0
346
+ };
347
+ thread_args[thread_idx] = (struct thread_args){
348
+ .stats = &stats[thread_idx], // Pass pointer to stats
349
+ .wq = wq,
350
+ .with_sleep = with_sleep,
351
+ .stop_flag = &stop_flag
352
+ };
353
+
354
+ char thread_name[32];
355
+ snprintf(thread_name, sizeof(thread_name), "STRESS%d-%d", prio, t);
356
+ threads[thread_idx] = nd_thread_create(
357
+ thread_name,
358
+ NETDATA_THREAD_OPTION_DONT_LOG | NETDATA_THREAD_OPTION_JOINABLE,
359
+ stress_thread,
360
+ &thread_args[thread_idx]);
361
+ thread_idx++;
362
+ }
363
+ }
364
+
365
+ // Let it run
366
+ time_t start = now_monotonic_sec();
367
+ fprintf(stderr, "Running...");
368
+ while(now_monotonic_sec() - start < TEST_DURATION_SEC) {
369
+ fprintf(stderr, ".");
370
+ sleep_usec(500000); // Print a dot every 0.5 seconds
371
+ }
372
+ fprintf(stderr, "\n");
373
+
374
+
375
+ fprintf(stderr, "Stopping threads...\n");
376
+ __atomic_store_n(&stop_flag, true, __ATOMIC_RELEASE);
377
+
378
+ // Wait for threads and collect stats
379
+ fprintf(stderr, "Waiting for %zu threads to finish...\n", total_threads);
380
+ for(size_t i = 0; i < total_threads; i++)
381
+ nd_thread_join(threads[i]);
382
+
383
+ // Print stats
384
+ print_thread_stats(stats, total_threads, TEST_DURATION_SEC * USEC_PER_SEC);
385
+
386
+// // Basic validation
387
+// for(size_t i = 0; i < total_threads - THREADS_PER_PRIORITY; i++) {
388
+// if(stats[i].executions < stats[i + THREADS_PER_PRIORITY].executions) {
389
+// fprintf(stderr, "ERROR: Higher priority thread got fewer executions!\n");
390
+// errors++;
391
+// }
392
+// }
393
+//
394
+// // Check fairness within same priority
395
+// for(size_t i = 0; i < total_threads; i += THREADS_PER_PRIORITY) {
396
+// for(size_t j = i + 1; j < i + THREADS_PER_PRIORITY; j++) {
397
+// double diff = (double)(stats[i].executions - stats[j].executions) /
398
+// (double)(stats[i].executions + stats[j].executions);
399
+// if(fabs(diff) > 0.1) { // allow 10% difference
400
+// fprintf(stderr, "ERROR: Unfair distribution within same priority!\n");
401
+// errors++;
402
+// }
403
+// }
404
+// }
405
+ }
406
+
407
+ waiting_queue_destroy(wq);
408
+ return errors;
409
+}
410
+
411
+int unittest_waiting_queue(void) {
412
+ int errors = unittest_functional();
413
+ errors += unittest_stress();
414
+
415
+ return errors;
416
+}
src/libnetdata/waiting-queue/waiting-queue.h
new
+58
@@ -0,0 +1,58 @@
1
+// SPDX-License-Identifier: GPL-3.0-or-later
2
+
3
+#ifndef NETDATA_WAITING_QUEUE_H
4
+#define NETDATA_WAITING_QUEUE_H
5
+
6
+#include "libnetdata/libnetdata.h"
7
+#include <uv.h>
8
+
9
+/*
10
+ * WAITING QUEUE
11
+ * Like a mutex, or a spinlock, but:
12
+ *
13
+ * 1. Waiters get a sequence number (FIFO)
14
+ * 2. FIFO is respected within each priority
15
+ * 3. Higher priority threads get in first
16
+ * 4. No wasting of resources, there are no spins
17
+ *
18
+ * When there are no other waiters, this is equivalent to 2 atomic
19
+ * operations for lock, and 2 for unlock.
20
+ *
21
+ * As lightweight and fast as it can be.
22
+ * About 0.5M thread switches/s per WAITING QUEUE, on modern hardware.
23
+ *
24
+ * Be careful: higher priority threads can starve the rest!
25
+ *
26
+ */
27
+
28
+typedef struct waiting_queue WAITING_QUEUE;
29
+
30
+typedef enum __attribute__((packed)) {
31
+ WAITING_QUEUE_PRIO_URGENT = 0, // will be first
32
+ WAITING_QUEUE_PRIO_HIGH, // will be second
33
+ WAITING_QUEUE_PRIO_NORMAL, // will be third
34
+ WAITING_QUEUE_PRIO_LOW, // will be last
35
+
36
+ // terminator
37
+ WAITING_QUEUE_PRIO_MAX,
38
+} WAITING_QUEUE_PRIORITY;
39
+
40
+// Initialize a waiting queue
41
+WAITING_QUEUE *waiting_queue_create(void);
42
+
43
+// Destroy a waiting queue - must be empty
44
+void waiting_queue_destroy(WAITING_QUEUE *wq);
45
+
46
+// Returns when it is our turn to run
47
+// Returns time spent waiting in microseconds
48
+usec_t waiting_queue_wait(WAITING_QUEUE *wq, WAITING_QUEUE_PRIORITY priority);
49
+
50
+// Mark that we are done - wakes up the next in line
51
+void waiting_queue_done(WAITING_QUEUE *wq);
52
+
53
+// Return the number of threads currently waiting
54
+size_t waiting_queue_waiting(WAITING_QUEUE *wq);
55
+
56
+int unittest_waiting_queue(void);
57
+
58
+#endif // NETDATA_WAITING_QUEUE_H
\ No newline at end of file
src/libnetdata/worker_utilization/worker_utilization.c
+6
-2
@@ -118,7 +118,9 @@ void worker_register(const char *name) {
118
119
workers_globals.memory += sizeof(struct worker) + strlen(worker->tag) + 1 + strlen(worker->workname) + 1;
120
121
+ JudyAllocThreadPulseReset();
122
Pvoid_t *PValue = JudyHSIns(&workers_globals.worknames_JudyHS, (void *)name, name_size, PJE0);
123
+ int64_t judy_mem = JudyAllocThreadPulseGetAndReset();
124
125
struct workers_workname *workname = *PValue;
126
if(!workname) {
@@ -127,7 +129,7 @@ void worker_register(const char *name) {
129
workname->base = NULL;
130
*PValue = workname;
131
130
- workers_globals.memory += sizeof(struct workers_workname) + JUDYHS_INDEX_SIZE_ESTIMATE(name_size);
132
+ workers_globals.memory = (int64_t)workers_globals.memory + (int64_t)sizeof(struct workers_workname) + judy_mem;
133
}
134
135
spinlock_lock(&workname->spinlock);
@@ -176,9 +178,11 @@ void worker_unregister(void) {
178
spinlock_unlock(&workname->spinlock);
179
180
if(!workname->base) {
181
+ JudyAllocThreadPulseReset();
182
JudyHSDel(&workers_globals.worknames_JudyHS, (void *) worker->workname, workname_size, PJE0);
183
+ int64_t judy_mem = JudyAllocThreadPulseGetAndReset();
184
freez(workname);
181
- workers_globals.memory -= sizeof(struct workers_workname) + JUDYHS_INDEX_SIZE_ESTIMATE(workname_size);
185
+ workers_globals.memory = (int64_t)workers_globals.memory - (int64_t)sizeof(struct workers_workname) + judy_mem;
186
}
187
}
188
workers_globals.memory -= sizeof(struct worker) + strlen(worker->tag) + 1 + strlen(worker->workname) + 1;
src/ml/ml_config.cc
+1
-1
@@ -45,7 +45,7 @@ void ml_config_load(ml_config_t *cfg) {
45
std::string anomaly_detection_grouping_method = config_get(config_section_ml, "anomaly detection grouping method", "average");
46
time_t anomaly_detection_query_duration = config_get_duration_seconds(config_section_ml, "anomaly detection grouping duration", 5 * 60);
47
48
- size_t num_worker_threads = stream_conf_is_parent(false) ? netdata_conf_cpus() / 4 : 1;
48
+ size_t num_worker_threads = netdata_conf_is_parent() ? netdata_conf_cpus() / 4 : 1;
49
if (num_worker_threads < 1) num_worker_threads = 1;
50
else if (num_worker_threads > 256) num_worker_threads = 256;
51
num_worker_threads = config_get_number(config_section_ml, "num training threads", num_worker_threads);
src/plugins.d/gperf-config.txt
+13
-13
@@ -73,36 +73,36 @@ HOST_LABEL, PLUGINSD_KEYWORD_ID_HOST_LABEL, PARSER_INIT_PLUGINSD|PARSE
73
#
74
# Common keywords
75
#
76
-BEGIN, PLUGINSD_KEYWORD_ID_BEGIN, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 8
77
-CHART, PLUGINSD_KEYWORD_ID_CHART, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 9
76
+BEGIN, PLUGINSD_KEYWORD_ID_BEGIN, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING|PARSER_REP_DATA, WORKER_PARSER_FIRST_JOB + 8
77
+CHART, PLUGINSD_KEYWORD_ID_CHART, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING|PARSER_REP_METADATA|PARSER_REP_REPLICATION, WORKER_PARSER_FIRST_JOB + 9
78
CLABEL, PLUGINSD_KEYWORD_ID_CLABEL, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 10
79
CLABEL_COMMIT, PLUGINSD_KEYWORD_ID_CLABEL_COMMIT, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 11
80
DIMENSION, PLUGINSD_KEYWORD_ID_DIMENSION, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 12
81
-END, PLUGINSD_KEYWORD_ID_END, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 13
81
+END, PLUGINSD_KEYWORD_ID_END, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING|PARSER_REP_DATA, WORKER_PARSER_FIRST_JOB + 13
82
FUNCTION, PLUGINSD_KEYWORD_ID_FUNCTION, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 14
83
FUNCTION_RESULT_BEGIN, PLUGINSD_KEYWORD_ID_FUNCTION_RESULT_BEGIN, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 15
84
FUNCTION_PROGRESS, PLUGINSD_KEYWORD_ID_FUNCTION_PROGRESS, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 16
85
LABEL, PLUGINSD_KEYWORD_ID_LABEL, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 17
86
OVERWRITE, PLUGINSD_KEYWORD_ID_OVERWRITE, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 18
87
-SET, PLUGINSD_KEYWORD_ID_SET, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 19
87
+SET, PLUGINSD_KEYWORD_ID_SET, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING|PARSER_REP_DATA, WORKER_PARSER_FIRST_JOB + 19
88
VARIABLE, PLUGINSD_KEYWORD_ID_VARIABLE, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 20
89
CONFIG, PLUGINSD_KEYWORD_ID_CONFIG, PARSER_INIT_PLUGINSD|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 21
90
#
91
# Streaming only keywords
92
#
93
CLAIMED_ID, PLUGINSD_KEYWORD_ID_CLAIMED_ID, PARSER_INIT_STREAMING|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 22
94
-BEGIN2, PLUGINSD_KEYWORD_ID_BEGIN2, PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 23
95
-SET2, PLUGINSD_KEYWORD_ID_SET2, PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 24
96
-END2, PLUGINSD_KEYWORD_ID_END2, PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 25
94
+BEGIN2, PLUGINSD_KEYWORD_ID_BEGIN2, PARSER_INIT_STREAMING|PARSER_REP_DATA, WORKER_PARSER_FIRST_JOB + 23
95
+SET2, PLUGINSD_KEYWORD_ID_SET2, PARSER_INIT_STREAMING|PARSER_REP_DATA, WORKER_PARSER_FIRST_JOB + 24
96
+END2, PLUGINSD_KEYWORD_ID_END2, PARSER_INIT_STREAMING|PARSER_REP_DATA, WORKER_PARSER_FIRST_JOB + 25
97
#
98
# Streaming Replication keywords
99
#
100
-CHART_DEFINITION_END, PLUGINSD_KEYWORD_ID_CHART_DEFINITION_END, PARSER_INIT_STREAMING|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 26
101
-RBEGIN, PLUGINSD_KEYWORD_ID_RBEGIN, PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 27
102
-RDSTATE, PLUGINSD_KEYWORD_ID_RDSTATE, PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 28
103
-REND, PLUGINSD_KEYWORD_ID_REND, PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 29
104
-RSET, PLUGINSD_KEYWORD_ID_RSET, PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 30
105
-RSSTATE, PLUGINSD_KEYWORD_ID_RSSTATE, PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 31
100
+CHART_DEFINITION_END, PLUGINSD_KEYWORD_ID_CHART_DEFINITION_END, PARSER_INIT_STREAMING|PARSER_REP_REPLICATION|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 26
101
+RBEGIN, PLUGINSD_KEYWORD_ID_RBEGIN, PARSER_INIT_STREAMING|PARSER_REP_REPLICATION|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 27
102
+RSET, PLUGINSD_KEYWORD_ID_RSET, PARSER_INIT_STREAMING|PARSER_REP_REPLICATION|PARSER_REP_DATA, WORKER_PARSER_FIRST_JOB + 28
103
+REND, PLUGINSD_KEYWORD_ID_REND, PARSER_INIT_STREAMING|PARSER_REP_REPLICATION|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 29
104
+RDSTATE, PLUGINSD_KEYWORD_ID_RDSTATE, PARSER_INIT_STREAMING|PARSER_REP_REPLICATION|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 30
105
+RSSTATE, PLUGINSD_KEYWORD_ID_RSSTATE, PARSER_INIT_STREAMING|PARSER_REP_REPLICATION|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 31
106
#
107
# JSON
108
#
src/plugins.d/gperf-hashtable.h
+16
-16
@@ -1,5 +1,5 @@
1
/* ANSI-C code produced by gperf version 3.1 */
2
-/* Command-line: gperf --multiple-iterations=1000 --output-file=gperf-hashtable.h gperf-config.txt */
2
+/* Command-line: gperf --multiple-iterations=1000 --output-file=gperf-hashtable.h -r gperf-config.txt */
3
/* Computed positions: -k'1-2' */
4
5
#if !((' ' == 32) && ('!' == 33) && ('"' == 34) && ('#' == 35) \
@@ -135,11 +135,11 @@ static const PARSER_KEYWORD gperf_keywords[] =
135
#line 69 "gperf-config.txt"
136
{"HOST", PLUGINSD_KEYWORD_ID_HOST, PARSER_INIT_PLUGINSD|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 4},
137
#line 103 "gperf-config.txt"
138
- {"REND", PLUGINSD_KEYWORD_ID_REND, PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 29},
138
+ {"REND", PLUGINSD_KEYWORD_ID_REND, PARSER_INIT_STREAMING|PARSER_REP_REPLICATION|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 29},
139
#line 68 "gperf-config.txt"
140
{"EXIT", PLUGINSD_KEYWORD_ID_EXIT, PARSER_INIT_PLUGINSD, WORKER_PARSER_FIRST_JOB + 3},
141
#line 77 "gperf-config.txt"
142
- {"CHART", PLUGINSD_KEYWORD_ID_CHART, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 9},
142
+ {"CHART", PLUGINSD_KEYWORD_ID_CHART, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING|PARSER_REP_METADATA|PARSER_REP_REPLICATION, WORKER_PARSER_FIRST_JOB + 9},
143
#line 89 "gperf-config.txt"
144
{"CONFIG", PLUGINSD_KEYWORD_ID_CONFIG, PARSER_INIT_PLUGINSD|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 21},
145
#line 86 "gperf-config.txt"
@@ -148,8 +148,8 @@ static const PARSER_KEYWORD gperf_keywords[] =
148
{"HOST_LABEL", PLUGINSD_KEYWORD_ID_HOST_LABEL, PARSER_INIT_PLUGINSD|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 7},
149
#line 70 "gperf-config.txt"
150
{"HOST_DEFINE", PLUGINSD_KEYWORD_ID_HOST_DEFINE, PARSER_INIT_PLUGINSD|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 5},
151
-#line 102 "gperf-config.txt"
152
- {"RDSTATE", PLUGINSD_KEYWORD_ID_RDSTATE, PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 28},
151
+#line 104 "gperf-config.txt"
152
+ {"RDSTATE", PLUGINSD_KEYWORD_ID_RDSTATE, PARSER_INIT_STREAMING|PARSER_REP_REPLICATION|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 30},
153
{(char*)0,0,PARSER_INIT_PLUGINSD,0},
154
#line 118 "gperf-config.txt"
155
{"DELETE_JOB", PLUGINSD_KEYWORD_ID_DELETE_JOB, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 38},
@@ -162,17 +162,17 @@ static const PARSER_KEYWORD gperf_keywords[] =
162
#line 117 "gperf-config.txt"
163
{"REPORT_JOB_STATUS", PLUGINSD_KEYWORD_ID_REPORT_JOB_STATUS, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 37},
164
#line 87 "gperf-config.txt"
165
- {"SET", PLUGINSD_KEYWORD_ID_SET, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 19},
165
+ {"SET", PLUGINSD_KEYWORD_ID_SET, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING|PARSER_REP_DATA, WORKER_PARSER_FIRST_JOB + 19},
166
#line 95 "gperf-config.txt"
167
- {"SET2", PLUGINSD_KEYWORD_ID_SET2, PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 24},
168
-#line 104 "gperf-config.txt"
169
- {"RSET", PLUGINSD_KEYWORD_ID_RSET, PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 30},
167
+ {"SET2", PLUGINSD_KEYWORD_ID_SET2, PARSER_INIT_STREAMING|PARSER_REP_DATA, WORKER_PARSER_FIRST_JOB + 24},
168
+#line 102 "gperf-config.txt"
169
+ {"RSET", PLUGINSD_KEYWORD_ID_RSET, PARSER_INIT_STREAMING|PARSER_REP_REPLICATION|PARSER_REP_DATA, WORKER_PARSER_FIRST_JOB + 28},
170
#line 100 "gperf-config.txt"
171
- {"CHART_DEFINITION_END", PLUGINSD_KEYWORD_ID_CHART_DEFINITION_END, PARSER_INIT_STREAMING|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 26},
171
+ {"CHART_DEFINITION_END", PLUGINSD_KEYWORD_ID_CHART_DEFINITION_END, PARSER_INIT_STREAMING|PARSER_REP_REPLICATION|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 26},
172
#line 115 "gperf-config.txt"
173
{"DYNCFG_REGISTER_JOB", PLUGINSD_KEYWORD_ID_DYNCFG_REGISTER_JOB, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 35},
174
#line 105 "gperf-config.txt"
175
- {"RSSTATE", PLUGINSD_KEYWORD_ID_RSSTATE, PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 31},
175
+ {"RSSTATE", PLUGINSD_KEYWORD_ID_RSSTATE, PARSER_INIT_STREAMING|PARSER_REP_REPLICATION|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 31},
176
#line 78 "gperf-config.txt"
177
{"CLABEL", PLUGINSD_KEYWORD_ID_CLABEL, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 10},
178
#line 114 "gperf-config.txt"
@@ -184,17 +184,17 @@ static const PARSER_KEYWORD gperf_keywords[] =
184
#line 93 "gperf-config.txt"
185
{"CLAIMED_ID", PLUGINSD_KEYWORD_ID_CLAIMED_ID, PARSER_INIT_STREAMING|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 22},
186
#line 81 "gperf-config.txt"
187
- {"END", PLUGINSD_KEYWORD_ID_END, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 13},
187
+ {"END", PLUGINSD_KEYWORD_ID_END, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING|PARSER_REP_DATA, WORKER_PARSER_FIRST_JOB + 13},
188
#line 96 "gperf-config.txt"
189
- {"END2", PLUGINSD_KEYWORD_ID_END2, PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 25},
189
+ {"END2", PLUGINSD_KEYWORD_ID_END2, PARSER_INIT_STREAMING|PARSER_REP_DATA, WORKER_PARSER_FIRST_JOB + 25},
190
#line 79 "gperf-config.txt"
191
{"CLABEL_COMMIT", PLUGINSD_KEYWORD_ID_CLABEL_COMMIT, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 11},
192
#line 76 "gperf-config.txt"
193
- {"BEGIN", PLUGINSD_KEYWORD_ID_BEGIN, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 8},
193
+ {"BEGIN", PLUGINSD_KEYWORD_ID_BEGIN, PARSER_INIT_PLUGINSD|PARSER_INIT_STREAMING|PARSER_REP_DATA, WORKER_PARSER_FIRST_JOB + 8},
194
#line 94 "gperf-config.txt"
195
- {"BEGIN2", PLUGINSD_KEYWORD_ID_BEGIN2, PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 23},
195
+ {"BEGIN2", PLUGINSD_KEYWORD_ID_BEGIN2, PARSER_INIT_STREAMING|PARSER_REP_DATA, WORKER_PARSER_FIRST_JOB + 23},
196
#line 101 "gperf-config.txt"
197
- {"RBEGIN", PLUGINSD_KEYWORD_ID_RBEGIN, PARSER_INIT_STREAMING, WORKER_PARSER_FIRST_JOB + 27},
197
+ {"RBEGIN", PLUGINSD_KEYWORD_ID_RBEGIN, PARSER_INIT_STREAMING|PARSER_REP_REPLICATION|PARSER_REP_METADATA, WORKER_PARSER_FIRST_JOB + 27},
198
#line 67 "gperf-config.txt"
199
{"DISABLE", PLUGINSD_KEYWORD_ID_DISABLE, PARSER_INIT_PLUGINSD, WORKER_PARSER_FIRST_JOB + 2},
200
#line 84 "gperf-config.txt"
src/plugins.d/pluginsd_parser.c
+4
-86
@@ -1,9 +1,7 @@
1
// SPDX-License-Identifier: GPL-3.0-or-later
2
3
#include "pluginsd_internals.h"
4
-#include "streaming/replication.h"
5
-#include "streaming/stream-waiting-list.h"
6
-#include "web/api/queries/backfill.h"
4
+#include "streaming/stream-replication-receiver.h"
5
6
static inline PARSER_RC pluginsd_set(char **words, size_t num_words, PARSER *parser) {
7
int idx = 1;
@@ -67,7 +65,7 @@ static inline PARSER_RC pluginsd_begin(char **words, size_t num_words, PARSER *p
65
rrdhost_hostname(host), rrdset_id(st),
66
st->last_collected_time.tv_sec * USEC_PER_SEC + st->last_collected_time.tv_usec,
67
st->last_updated.tv_sec * USEC_PER_SEC + st->last_updated.tv_usec,
70
- microseconds
68
+ (long long unsigned)microseconds
69
);
70
}
71
#endif
@@ -189,7 +187,7 @@ static inline PARSER_RC pluginsd_host_define_end(char **words __maybe_unused, si
187
netdata_configured_utc_offset,
188
program_name,
189
NETDATA_VERSION,
192
- default_rrd_update_every,
190
+ nd_profile.update_every,
191
default_rrd_history_entries,
192
default_rrd_memory_mode,
193
health_plugin_enabled(),
@@ -205,7 +203,7 @@ static inline PARSER_RC pluginsd_host_define_end(char **words __maybe_unused, si
203
204
rrdhost_option_set(host, RRDHOST_OPTION_VIRTUAL_HOST);
205
rrdhost_flag_set(host, RRDHOST_FLAG_COLLECTOR_ONLINE);
208
- rrdhost_state_connected(host);
206
+ object_state_activate(&host->state_id);
207
ml_host_start(host);
208
dyncfg_host_init(host);
209
@@ -376,79 +374,6 @@ static inline PARSER_RC pluginsd_chart(char **words, size_t num_words, PARSER *p
374
return PARSER_RC_OK;
375
}
376
379
-static bool backfill_callback(size_t successful_dims __maybe_unused, size_t failed_dims __maybe_unused, struct backfill_request_data *brd) {
380
- if(!rrdhost_state_acquire(brd->host, brd->rrdhost_receiver_state_id))
381
- return false;
382
-
383
- bool rc = replicate_chart_request(send_to_plugin, brd->parser, brd->host, brd->st,
384
- brd->first_entry_child, brd->last_entry_child, brd->child_wall_clock_time,
385
- 0, 0);
386
- if (rc) {
387
- rrdset_flag_set(brd->st, RRDSET_FLAG_RECEIVER_REPLICATION_IN_PROGRESS);
388
- rrdset_flag_clear(brd->st, RRDSET_FLAG_RECEIVER_REPLICATION_FINISHED);
389
- rrdhost_receiver_replicating_charts_plus_one(brd->st->rrdhost);
390
- }
391
- else {
392
- netdata_log_error(
393
- "PLUGINSD: 'host:%s' failed to initiate replication for 'chart:%s'",
394
- rrdhost_hostname(brd->host),
395
- rrdset_id(brd->st));
396
- }
397
-
398
- rrdhost_state_release(brd->host);
399
- return rc;
400
-}
401
-
402
-static inline PARSER_RC pluginsd_chart_definition_end(char **words, size_t num_words, PARSER *parser) {
403
- const char *first_entry_txt = get_word(words, num_words, 1);
404
- const char *last_entry_txt = get_word(words, num_words, 2);
405
- const char *wall_clock_time_txt = get_word(words, num_words, 3);
406
-
407
- RRDHOST *host = pluginsd_require_scope_host(parser, PLUGINSD_KEYWORD_CHART_DEFINITION_END);
408
- if(!host) return PLUGINSD_DISABLE_PLUGIN(parser, NULL, NULL);
409
-
410
- RRDSET *st = pluginsd_require_scope_chart(parser, PLUGINSD_KEYWORD_CHART_DEFINITION_END, PLUGINSD_KEYWORD_CHART);
411
- if(!st) return PLUGINSD_DISABLE_PLUGIN(parser, NULL, NULL);
412
-
413
- time_t first_entry_child = (first_entry_txt && *first_entry_txt) ? (time_t)str2ul(first_entry_txt) : 0;
414
- time_t last_entry_child = (last_entry_txt && *last_entry_txt) ? (time_t)str2ul(last_entry_txt) : 0;
415
- time_t child_wall_clock_time = (wall_clock_time_txt && *wall_clock_time_txt) ? (time_t)str2ul(wall_clock_time_txt) : now_realtime_sec();
416
-
417
- bool ok = true;
418
- if(!rrdset_flag_check(st, RRDSET_FLAG_RECEIVER_REPLICATION_IN_PROGRESS)) {
419
-
420
-#ifdef NETDATA_LOG_REPLICATION_REQUESTS
421
- st->replay.start_streaming = false;
422
- st->replay.after = 0;
423
- st->replay.before = 0;
424
-#endif
425
-
426
- struct backfill_request_data brd = {
427
- .rrdhost_receiver_state_id = rrdhost_state_id(host),
428
- .parser = parser,
429
- .host = host,
430
- .st = st,
431
- .first_entry_child = first_entry_child,
432
- .last_entry_child = last_entry_child,
433
- .child_wall_clock_time = child_wall_clock_time,
434
- };
435
-
436
- ok = backfill_request_add(st, backfill_callback, &brd);
437
- if(!ok)
438
- ok = backfill_callback(0, 0, &brd);
439
- }
440
-#ifdef NETDATA_LOG_REPLICATION_REQUESTS
441
- else {
442
- internal_error(true, "REPLAY: 'host:%s/chart:%s' not sending duplicate replication request",
443
- rrdhost_hostname(st->rrdhost), rrdset_id(st));
444
- }
445
-#endif
446
-
447
- stream_thread_received_metadata();
448
-
449
- return ok ? PARSER_RC_OK : PARSER_RC_ERROR;
450
-}
451
-
377
static inline PARSER_RC pluginsd_dimension(char **words, size_t num_words, PARSER *parser) {
378
int idx = 1;
379
ssize_t slot = pluginsd_parse_rrd_slot(words, num_words);
@@ -1179,13 +1104,6 @@ void pluginsd_process_cleanup(PARSER *parser) {
1104
pluginsd_cleanup_v2(parser);
1105
pluginsd_host_define_cleanup(parser);
1106
1182
-#ifdef NETDATA_LOG_STREAM_RECEIVE
1183
- if(parser->user.stream_log_fp) {
1184
- fclose(parser->user.stream_log_fp);
1185
- parser->user.stream_log_fp = NULL;
1186
- }
1187
-#endif
1188
-
1107
parser_destroy(parser);
1108
}
1109
src/plugins.d/pluginsd_parser.h
+16
-11
@@ -5,6 +5,10 @@
5
6
#include "daemon/common.h"
7
8
+#ifdef NETDATA_LOG_STREAM_RECEIVER
9
+#include "streaming/stream-receiver-internals.h"
10
+#endif
11
+
12
#define WORKER_PARSER_FIRST_JOB 36
13
14
// this has to be in-sync with the same at stream-thread.c
@@ -31,7 +35,9 @@ typedef enum __attribute__ ((__packed__)) parser_input_type {
35
typedef enum __attribute__ ((__packed__)) {
36
PARSER_INIT_PLUGINSD = (1 << 1),
37
PARSER_INIT_STREAMING = (1 << 2),
34
- PARSER_REP_METADATA = (1 << 3),
38
+ PARSER_REP_REPLICATION = (1 << 3),
39
+ PARSER_REP_METADATA = (1 << 4),
40
+ PARSER_REP_DATA = (1 << 5),
41
} PARSER_REPERTOIRE;
42
43
struct parser;
@@ -56,9 +62,8 @@ typedef struct parser_user_object {
62
size_t data_collections_count;
63
int enabled;
64
59
-#ifdef NETDATA_LOG_STREAM_RECEIVE
60
- FILE *stream_log_fp;
61
- PARSER_REPERTOIRE stream_log_repertoire;
65
+#ifdef NETDATA_LOG_STREAM_RECEIVER
66
+ void *rpt;
67
#endif
68
69
STREAM_CAPABILITIES capabilities; // receiver capabilities
@@ -172,8 +177,8 @@ bool parser_reconstruct_instance(BUFFER *wb, void *ptr);
177
bool parser_reconstruct_context(BUFFER *wb, void *ptr);
178
179
static inline int parser_action(PARSER *parser, char *input) {
175
-#ifdef NETDATA_LOG_STREAM_RECEIVE
176
- static __thread char line[PLUGINSD_LINE_MAX + 1];
180
+#ifdef NETDATA_LOG_STREAM_RECEIVER
181
+ char line[1024];
182
strncpyz(line, input, sizeof(line) - 1);
183
#endif
184
@@ -222,11 +227,6 @@ static inline int parser_action(PARSER *parser, char *input) {
227
if(likely(parser->keyword)) {
228
worker_is_busy(parser->keyword->worker_job_id);
229
225
-#ifdef NETDATA_LOG_STREAM_RECEIVE
226
- if(parser->user.stream_log_fp && parser->keyword->repertoire & parser->user.stream_log_repertoire)
227
- fprintf(parser->user.stream_log_fp, "%s", line);
228
-#endif
229
-
230
rc = parser_execute(parser, parser->keyword, parser->line.words, parser->line.num_words);
231
// rc = (*t->func)(words, num_words, parser);
232
worker_is_idle();
@@ -241,6 +241,11 @@ static inline int parser_action(PARSER *parser, char *input) {
241
command, parser->line.count, buffer_tostring(wb));
242
}
243
244
+#ifdef NETDATA_LOG_STREAM_RECEIVER
245
+ if((parser->keyword->repertoire & PARSER_REP_REPLICATION) && !(parser->keyword->repertoire & PARSER_REP_DATA))
246
+ stream_receiver_log_payload(parser->user.rpt, line, STREAM_TRAFFIC_TYPE_REPLICATION, true);
247
+#endif
248
+
249
line_splitter_reset(&parser->line);
250
return (rc == PARSER_RC_ERROR || rc == PARSER_RC_STOP);
251
}
src/plugins.d/pluginsd_replication.c
+143
-31
@@ -2,8 +2,109 @@
2
3
#include "pluginsd_replication.h"
4
#include "streaming/stream-receiver-internals.h"
5
-#include "streaming/replication.h"
5
+#include "streaming/stream-replication-receiver.h"
6
#include "streaming/stream-waiting-list.h"
7
+#include "web/api/queries/backfill.h"
8
+
9
+static bool backfill_callback(size_t successful_dims __maybe_unused, size_t failed_dims __maybe_unused, struct backfill_request_data *brd) {
10
+ if(!object_state_acquire(&brd->host->state_id, brd->host_state_id)) {
11
+ // this may happen because the host got reconnected
12
+
13
+ nd_log(NDLS_DAEMON, NDLP_DEBUG,
14
+ "PLUGINSD REPLAY ERROR: 'host:%s' failed to acquire host for sending replication"
15
+ " command for 'chart:%s'",
16
+ rrdhost_hostname(brd->host),
17
+ rrdset_id(brd->st));
18
+
19
+ return false;
20
+ }
21
+
22
+ __atomic_sub_fetch(&brd->host->stream.rcv.status.replication.backfill_pending, 1, __ATOMIC_RELAXED);
23
+
24
+ bool rc = replicate_chart_request(send_to_plugin, brd->parser, brd->host, brd->st,
25
+ brd->first_entry_child, brd->last_entry_child, brd->child_wall_clock_time,
26
+ 0, 0);
27
+
28
+ if (!rc) {
29
+ nd_log(NDLS_DAEMON, NDLP_ERR,
30
+ "PLUGINSD REPLAY ERROR: 'host:%s' failed to initiate replication for 'chart:%s' - replication may not proceed for this instance.",
31
+ rrdhost_hostname(brd->host),
32
+ rrdset_id(brd->st));
33
+ }
34
+
35
+ object_state_release(&brd->host->state_id);
36
+ return rc;
37
+}
38
+
39
+PARSER_RC pluginsd_chart_definition_end(char **words, size_t num_words, PARSER *parser) {
40
+ const char *first_entry_txt = get_word(words, num_words, 1);
41
+ const char *last_entry_txt = get_word(words, num_words, 2);
42
+ const char *wall_clock_time_txt = get_word(words, num_words, 3);
43
+
44
+ RRDHOST *host = pluginsd_require_scope_host(parser, PLUGINSD_KEYWORD_CHART_DEFINITION_END);
45
+ if(!host) return PLUGINSD_DISABLE_PLUGIN(parser, NULL, NULL);
46
+
47
+ RRDSET *st = pluginsd_require_scope_chart(parser, PLUGINSD_KEYWORD_CHART_DEFINITION_END, PLUGINSD_KEYWORD_CHART);
48
+ if(!st) return PLUGINSD_DISABLE_PLUGIN(parser, NULL, NULL);
49
+
50
+ time_t first_entry_child = (first_entry_txt && *first_entry_txt) ? (time_t)str2ul(first_entry_txt) : 0;
51
+ time_t last_entry_child = (last_entry_txt && *last_entry_txt) ? (time_t)str2ul(last_entry_txt) : 0;
52
+ time_t child_wall_clock_time = (wall_clock_time_txt && *wall_clock_time_txt) ? (time_t)str2ul(wall_clock_time_txt) : now_realtime_sec();
53
+
54
+ bool ok = true;
55
+ RRDSET_FLAGS old = rrdset_flag_set_and_clear(
56
+ st, RRDSET_FLAG_RECEIVER_REPLICATION_IN_PROGRESS, RRDSET_FLAG_RECEIVER_REPLICATION_FINISHED);
57
+
58
+ if(!(old & RRDSET_FLAG_RECEIVER_REPLICATION_IN_PROGRESS)) {
59
+ rrdhost_receiver_replicating_charts_plus_one(st->rrdhost);
60
+ __atomic_add_fetch(&host->stream.rcv.status.replication.counter_in, 1, __ATOMIC_RELAXED);
61
+
62
+#ifdef REPLICATION_TRACKING
63
+ st->stream.rcv.who = REPLAY_WHO_ME;
64
+#endif
65
+
66
+#ifdef NETDATA_LOG_REPLICATION_REQUESTS
67
+ st->replay.start_streaming = false;
68
+ st->replay.after = 0;
69
+ st->replay.before = 0;
70
+#endif
71
+
72
+ struct backfill_request_data brd = {
73
+ .host_state_id = object_state_id(&host->state_id),
74
+ .parser = parser,
75
+ .host = host,
76
+ .st = st,
77
+ .first_entry_child = first_entry_child,
78
+ .last_entry_child = last_entry_child,
79
+ .child_wall_clock_time = child_wall_clock_time,
80
+ };
81
+
82
+ __atomic_add_fetch(&host->stream.rcv.status.replication.backfill_pending, 1, __ATOMIC_RELAXED);
83
+
84
+ if(!rrdset_flag_check(st, RRDSET_FLAG_BACKFILLED_HIGH_TIERS)) {
85
+ ok = backfill_request_add(st, backfill_callback, &brd);
86
+ if (!ok)
87
+ ok = backfill_callback(0, 0, &brd);
88
+ else
89
+ rrdset_flag_set(st, RRDSET_FLAG_BACKFILLED_HIGH_TIERS);
90
+ }
91
+ else
92
+ ok = backfill_callback(0, 0, &brd);
93
+ }
94
+ else {
95
+ // this is normal, since dimensions may be added to a chart,
96
+ // and the child will send another CHART_DEFINITION_END command.
97
+
98
+#ifdef NETDATA_LOG_REPLICATION_REQUESTS
99
+ internal_error(true, "REPLAY: 'host:%s/chart:%s' not sending duplicate replication request",
100
+ rrdhost_hostname(st->rrdhost), rrdset_id(st));
101
+#endif
102
+ }
103
+
104
+ stream_thread_received_metadata();
105
+
106
+ return ok ? PARSER_RC_OK : PARSER_RC_ERROR;
107
+}
108
109
PARSER_RC pluginsd_replay_begin(char **words, size_t num_words, PARSER *parser) {
110
int idx = 1;
@@ -89,12 +190,13 @@ PARSER_RC pluginsd_replay_begin(char **words, size_t num_words, PARSER *parser)
190
return PARSER_RC_OK;
191
}
192
92
- netdata_log_error("PLUGINSD REPLAY ERROR: 'host:%s/chart:%s' got a " PLUGINSD_KEYWORD_REPLAY_BEGIN
93
- " from %ld to %ld, but timestamps are invalid "
94
- "(now is %ld [%s], tolerance %ld). Ignoring " PLUGINSD_KEYWORD_REPLAY_SET,
95
- rrdhost_hostname(st->rrdhost), rrdset_id(st), start_time, end_time,
96
- wall_clock_time, wall_clock_comes_from_child ? "child wall clock" : "parent wall clock",
97
- tolerance);
193
+ nd_log(NDLS_DAEMON, NDLP_ERR,
194
+ "PLUGINSD REPLAY ERROR: 'host:%s/chart:%s' got a " PLUGINSD_KEYWORD_REPLAY_BEGIN
195
+ " from %ld to %ld, but timestamps are invalid "
196
+ "(now is %ld [%s], tolerance %ld). Ignoring " PLUGINSD_KEYWORD_REPLAY_SET,
197
+ rrdhost_hostname(st->rrdhost), rrdset_id(st), start_time, end_time,
198
+ wall_clock_time, wall_clock_comes_from_child ? "child wall clock" : "parent wall clock",
199
+ tolerance);
200
}
201
202
// the child sends an RBEGIN without any parameters initially
@@ -127,7 +229,7 @@ PARSER_RC pluginsd_replay_set(char **words, size_t num_words, PARSER *parser) {
229
if(!parser->user.replay.rset_enabled) {
230
nd_log_limit_static_thread_var(erl, 1, 0);
231
nd_log_limit(&erl, NDLS_COLLECTORS, NDLP_ERR,
130
- "PLUGINSD: 'host:%s/chart:%s' got a %s but it is disabled by %s errors",
232
+ "PLUGINSD REPLAY ERROR: 'host:%s/chart:%s' got a %s but it is disabled by %s errors",
233
rrdhost_hostname(host), rrdset_id(st), PLUGINSD_KEYWORD_REPLAY_SET, PLUGINSD_KEYWORD_REPLAY_BEGIN);
234
235
// we have to return OK here
@@ -140,14 +242,13 @@ PARSER_RC pluginsd_replay_set(char **words, size_t num_words, PARSER *parser) {
242
st->pluginsd.set = true;
243
244
if (unlikely(!parser->user.replay.start_time || !parser->user.replay.end_time)) {
143
- netdata_log_error("PLUGINSD: 'host:%s/chart:%s/dim:%s' got a %s with invalid timestamps %ld to %ld from a %s. Disabling it.",
144
- rrdhost_hostname(host),
145
- rrdset_id(st),
146
- dimension,
147
- PLUGINSD_KEYWORD_REPLAY_SET,
148
- parser->user.replay.start_time,
149
- parser->user.replay.end_time,
150
- PLUGINSD_KEYWORD_REPLAY_BEGIN);
245
+
246
+ nd_log(NDLS_DAEMON, NDLP_ERR,
247
+ "PLUGINSD REPLAY ERROR: 'host:%s/chart:%s/dim:%s' got a %s with "
248
+ "invalid timestamps %ld to %ld from a %s. Disabling it.",
249
+ rrdhost_hostname(host), rrdset_id(st), dimension, PLUGINSD_KEYWORD_REPLAY_SET,
250
+ parser->user.replay.start_time, parser->user.replay.end_time, PLUGINSD_KEYWORD_REPLAY_BEGIN);
251
+
252
return PLUGINSD_DISABLE_PLUGIN(parser, NULL, NULL);
253
}
254
@@ -177,7 +278,7 @@ PARSER_RC pluginsd_replay_set(char **words, size_t num_words, PARSER *parser) {
278
else {
279
nd_log_limit_static_global_var(erl, 1, 0);
280
nd_log_limit(&erl, NDLS_COLLECTORS, NDLP_WARNING,
180
- "PLUGINSD: 'host:%s/chart:%s/dim:%s' has the ARCHIVED flag set, but it is replicated. "
281
+ "PLUGINSD REPLAY ERROR: 'host:%s/chart:%s/dim:%s' has the ARCHIVED flag set, but it is replicated. "
282
"Ignoring data.",
283
rrdhost_hostname(st->rrdhost), rrdset_id(st), rrddim_name(rd));
284
}
@@ -219,7 +320,7 @@ PARSER_RC pluginsd_replay_rrddim_collection_state(char **words, size_t num_words
320
usec_t last_collected_ut = last_collected_ut_str ? str2ull_encoded(last_collected_ut_str) : 0;
321
if(last_collected_ut > dim_last_collected_ut) {
322
rd->collector.last_collected_time.tv_sec = (time_t)(last_collected_ut / USEC_PER_SEC);
222
- rd->collector.last_collected_time.tv_usec = (last_collected_ut % USEC_PER_SEC);
323
+ rd->collector.last_collected_time.tv_usec = (suseconds_t)(last_collected_ut % USEC_PER_SEC);
324
}
325
326
rd->collector.last_collected_value = last_collected_value_str ? str2ll_encoded(last_collected_value_str) : 0;
@@ -247,14 +348,14 @@ PARSER_RC pluginsd_replay_rrdset_collection_state(char **words, size_t num_words
348
usec_t last_collected_ut = last_collected_ut_str ? str2ull_encoded(last_collected_ut_str) : 0;
349
if(last_collected_ut > chart_last_collected_ut) {
350
st->last_collected_time.tv_sec = (time_t)(last_collected_ut / USEC_PER_SEC);
250
- st->last_collected_time.tv_usec = (last_collected_ut % USEC_PER_SEC);
351
+ st->last_collected_time.tv_usec = (suseconds_t)(last_collected_ut % USEC_PER_SEC);
352
}
353
354
usec_t chart_last_updated_ut = (usec_t)st->last_updated.tv_sec * USEC_PER_SEC + (usec_t)st->last_updated.tv_usec;
355
usec_t last_updated_ut = last_updated_ut_str ? str2ull_encoded(last_updated_ut_str) : 0;
356
if(last_updated_ut > chart_last_updated_ut) {
357
st->last_updated.tv_sec = (time_t)(last_updated_ut / USEC_PER_SEC);
257
- st->last_updated.tv_usec = (last_updated_ut % USEC_PER_SEC);
358
+ st->last_updated.tv_usec = (suseconds_t)(last_updated_ut % USEC_PER_SEC);
359
}
360
361
st->counter++;
@@ -265,7 +366,7 @@ PARSER_RC pluginsd_replay_rrdset_collection_state(char **words, size_t num_words
366
367
PARSER_RC pluginsd_replay_end(char **words, size_t num_words, PARSER *parser) {
368
if (num_words < 7) { // accepts 7, but the 7th is optional
268
- netdata_log_error("REPLAY: malformed " PLUGINSD_KEYWORD_REPLAY_END " command");
369
+ nd_log(NDLS_DAEMON, NDLP_ERR, "REPLAY: malformed " PLUGINSD_KEYWORD_REPLAY_END " command");
370
return PARSER_RC_ERROR;
371
}
372
@@ -291,6 +392,7 @@ PARSER_RC pluginsd_replay_end(char **words, size_t num_words, PARSER *parser) {
392
393
RRDHOST *host = pluginsd_require_scope_host(parser, PLUGINSD_KEYWORD_REPLAY_END);
394
if(!host) return PLUGINSD_DISABLE_PLUGIN(parser, NULL, NULL);
395
+ __atomic_add_fetch(&host->stream.rcv.status.replication.counter_in, 1, __ATOMIC_RELAXED);
396
397
RRDSET *st = pluginsd_require_scope_chart(parser, PLUGINSD_KEYWORD_REPLAY_END, PLUGINSD_KEYWORD_REPLAY_BEGIN);
398
if(!st) return PLUGINSD_DISABLE_PLUGIN(parser, NULL, NULL);
@@ -310,7 +412,7 @@ PARSER_RC pluginsd_replay_end(char **words, size_t num_words, PARSER *parser) {
412
413
if(parser->user.replay.rset_enabled && st->rrdhost->receiver) {
414
time_t now = now_realtime_sec();
313
- time_t started = st->rrdhost->receiver->replication_first_time_t;
415
+ time_t started = st->rrdhost->receiver->replication.first_time_s;
416
time_t current = parser->user.replay.end_time;
417
418
if(started && current > started) {
@@ -340,20 +442,25 @@ PARSER_RC pluginsd_replay_end(char **words, size_t num_words, PARSER *parser) {
442
#endif
443
444
if (start_streaming) {
445
+#ifdef REPLICATION_TRACKING
446
+ st->stream.rcv.who = REPLAY_WHO_FINISHED;
447
+#endif
448
+
449
if (st->update_every != update_every_child)
450
rrdset_set_update_every_s(st, update_every_child);
451
346
- if(rrdset_flag_check(st, RRDSET_FLAG_RECEIVER_REPLICATION_IN_PROGRESS)) {
347
- rrdset_flag_set(st, RRDSET_FLAG_RECEIVER_REPLICATION_FINISHED);
348
- rrdset_flag_clear(st, RRDSET_FLAG_RECEIVER_REPLICATION_IN_PROGRESS);
349
- rrdset_flag_clear(st, RRDSET_FLAG_SYNC_CLOCK);
452
+ RRDSET_FLAGS old = rrdset_flag_set_and_clear(
453
+ st, RRDSET_FLAG_RECEIVER_REPLICATION_FINISHED,
454
+ RRDSET_FLAG_RECEIVER_REPLICATION_IN_PROGRESS | RRDSET_FLAG_SYNC_CLOCK);
455
+
456
+ if(!(old & RRDSET_FLAG_RECEIVER_REPLICATION_FINISHED))
457
rrdhost_receiver_replicating_charts_minus_one(st->rrdhost);
351
- }
352
-#ifdef NETDATA_LOG_REPLICATION_REQUESTS
458
+
459
else
354
- internal_error(true, "REPLAY ERROR: 'host:%s/chart:%s' got a " PLUGINSD_KEYWORD_REPLAY_END " with enable_streaming = true, but there is no replication in progress for this chart.",
355
- rrdhost_hostname(host), rrdset_id(st));
356
-#endif
460
+ nd_log(NDLS_DAEMON, NDLP_WARNING,
461
+ "PLUGINSD REPLAY ERROR: 'host:%s/chart:%s' got a " PLUGINSD_KEYWORD_REPLAY_END " "
462
+ "with enable_streaming = true, but there was no replication in progress for this chart.",
463
+ rrdhost_hostname(host), rrdset_id(st));
464
465
pluginsd_clear_scope_chart(parser, PLUGINSD_KEYWORD_REPLAY_END);
466
@@ -365,6 +472,10 @@ PARSER_RC pluginsd_replay_end(char **words, size_t num_words, PARSER *parser) {
472
return PARSER_RC_OK;
473
}
474
475
+#ifdef REPLICATION_TRACKING
476
+ st->stream.rcv.who = REPLAY_WHO_ME;
477
+#endif
478
+
479
pluginsd_clear_scope_chart(parser, PLUGINSD_KEYWORD_REPLAY_END);
480
481
rrdcontext_updated_retention_rrdset(st);
@@ -372,5 +483,6 @@ PARSER_RC pluginsd_replay_end(char **words, size_t num_words, PARSER *parser) {
483
bool ok = replicate_chart_request(send_to_plugin, parser, host, st,
484
first_entry_child, last_entry_child, child_world_time,
485
first_entry_requested, last_entry_requested);
486
+
487
return ok ? PARSER_RC_OK : PARSER_RC_ERROR;
488
}
src/plugins.d/pluginsd_replication.h
+1
@@ -10,5 +10,6 @@ PARSER_RC pluginsd_replay_set(char **words, size_t num_words, PARSER *parser);
10
PARSER_RC pluginsd_replay_rrddim_collection_state(char **words, size_t num_words, PARSER *parser);
11
PARSER_RC pluginsd_replay_rrdset_collection_state(char **words, size_t num_words, PARSER *parser);
12
PARSER_RC pluginsd_replay_end(char **words, size_t num_words, PARSER *parser);
13
+PARSER_RC pluginsd_chart_definition_end(char **words, size_t num_words, PARSER *parser);
14
15
#endif //NETDATA_PLUGINSD_REPLICATION_H
src/registry/registry_init.c
+4
-4
@@ -149,28 +149,28 @@ int registry_init(void) {
149
®istry.aral_stats,
150
"registry_persons",
151
&netdata_configured_cache_dir,
152
- use_mmap, true);
152
+ use_mmap, true, true);
153
154
registry.machines_aral = aral_create("registry_machines", sizeof(REGISTRY_MACHINE),
155
min_page_size / sizeof(REGISTRY_MACHINE), max_page_size,
156
®istry.aral_stats,
157
"registry_machines",
158
&netdata_configured_cache_dir,
159
- use_mmap, true);
159
+ use_mmap, true, true);
160
161
registry.person_urls_aral = aral_create("registry_person_urls", sizeof(REGISTRY_PERSON_URL),
162
min_page_size / sizeof(REGISTRY_PERSON_URL), max_page_size,
163
®istry.aral_stats,
164
"registry_person_urls",
165
&netdata_configured_cache_dir,
166
- use_mmap, true);
166
+ use_mmap, true, true);
167
168
registry.machine_urls_aral = aral_create("registry_machine_urls", sizeof(REGISTRY_MACHINE_URL),
169
min_page_size / sizeof(REGISTRY_MACHINE_URL), max_page_size,
170
®istry.aral_stats,
171
"registry_machine_urls",
172
&netdata_configured_cache_dir,
173
- use_mmap, true);
173
+ use_mmap, true, true);
174
175
registry_log_open();
176
registry_db_load();
src/streaming/protocol/command-chart-definition.c
+3
-4
@@ -116,11 +116,10 @@ bool stream_sender_send_rrdset_definition(BUFFER *wb, RRDSET *st) {
116
(unsigned long long)db_last_time_t,
117
(unsigned long long)now);
118
119
- if(!rrdset_flag_check(st, RRDSET_FLAG_SENDER_REPLICATION_IN_PROGRESS)) {
120
- rrdset_flag_set(st, RRDSET_FLAG_SENDER_REPLICATION_IN_PROGRESS);
121
- rrdset_flag_clear(st, RRDSET_FLAG_SENDER_REPLICATION_FINISHED);
119
+ RRDSET_FLAGS old = rrdset_flag_set_and_clear(st, RRDSET_FLAG_SENDER_REPLICATION_IN_PROGRESS, RRDSET_FLAG_SENDER_REPLICATION_FINISHED);
120
+ if(!(old & RRDSET_FLAG_SENDER_REPLICATION_IN_PROGRESS))
121
rrdhost_sender_replicating_charts_plus_one(st->rrdhost);
123
- }
122
+
123
replication_progress = true;
124
125
#ifdef NETDATA_LOG_REPLICATION_REQUESTS
src/streaming/protocol/command-nodeid.c
+6
-6
@@ -52,7 +52,7 @@ void stream_sender_get_node_and_claim_id_from_parent(struct sender_state *s) {
52
if (uuid_parse(claim_id_str ? claim_id_str : "", claim_id.uuid) != 0) {
53
nd_log(NDLS_DAEMON, NDLP_ERR,
54
"STREAM SND '%s' [to %s] received invalid claim id '%s'",
55
- rrdhost_hostname(s->host), s->connected_to,
55
+ rrdhost_hostname(s->host), s->remote_ip,
56
claim_id_str ? claim_id_str : "(unset)");
57
return;
58
}
@@ -61,7 +61,7 @@ void stream_sender_get_node_and_claim_id_from_parent(struct sender_state *s) {
61
if(uuid_parse(node_id_str ? node_id_str : "", node_id.uuid) != 0) {
62
nd_log(NDLS_DAEMON, NDLP_ERR,
63
"STREAM SND '%s' [to %s] received an invalid node id '%s'",
64
- rrdhost_hostname(s->host), s->connected_to,
64
+ rrdhost_hostname(s->host), s->remote_ip,
65
node_id_str ? node_id_str : "(unset)");
66
return;
67
}
@@ -69,14 +69,14 @@ void stream_sender_get_node_and_claim_id_from_parent(struct sender_state *s) {
69
if (!UUIDiszero(s->host->aclk.claim_id_of_parent) && !UUIDeq(s->host->aclk.claim_id_of_parent, claim_id))
70
nd_log(NDLS_DAEMON, NDLP_INFO,
71
"STREAM SND '%s' [to %s] changed parent's claim id to %s",
72
- rrdhost_hostname(s->host), s->connected_to,
72
+ rrdhost_hostname(s->host), s->remote_ip,
73
claim_id_str ? claim_id_str : "(unset)");
74
75
if(!UUIDiszero(s->host->node_id) && !UUIDeq(s->host->node_id, node_id)) {
76
if(claimed) {
77
nd_log(NDLS_DAEMON, NDLP_WARNING,
78
"STREAM SND '%s' [to %s] parent reports different node id '%s', but we are claimed. Ignoring it.",
79
- rrdhost_hostname(s->host), s->connected_to,
79
+ rrdhost_hostname(s->host), s->remote_ip,
80
node_id_str ? node_id_str : "(unset)");
81
return;
82
}
@@ -84,7 +84,7 @@ void stream_sender_get_node_and_claim_id_from_parent(struct sender_state *s) {
84
update_node_id = true;
85
nd_log(NDLS_DAEMON, NDLP_WARNING,
86
"STREAM SND '%s' [to %s] changed node id to %s",
87
- rrdhost_hostname(s->host), s->connected_to,
87
+ rrdhost_hostname(s->host), s->remote_ip,
88
node_id_str ? node_id_str : "(unset)");
89
}
90
}
@@ -92,7 +92,7 @@ void stream_sender_get_node_and_claim_id_from_parent(struct sender_state *s) {
92
if(!url || !*url) {
93
nd_log(NDLS_DAEMON, NDLP_ERR,
94
"STREAM SND '%s' [to %s] received an invalid cloud URL '%s'",
95
- rrdhost_hostname(s->host), s->connected_to,
95
+ rrdhost_hostname(s->host), s->remote_ip,
96
url ? url : "(unset)");
97
return;
98
}
src/streaming/replication.h
deleted
-47
@@ -1,47 +0,0 @@
1
-// SPDX-License-Identifier: GPL-3.0-or-later
2
-
3
-#ifndef REPLICATION_H
4
-#define REPLICATION_H
5
-
6
-#include "daemon/common.h"
7
-#include "stream-circular-buffer.h"
8
-
9
-#ifdef __cplusplus
10
-extern "C" {
11
-#endif
12
-
13
-struct parser;
14
-
15
-struct replication_query_statistics {
16
- SPINLOCK spinlock;
17
- size_t queries_started;
18
- size_t queries_finished;
19
- size_t points_read;
20
- size_t points_generated;
21
-};
22
-
23
-struct replication_query_statistics replication_get_query_statistics(void);
24
-
25
-bool replicate_chart_response(RRDHOST *rh, RRDSET *rs, bool start_streaming, time_t after, time_t before);
26
-
27
-typedef ssize_t (*send_command)(const char *txt, struct parser *parser, STREAM_TRAFFIC_TYPE type);
28
-
29
-bool replicate_chart_request(send_command callback, struct parser *parser,
30
- RRDHOST *rh, RRDSET *rs,
31
- time_t child_first_entry, time_t child_last_entry, time_t child_wall_clock_time,
32
- time_t response_first_start_time, time_t response_last_end_time);
33
-
34
-void replication_sender_init(struct sender_state *sender);
35
-void replication_cleanup_sender(struct sender_state *sender);
36
-void replication_sender_delete_pending_requests(struct sender_state *sender);
37
-void replication_add_request(struct sender_state *sender, const char *chart_id, time_t after, time_t before, bool start_streaming);
38
-void replication_recalculate_buffer_used_ratio_unsafe(struct sender_state *s);
39
-
40
-size_t replication_allocated_memory(void);
41
-size_t replication_allocated_buffers(void);
42
-
43
-#ifdef __cplusplus
44
-}
45
-#endif
46
-
47
-#endif /* REPLICATION_H */
src/streaming/stream-capabilities.c
+4
-2
@@ -81,7 +81,7 @@ void log_receiver_capabilities(struct receiver_state *rpt) {
81
stream_capabilities_to_string(wb, rpt->capabilities);
82
83
nd_log_daemon(NDLP_INFO, "STREAM RCV '%s' [from [%s]:%s]: established link with negotiated capabilities: %s",
84
- rrdhost_hostname(rpt->host), rpt->client_ip, rpt->client_port, buffer_tostring(wb));
84
+ rrdhost_hostname(rpt->host), rpt->remote_ip, rpt->remote_port, buffer_tostring(wb));
85
86
buffer_free(wb);
87
}
@@ -91,7 +91,7 @@ void log_sender_capabilities(struct sender_state *s) {
91
stream_capabilities_to_string(wb, s->capabilities);
92
93
nd_log_daemon(NDLP_INFO, "STREAM SND '%s' [to %s]: established link with negotiated capabilities: %s",
94
- rrdhost_hostname(s->host), s->connected_to, buffer_tostring(wb));
94
+ rrdhost_hostname(s->host), s->remote_ip, buffer_tostring(wb));
95
96
buffer_free(wb);
97
}
@@ -174,4 +174,6 @@ void check_local_streaming_capabilities(void) {
174
ieee754_doubles = is_system_ieee754_double();
175
if(!ieee754_doubles)
176
globally_disabled_capabilities |= STREAM_CAP_IEEE754;
177
+ else
178
+ globally_disabled_capabilities &= ~STREAM_CAP_IEEE754;
179
}
src/streaming/stream-circular-buffer.c
+21
-9
@@ -6,7 +6,9 @@
6
struct stream_circular_buffer {
7
struct circular_buffer *cb;
8
STREAM_CIRCULAR_BUFFER_STATS stats;
9
- usec_t last_reset_ut;
9
+
10
+ usec_t last_recreate_ut; // recreates are only used to shrink the buffer, they are normal during operation
11
+ usec_t last_sent_ut; // the last time we removed or flushed data from the buffer
12
13
struct {
14
// the current max size of the buffer
@@ -17,7 +19,7 @@ struct stream_circular_buffer {
19
20
// the last time we flushed the buffer
21
// by monitoring this we can know if the system was reconnected
20
- usec_t since_ut;
22
+ usec_t last_flush_ut;
23
} atomic;
24
};
25
@@ -52,9 +54,11 @@ bool stream_circular_buffer_set_max_size_unsafe(STREAM_CIRCULAR_BUFFER *scb, siz
54
}
55
56
void stream_circular_buffer_flush_unsafe(STREAM_CIRCULAR_BUFFER *scb, size_t buffer_max_size) {
55
- __atomic_store_n(&((scb)->atomic.since_ut), now_monotonic_usec(), __ATOMIC_RELAXED);
57
+ usec_t now_ut = now_monotonic_usec();
58
+ __atomic_store_n(&scb->atomic.last_flush_ut, now_ut, __ATOMIC_RELAXED);
59
60
// flush the output buffer from any data it may have
61
+ scb->last_sent_ut = now_ut;
62
cbuffer_flush(scb->cb);
63
memset(&scb->stats, 0, sizeof(scb->stats));
64
stream_circular_buffer_set_max_size_unsafe(scb, buffer_max_size, true);
@@ -62,7 +66,7 @@ void stream_circular_buffer_flush_unsafe(STREAM_CIRCULAR_BUFFER *scb, size_t buf
66
}
67
68
inline size_t stream_sender_get_buffer_used_percent(STREAM_CIRCULAR_BUFFER *scb) {
65
- return __atomic_load_n(&((scb)->atomic.buffer_ratio), __ATOMIC_RELAXED);
69
+ return __atomic_load_n(&scb->atomic.buffer_ratio, __ATOMIC_RELAXED);
70
}
71
72
size_t stream_circular_buffer_get_max_size(STREAM_CIRCULAR_BUFFER *scb) {
@@ -70,10 +74,10 @@ size_t stream_circular_buffer_get_max_size(STREAM_CIRCULAR_BUFFER *scb) {
74
}
75
76
void stream_circular_buffer_recreate_timed_unsafe(STREAM_CIRCULAR_BUFFER *scb, usec_t now_ut, bool force) {
73
- if(!force && (scb->stats.bytes_outstanding || now_ut - scb->last_reset_ut < 300 * USEC_PER_SEC))
77
+ if(!force && (scb->stats.bytes_outstanding || now_ut - scb->last_recreate_ut < 300 * USEC_PER_SEC))
78
return;
79
76
- scb->last_reset_ut = now_ut;
80
+ scb->last_recreate_ut = now_ut;
81
82
scb->stats.recreates++; // we increase even if we don't do it, to have sender_start() recreate its buffers
83
@@ -84,11 +88,18 @@ void stream_circular_buffer_recreate_timed_unsafe(STREAM_CIRCULAR_BUFFER *scb, u
88
}
89
}
90
87
-inline usec_t stream_circular_buffer_get_since_ut(STREAM_CIRCULAR_BUFFER *scb) {
88
- return __atomic_load_n(&((scb)->atomic.since_ut), __ATOMIC_RELAXED);
91
+inline usec_t stream_circular_buffer_last_flush_ut(STREAM_CIRCULAR_BUFFER *scb) {
92
+ return __atomic_load_n(&((scb)->atomic.last_flush_ut), __ATOMIC_RELAXED);
93
+}
94
+
95
+inline usec_t stream_circular_buffer_last_sent_ut(STREAM_CIRCULAR_BUFFER *scb) {
96
+ // this is ok without locks and atomics, since only the stream threads
97
+ // can actually remove data and call this
98
+ return scb->last_sent_ut;
99
}
100
101
void stream_circular_buffer_destroy(STREAM_CIRCULAR_BUFFER *scb) {
102
+ if(!scb) return;
103
cbuffer_free(scb->cb);
104
freez(scb);
105
}
@@ -118,7 +129,8 @@ size_t stream_circular_buffer_get_unsafe(STREAM_CIRCULAR_BUFFER *scb, char **chu
129
}
130
131
// removes data from the beginning of the circular buffer
121
-void stream_circular_buffer_del_unsafe(STREAM_CIRCULAR_BUFFER *scb, size_t bytes) {
132
+void stream_circular_buffer_del_unsafe(STREAM_CIRCULAR_BUFFER *scb, size_t bytes, usec_t now_ut) {
133
+ scb->last_sent_ut = now_ut ? now_ut : now_monotonic_usec();
134
scb->stats.sends++;
135
scb->stats.bytes_sent += bytes;
136
cbuffer_remove_unsafe(scb->cb, bytes);
src/streaming/stream-circular-buffer.h
+5
-2
@@ -70,7 +70,10 @@ size_t stream_circular_buffer_get_max_size(STREAM_CIRCULAR_BUFFER *scb);
70
size_t stream_sender_get_buffer_used_percent(STREAM_CIRCULAR_BUFFER *scb);
71
72
// return the monotonic timestamp of the last time the buffer was created
73
-usec_t stream_circular_buffer_get_since_ut(STREAM_CIRCULAR_BUFFER *scb);
73
+usec_t stream_circular_buffer_last_flush_ut(STREAM_CIRCULAR_BUFFER *scb);
74
+
75
+// return the monotonic timestamp of the last time we removed data from the buffer
76
+usec_t stream_circular_buffer_last_sent_ut(STREAM_CIRCULAR_BUFFER *scb);
77
78
// --------------------------------------------------------------------------------------------------------------------
79
// data operations (add, get, remove data from/to the buffer)
@@ -86,7 +89,7 @@ size_t stream_circular_buffer_get_unsafe(STREAM_CIRCULAR_BUFFER *scb, char **chu
89
90
// removes data from the beginning of circular buffer
91
// it updates the statistics
89
-void stream_circular_buffer_del_unsafe(STREAM_CIRCULAR_BUFFER *scb, size_t bytes);
92
+void stream_circular_buffer_del_unsafe(STREAM_CIRCULAR_BUFFER *scb, size_t bytes, usec_t now_ut);
93
94
#ifdef __cplusplus
95
}
src/streaming/stream-conf.c
+9
-5
@@ -90,11 +90,12 @@ bool stream_conf_receiver_needs_dbengine(void) {
90
}
91
92
void stream_conf_load() {
93
- static bool loaded = false;
94
- if(loaded) return;
95
- loaded = true;
93
+ static bool run = false;
94
+ if(run) return;
95
+ run = true;
96
97
stream_conf_load_internal();
98
+ check_local_streaming_capabilities();
99
100
stream_send.enabled =
101
appconfig_get_boolean(&stream_config, CONFIG_SECTION_STREAM, "enabled", stream_send.enabled);
@@ -199,6 +200,10 @@ bool stream_conf_is_parent(bool recheck) {
200
return rc;
201
}
202
203
+bool stream_conf_is_child(void) {
204
+ return stream_send.enabled;
205
+}
206
+
207
void stream_conf_receiver_config(struct receiver_state *rpt, struct stream_receiver_config *config, const char *api_key, const char *machine_guid) {
208
config->mode = rrd_memory_mode_id(
209
appconfig_get(&stream_config, machine_guid, "db",
@@ -209,8 +214,7 @@ void stream_conf_receiver_config(struct receiver_state *rpt, struct stream_recei
214
netdata_log_error("STREAM RCV '%s' [from [%s]:%s]: "
215
"dbengine is not enabled, falling back to default."
216
, rpt->hostname
212
- , rpt->client_ip, rpt->client_port
213
- );
217
+ , rpt->remote_ip, rpt->remote_port);
218
config->mode = default_rrd_memory_mode;
219
}
220
src/streaming/stream-conf.h
+1
@@ -86,6 +86,7 @@ void stream_conf_receiver_config(struct receiver_state *rpt, struct stream_recei
86
void stream_conf_load();
87
bool stream_conf_receiver_needs_dbengine();
88
bool stream_conf_is_parent(bool recheck);
89
+bool stream_conf_is_child(void);
90
91
bool stream_conf_is_key_type(const char *api_key, const char *type);
92
bool stream_conf_api_key_is_enabled(const char *api_key, bool enabled);
src/streaming/stream-connector.c
+10
-18
@@ -261,7 +261,7 @@ stream_connect_validate_first_response(RRDHOST *host, struct sender_state *s, ch
261
262
nd_log(NDLS_DAEMON, priority,
263
"STREAM CONNECT '%s' [to %s]: %s - will retry in %d secs, at %s",
264
- rrdhost_hostname(host), s->connected_to, error, delay, buf);
264
+ rrdhost_hostname(host), s->remote_ip, error, delay, buf);
265
266
return false;
267
}
@@ -282,7 +282,7 @@ bool stream_connect(struct sender_state *s, uint16_t default_port, time_t timeou
282
283
if(!stream_parent_connect_to_one(
284
&s->sock, host, default_port, timeout,
285
- s->connected_to, sizeof(s->connected_to) - 1,
285
+ s->remote_ip, sizeof(s->remote_ip) - 1,
286
&host->stream.snd.parents.current)) {
287
288
if(s->sock.error != ND_SOCK_ERR_NO_DESTINATION_AVAILABLE)
@@ -305,7 +305,7 @@ bool stream_connect(struct sender_state *s, uint16_t default_port, time_t timeou
305
buffer_key_value_urlencode(wb, "&hostname", rrdhost_hostname(host));
306
buffer_key_value_urlencode(wb, "®istry_hostname", rrdhost_registry_hostname(host));
307
buffer_key_value_urlencode(wb, "&machine_guid", host->machine_guid);
308
- buffer_sprintf(wb, "&update_every=%d", default_rrd_update_every);
308
+ buffer_sprintf(wb, "&update_every=%d", (int)nd_profile.update_every);
309
buffer_key_value_urlencode(wb, "&os", rrdhost_os(host));
310
buffer_key_value_urlencode(wb, "&timezone", rrdhost_timezone(host));
311
buffer_key_value_urlencode(wb, "&abbrev_timezone", rrdhost_abbrev_timezone(host));
@@ -346,7 +346,7 @@ bool stream_connect(struct sender_state *s, uint16_t default_port, time_t timeou
346
347
nd_log(NDLS_DAEMON, NDLP_ERR,
348
"STREAM CONNECT '%s' [to %s]: failed to send HTTP header to remote netdata.",
349
- rrdhost_hostname(host), s->connected_to);
349
+ rrdhost_hostname(host), s->remote_ip);
350
351
stream_parent_set_reconnect_delay(
352
host->stream.snd.parents.current, STREAM_HANDSHAKE_ERROR_SEND_TIMEOUT, 60);
@@ -368,7 +368,7 @@ bool stream_connect(struct sender_state *s, uint16_t default_port, time_t timeou
368
369
nd_log(NDLS_DAEMON, NDLP_ERR,
370
"STREAM CONNECT '%s' [to %s]: remote netdata does not respond.",
371
- rrdhost_hostname(host), s->connected_to);
371
+ rrdhost_hostname(host), s->remote_ip);
372
373
stream_parent_set_reconnect_delay(
374
host->stream.snd.parents.current, STREAM_HANDSHAKE_ERROR_RECEIVE_TIMEOUT, 30);
@@ -377,18 +377,6 @@ bool stream_connect(struct sender_state *s, uint16_t default_port, time_t timeou
377
}
378
response[bytes] = '\0';
379
380
- if(sock_setnonblock(s->sock.fd) < 0)
381
- nd_log(NDLS_DAEMON, NDLP_WARNING,
382
- "STREAM CONNECT '%s' [to %s]: cannot set non-blocking mode for socket.",
383
- rrdhost_hostname(host), s->connected_to);
384
-
385
- sock_setcloexec(s->sock.fd);
386
-
387
- if(sock_enlarge_out(s->sock.fd) < 0)
388
- nd_log(NDLS_DAEMON, NDLP_WARNING,
389
- "STREAM CONNECT '%s' [to %s]: cannot enlarge the socket buffer.",
390
- rrdhost_hostname(host), s->connected_to);
391
-
380
if(!stream_connect_validate_first_response(host, s, response, bytes)) {
381
nd_sock_close(&s->sock);
382
return false;
@@ -406,7 +394,7 @@ bool stream_connect(struct sender_state *s, uint16_t default_port, time_t timeou
394
395
nd_log(NDLS_DAEMON, NDLP_DEBUG,
396
"STREAM CONNECT '%s' [to %s]: connected to parent...",
409
- rrdhost_hostname(host), s->connected_to);
397
+ rrdhost_hostname(host), s->remote_ip);
398
399
return true;
400
}
@@ -512,6 +500,10 @@ static void stream_connector_remove(struct sender_state *s) {
500
struct connector *sc = stream_connector_get(s);
501
__atomic_sub_fetch(&sc->nodes, 1, __ATOMIC_RELAXED);
502
503
+ nd_log(NDLS_DAEMON, NDLP_NOTICE,
504
+ "STREAM CNT '%s' [to %s]: streaming connector removed host: %s (signaled to stop)",
505
+ rrdhost_hostname(s->host), s->remote_ip, stream_handshake_error_to_string(s->exit.reason));
506
+
507
stream_sender_remove(s);
508
}
509
src/streaming/stream-control.c
+5
-18
@@ -2,26 +2,13 @@
2
3
#include "stream-control.h"
4
#include "stream.h"
5
-#include "replication.h"
5
+#include "stream-replication-sender.h"
6
7
static struct {
8
- CACHE_LINE_PADDING();
9
-
10
- uint32_t backfill_runners;
11
-
12
- CACHE_LINE_PADDING();
13
-
14
- uint32_t replication_runners;
15
-
16
- CACHE_LINE_PADDING();
17
-
18
- uint32_t user_data_queries_runners;
19
-
20
- CACHE_LINE_PADDING();
21
-
22
- uint32_t user_weights_queries_runners;
23
-
24
- CACHE_LINE_PADDING();
8
+ PAD64(uint32_t) backfill_runners;
9
+ PAD64(uint32_t) replication_runners;
10
+ PAD64(uint32_t) user_data_queries_runners;
11
+ PAD64(uint32_t) user_weights_queries_runners;
12
} sc;
13
14
// --------------------------------------------------------------------------------------------------------------------
src/streaming/stream-handshake.c
+1
@@ -51,6 +51,7 @@ static struct {
51
{STREAM_HANDSHAKE_CONNECTED, "CONNECTED"},
52
{STREAM_HANDSHAKE_EXITING, "EXITING"},
53
{STREAM_HANDSHAKE_NO_STREAM_INFO, "NO STREAM INFO"},
54
+ {STREAM_HANDSHAKE_REPLICATION_STALLED, "REPLICATION STALLED"},
55
{ 0, NULL },
56
};
57
src/streaming/stream-handshake.h
+1
@@ -88,6 +88,7 @@ typedef enum {
88
STREAM_HANDSHAKE_CONNECTED = -39,
89
STREAM_HANDSHAKE_EXITING = -40,
90
STREAM_HANDSHAKE_NO_STREAM_INFO = -41,
91
+ STREAM_HANDSHAKE_REPLICATION_STALLED = -42,
92
93
} STREAM_HANDSHAKE;
94
src/streaming/stream-parents.c
+3
-3
@@ -1,7 +1,7 @@
1
// SPDX-License-Identifier: GPL-3.0-or-later
2
3
#include "stream-sender-internals.h"
4
-#include "replication.h"
4
+#include "stream-replication-sender.h"
5
6
#define TIME_TO_CONSIDER_PARENTS_SIMILAR 120
7
@@ -607,7 +607,7 @@ bool stream_parent_connect_to_one_unsafe(
607
d->since_ut = now_ut;
608
d->banned_permanently = true;
609
skipped_not_useful++;
610
- nd_log(NDLS_DAEMON, NDLP_NOTICE,
610
+ nd_log(NDLS_DAEMON, NDLP_INFO,
611
"STREAM PARENTS '%s': destination '%s' is banned permanently because it is the origin server",
612
rrdhost_hostname(host), string2str(d->destination));
613
continue;
@@ -635,7 +635,7 @@ bool stream_parent_connect_to_one_unsafe(
635
d->since_ut = now_ut;
636
d->banned_for_this_session = true;
637
skipped_not_useful++;
638
- nd_log(NDLS_DAEMON, NDLP_NOTICE,
638
+ nd_log(NDLS_DAEMON, NDLP_INFO,
639
"STREAM PARENTS '%s': destination '%s' is banned for this session, because it is in our path before us.",
640
rrdhost_hostname(host), string2str(d->destination));
641
continue;
src/streaming/stream-receiver-connection.c
+38
-35
@@ -4,7 +4,7 @@
4
#include "stream-thread.h"
5
#include "stream-receiver-internals.h"
6
#include "web/server/h2o/http_server.h"
7
-#include "replication.h"
7
+#include "stream-replication-sender.h"
8
9
// --------------------------------------------------------------------------------------------------------------------
10
@@ -12,8 +12,8 @@ void stream_receiver_log_status(struct receiver_state *rpt, const char *msg, con
12
// this function may be called BEFORE we spawn the receiver thread
13
// so, we need to add the fields again (it does not harm)
14
ND_LOG_STACK lgs[] = {
15
- ND_LOG_FIELD_TXT(NDF_SRC_IP, rpt->client_ip),
16
- ND_LOG_FIELD_TXT(NDF_SRC_PORT, rpt->client_port),
15
+ ND_LOG_FIELD_TXT(NDF_SRC_IP, rpt->remote_ip),
16
+ ND_LOG_FIELD_TXT(NDF_SRC_PORT, rpt->remote_port),
17
ND_LOG_FIELD_TXT(NDF_NIDL_NODE, (rpt->hostname && *rpt->hostname) ? rpt->hostname : ""),
18
ND_LOG_FIELD_TXT(NDF_RESPONSE_CODE, status),
19
ND_LOG_FIELD_UUID(NDF_MESSAGE_ID, &streaming_from_child_msgid),
@@ -28,8 +28,7 @@ void stream_receiver_log_status(struct receiver_state *rpt, const char *msg, con
28
29
nd_log(NDLS_DAEMON, priority, "STREAM RCV '%s' [from [%s]:%s]: %s %s%s%s"
30
, (rpt->hostname && *rpt->hostname) ? rpt->hostname : ""
31
- , rpt->client_ip, rpt->client_port
32
- , msg
31
+ , rpt->remote_ip, rpt->remote_port, msg
32
, rpt->exit.reason != STREAM_HANDSHAKE_NEVER?" (":""
33
, stream_handshake_error_to_string(rpt->exit.reason)
34
, rpt->exit.reason != STREAM_HANDSHAKE_NEVER?")":""
@@ -54,18 +53,31 @@ void stream_receiver_free(struct receiver_state *rpt) {
53
freez(rpt->os);
54
freez(rpt->timezone);
55
freez(rpt->abbrev_timezone);
57
- freez(rpt->client_ip);
58
- freez(rpt->client_port);
56
+ freez(rpt->remote_ip);
57
+ freez(rpt->remote_port);
58
freez(rpt->program_name);
59
freez(rpt->program_version);
60
61
+ string_freez(rpt->config.send.api_key);
62
+ string_freez(rpt->config.send.parents);
63
+ string_freez(rpt->config.send.charts_matching);
64
+
65
+ buffer_free(rpt->thread.line_buffer);
66
+ rpt->thread.line_buffer = NULL;
67
+
68
freez(rpt->thread.compressed.buf);
69
rpt->thread.compressed.buf = NULL;
70
rpt->thread.compressed.size = 0;
71
66
- string_freez(rpt->config.send.api_key);
67
- string_freez(rpt->config.send.parents);
68
- string_freez(rpt->config.send.charts_matching);
72
+ rpt->thread.send_to_child.msg.session = 0;
73
+ rpt->thread.send_to_child.msg.meta = NULL;
74
+ stream_circular_buffer_destroy(rpt->thread.send_to_child.scb);
75
+ rpt->thread.send_to_child.scb = NULL;
76
+
77
+#ifdef NETDATA_LOG_STREAM_RECEIVER
78
+ if(rpt->log.fp)
79
+ fclose(rpt->log.fp);
80
+#endif
81
82
freez(rpt);
83
}
@@ -89,7 +101,7 @@ static int stream_receiver_response_too_busy_now(struct web_client *w) {
101
}
102
103
static void stream_receiver_takeover_web_connection(struct web_client *w, struct receiver_state *rpt) {
92
- rpt->sock.fd = w->ifd;
104
+ rpt->sock.fd = w->fd;
105
rpt->sock.ssl = w->ssl;
106
107
w->ssl = NETDATA_SSL_UNSET_CONNECTION;
@@ -99,12 +111,8 @@ static void stream_receiver_takeover_web_connection(struct web_client *w, struct
111
if(web_server_mode == WEB_SERVER_MODE_STATIC_THREADED) {
112
web_client_flag_set(w, WEB_CLIENT_FLAG_DONT_CLOSE_SOCKET);
113
}
102
- else {
103
- if(w->ifd == w->ofd)
104
- w->ifd = w->ofd = -1;
105
- else
106
- w->ifd = -1;
107
- }
114
+ else
115
+ w->fd = -1;
116
117
buffer_flush(w->response.data);
118
}
@@ -189,9 +197,7 @@ static bool stream_receiver_send_first_response(struct receiver_state *rpt) {
197
"client willing to stream metrics for host '%s' with machine_guid '%s': "
198
"update every = %d, history = %d, memory mode = %s, health %s,%s"
199
, rpt->hostname
192
- , rpt->client_ip
193
- , rpt->client_port
194
- , rrdhost_hostname(rpt->host)
200
+ , rpt->remote_ip, rpt->remote_port, rrdhost_hostname(rpt->host)
201
, rpt->host->machine_guid
202
, rpt->host->rrd_update_every
203
, rpt->host->rrd_history_entries
@@ -231,10 +237,10 @@ static bool stream_receiver_send_first_response(struct receiver_state *rpt) {
237
#endif
238
{
239
// remove the non-blocking flag from the socket
234
- if(sock_delnonblock(rpt->sock.fd) < 0)
240
+ if(sock_setnonblock(rpt->sock.fd, false) != 0)
241
nd_log(NDLS_DAEMON, NDLP_ERR,
242
"STREAM RCV '%s' [from [%s]:%s]: cannot remove the non-blocking flag from socket %d",
237
- rrdhost_hostname(rpt->host), rpt->client_ip, rpt->client_port, rpt->sock.fd);
243
+ rrdhost_hostname(rpt->host), rpt->remote_ip, rpt->remote_port, rpt->sock.fd);
244
245
struct timeval timeout;
246
timeout.tv_sec = 600;
@@ -242,10 +248,10 @@ static bool stream_receiver_send_first_response(struct receiver_state *rpt) {
248
if (unlikely(setsockopt(rpt->sock.fd, SOL_SOCKET, SO_RCVTIMEO, &timeout, sizeof timeout) != 0))
249
nd_log(NDLS_DAEMON, NDLP_ERR,
250
"STREAM RCV '%s' [from [%s]:%s]: cannot set timeout for socket %d",
245
- rrdhost_hostname(rpt->host), rpt->client_ip, rpt->client_port, rpt->sock.fd);
251
+ rrdhost_hostname(rpt->host), rpt->remote_ip, rpt->remote_port, rpt->sock.fd);
252
}
253
248
- netdata_log_debug(D_STREAM, "Initial response to %s: %s", rpt->client_ip, initial_response);
254
+ netdata_log_debug(D_STREAM, "Initial response to %s: %s", rpt->remote_ip, initial_response);
255
#ifdef ENABLE_H2O
256
if (is_h2o_rrdpush(rpt)) {
257
h2o_stream_write(rpt->h2o_ctx, initial_response, strlen(initial_response));
@@ -279,7 +285,7 @@ int stream_receiver_accept_connection(struct web_client *w, char *decoded_query_
285
rpt->thread.compressed.size = COMPRESSION_MAX_CHUNK;
286
rpt->thread.compressed.buf = mallocz(rpt->thread.compressed.size);
287
rpt->connected_since_s = now_realtime_sec();
282
- rpt->last_msg_t = now_monotonic_sec();
288
+ rpt->thread.last_traffic_ut = now_monotonic_usec();
289
rpt->hops = 1;
290
291
rpt->capabilities = STREAM_CAP_INVALID;
@@ -294,10 +300,10 @@ int stream_receiver_accept_connection(struct web_client *w, char *decoded_query_
300
rrdhost_system_info_hops_set(rpt->system_info, rpt->hops);
301
302
nd_sock_init(&rpt->sock, netdata_ssl_web_server_ctx, false);
297
- rpt->client_ip = strdupz(w->client_ip);
298
- rpt->client_port = strdupz(w->client_port);
303
+ rpt->remote_ip = strdupz(w->client_ip);
304
+ rpt->remote_port = strdupz(w->client_port);
305
300
- rpt->config.update_every = default_rrd_update_every;
306
+ rpt->config.update_every = nd_profile.update_every;
307
308
// parse the parameters and fill rpt and rpt->system_info
309
@@ -380,8 +386,7 @@ int stream_receiver_accept_connection(struct web_client *w, char *decoded_query_
386
nd_log_daemon(NDLP_NOTICE, "STREAM RCV '%s' [from [%s]:%s]: "
387
"request has parameter '%s' = '%s', which is not used."
388
, (rpt->hostname && *rpt->hostname) ? rpt->hostname : "-"
383
- , rpt->client_ip, rpt->client_port
384
- , name, value);
389
+ , rpt->remote_ip, rpt->remote_port, name, value);
390
}
391
}
392
}
@@ -540,8 +545,7 @@ int stream_receiver_accept_connection(struct web_client *w, char *decoded_query_
545
(ssize_t)strlen(initial_response)) {
546
547
nd_log_daemon(NDLP_ERR, "STREAM RCV '%s' [from [%s]:%s]: failed to reply.",
543
- rpt->hostname, rpt->client_ip, rpt->client_port
544
- );
548
+ rpt->hostname, rpt->remote_ip, rpt->remote_port);
549
}
550
551
stream_receiver_free(rpt);
@@ -598,7 +602,7 @@ int stream_receiver_accept_connection(struct web_client *w, char *decoded_query_
602
if (host) {
603
rrdhost_receiver_lock(host);
604
if (host->receiver) {
601
- age = now_monotonic_sec() - host->receiver->last_msg_t;
605
+ age =(time_t)((now_monotonic_usec() - host->receiver->thread.last_traffic_ut) / USEC_PER_SEC);
606
607
if (age < 30)
608
receiver_working = true;
@@ -618,8 +622,7 @@ int stream_receiver_accept_connection(struct web_client *w, char *decoded_query_
622
nd_log_daemon(NDLP_NOTICE, "STREAM '%s' [receive from [%s]:%s]: "
623
"stopped previous stale receiver to accept this one."
624
, rpt->hostname
621
- , rpt->client_ip, rpt->client_port
622
- );
625
+ , rpt->remote_ip, rpt->remote_port);
626
}
627
628
if (receiver_working || receiver_stale) {
src/streaming/stream-receiver-internals.h
+41
-11
@@ -3,6 +3,16 @@
3
#ifndef NETDATA_STREAM_RECEIVER_INTERNALS_H
4
#define NETDATA_STREAM_RECEIVER_INTERNALS_H
5
6
+#include "libnetdata/libnetdata.h"
7
+
8
+#ifdef NETDATA_LOG_STREAM_RECEIVER
9
+#include "stream-traffic-types.h"
10
+struct receiver_state;
11
+void stream_receiver_log_payload(struct receiver_state *rpt, const char *payload, STREAM_TRAFFIC_TYPE type, bool inbound);
12
+#else
13
+#define stream_receiver_log_payload(s, payload, type, inbound) debug_dummy()
14
+#endif
15
+
16
#include "stream.h"
17
#include "stream-thread.h"
18
#include "stream-conf.h"
@@ -22,29 +32,24 @@ struct receiver_state {
32
char *registry_hostname;
33
char *machine_guid;
34
char *os;
25
- char *timezone; // Unused?
35
+ char *timezone; // Unused?
36
char *abbrev_timezone;
27
- char *client_ip; // Duplicated in pluginsd
28
- char *client_port; // Duplicated in pluginsd
29
- char *program_name; // Duplicated in pluginsd
37
+ char *remote_ip; // Duplicated in pluginsd
38
+ char *remote_port; // Duplicated in pluginsd
39
+ char *program_name; // Duplicated in pluginsd
40
char *program_version;
41
struct rrdhost_system_info *system_info;
32
- time_t last_msg_t;
42
time_t connected_since_s;
43
35
- struct buffered_reader reader;
36
-
44
struct {
38
- bool draining_input; // used exclusively by the stream thread
39
-
45
// The parser pointer is safe to read and use, only when having the host receiver lock.
46
// Without this lock, the data pointed by the pointer may vanish randomly.
47
// Also, since the receiver sets it when it starts, it should be read with
48
// an atomic read.
49
struct parser *parser;
50
struct plugind cd;
46
- BUFFER *buffer;
51
52
+ // compressed data input
53
struct {
54
bool enabled;
55
size_t start;
@@ -54,6 +59,12 @@ struct receiver_state {
59
struct decompressor_state decompressor;
60
} compressed;
61
62
+ // uncompressed data input (either directly or via the decompressor)
63
+ struct buffered_reader uncompressed;
64
+
65
+ // a single line of input (composed via uncompressed buffer input)
66
+ BUFFER *line_buffer;
67
+
68
struct {
69
SPINLOCK spinlock;
70
struct stream_opcode msg;
@@ -61,9 +72,19 @@ struct receiver_state {
72
STREAM_CIRCULAR_BUFFER *scb;
73
} send_to_child;
74
75
+ nd_poll_event_t wanted;
76
+ usec_t last_traffic_ut;
77
struct pollfd_meta meta;
78
} thread;
79
80
+ struct {
81
+ uint32_t last_counter_in; // copy from the host, to detect progress
82
+ uint32_t last_counter_out; // copy from the host, to detect progress
83
+ usec_t last_progress_ut; // last time we found some progress (monotonic)
84
+
85
+ time_t first_time_s;
86
+ } replication;
87
+
88
struct {
89
bool shutdown; // signal the streaming parser to exit
90
STREAM_HANDSHAKE reason;
@@ -71,7 +92,13 @@ struct receiver_state {
92
93
struct stream_receiver_config config;
94
74
- time_t replication_first_time_t;
95
+#ifdef NETDATA_LOG_STREAM_RECEIVER
96
+ struct {
97
+ struct timespec first_call;
98
+ SPINLOCK spinlock;
99
+ FILE *fp;
100
+ } log;
101
+#endif
102
103
#ifdef ENABLE_H2O
104
void *h2o_ctx;
@@ -93,4 +120,7 @@ bool stream_receiver_signal_to_stop_and_wait(RRDHOST *host, STREAM_HANDSHAKE rea
120
void stream_receiver_send_opcode(struct receiver_state *rpt, struct stream_opcode msg);
121
void stream_receiver_handle_op(struct stream_thread *sth, struct receiver_state *rpt, struct stream_opcode *msg);
122
123
+void stream_receiver_check_all_nodes_from_poll(struct stream_thread *sth, usec_t now_ut);
124
+void stream_receiver_replication_check_from_poll(struct stream_thread *sth, usec_t now_ut);
125
+
126
#endif //NETDATA_STREAM_RECEIVER_INTERNALS_H
src/streaming/stream-receiver.c
+556
-248
@@ -5,6 +5,66 @@
5
#include "stream-receiver-internals.h"
6
#include "web/server/h2o/http_server.h"
7
8
+#ifdef NETDATA_LOG_STREAM_RECEIVER
9
+void stream_receiver_log_payload(struct receiver_state *rpt, const char *payload, STREAM_TRAFFIC_TYPE type __maybe_unused, bool inbound) {
10
+ if (!rpt || type != STREAM_TRAFFIC_TYPE_REPLICATION) return; // not a streaming parser
11
+
12
+ spinlock_lock(&rpt->log.spinlock);
13
+
14
+ if (!rpt->log.fp) {
15
+ char filename[FILENAME_MAX + 1];
16
+ snprintfz(
17
+ filename, FILENAME_MAX, "/tmp/stream-receiver-%s.txt", rpt->host ? rrdhost_hostname(rpt->host) : "unknown");
18
+
19
+ rpt->log.fp = fopen(filename, "w");
20
+
21
+ // Align first_call to wall clock time
22
+ clock_gettime(CLOCK_REALTIME, &rpt->log.first_call);
23
+ rpt->log.first_call.tv_nsec = 0; // Align to the start of the second
24
+ }
25
+
26
+ if (rpt->log.fp) {
27
+ struct timespec now;
28
+ clock_gettime(CLOCK_REALTIME, &now);
29
+
30
+ time_t elapsed_sec = now.tv_sec - rpt->log.first_call.tv_sec;
31
+ long elapsed_nsec = now.tv_nsec - rpt->log.first_call.tv_nsec;
32
+
33
+ if (elapsed_nsec < 0) {
34
+ elapsed_sec--;
35
+ elapsed_nsec += 1000000000;
36
+ }
37
+
38
+ uint16_t days = elapsed_sec / 86400;
39
+ uint8_t hours = (elapsed_sec % 86400) / 3600;
40
+ uint8_t minutes = (elapsed_sec % 3600) / 60;
41
+ uint8_t seconds = elapsed_sec % 60;
42
+ uint16_t milliseconds = elapsed_nsec / 1000000;
43
+
44
+ char prefix[30];
45
+ snprintf(prefix, sizeof(prefix), "%03ud.%02u:%02u:%02u.%03u ",
46
+ days, hours, minutes, seconds, milliseconds);
47
+
48
+ const char *line_start = payload;
49
+ const char *line_end;
50
+
51
+ while (line_start && *line_start) {
52
+ line_end = strchr(line_start, '\n');
53
+ if (line_end) {
54
+ fprintf(rpt->log.fp, "%s%s%.*s\n", prefix, inbound ? "> " : "< ", (int)(line_end - line_start), line_start);
55
+ line_start = line_end + 1;
56
+ } else {
57
+ fprintf(rpt->log.fp, "%s%s%s\n", prefix, inbound ? "> " : "< ", line_start);
58
+ break;
59
+ }
60
+ }
61
+ }
62
+
63
+ fflush(rpt->log.fp);
64
+ spinlock_unlock(&rpt->log.spinlock);
65
+}
66
+#endif
67
+
68
static void stream_receiver_remove(struct stream_thread *sth, struct receiver_state *rpt, const char *why);
69
70
// When a child disconnects this is the maximum we will wait
@@ -97,16 +157,16 @@ static inline ssize_t read_stream(struct receiver_state *r, char* buffer, size_t
157
// --------------------------------------------------------------------------------------------------------------------
158
159
static inline ssize_t receiver_read_uncompressed(struct receiver_state *r) {
100
- internal_fatal(r->reader.read_buffer[r->reader.read_len] != '\0',
160
+ internal_fatal(r->thread.uncompressed.read_buffer[r->thread.uncompressed.read_len] != '\0',
161
"%s: read_buffer does not start with zero #2", __FUNCTION__ );
162
103
- ssize_t bytes = read_stream(r, r->reader.read_buffer + r->reader.read_len, sizeof(r->reader.read_buffer) - r->reader.read_len - 1);
163
+ ssize_t bytes = read_stream(r, r->thread.uncompressed.read_buffer + r->thread.uncompressed.read_len, sizeof(r->thread.uncompressed.read_buffer) - r->thread.uncompressed.read_len - 1);
164
if(bytes > 0) {
165
worker_set_metric(WORKER_RECEIVER_JOB_BYTES_READ, (NETDATA_DOUBLE)bytes);
166
worker_set_metric(WORKER_RECEIVER_JOB_BYTES_UNCOMPRESSED, (NETDATA_DOUBLE)bytes);
167
108
- r->reader.read_len += bytes;
109
- r->reader.read_buffer[r->reader.read_len] = '\0';
168
+ r->thread.uncompressed.read_len += bytes;
169
+ r->thread.uncompressed.read_buffer[r->thread.uncompressed.read_len] = '\0';
170
}
171
172
return bytes;
@@ -149,7 +209,7 @@ static inline decompressor_status_t receiver_feed_decompressor(struct receiver_s
209
if (unlikely(!compressed_message_size)) {
210
nd_log(NDLS_DAEMON, NDLP_ERR,
211
"STREAM RCV[x] '%s' [from [%s]:%s]: multiplexed uncompressed data in compressed stream!",
152
- rrdhost_hostname(r->host), r->client_ip, r->client_port);
212
+ rrdhost_hostname(r->host), r->remote_ip, r->remote_port);
213
return DECOMPRESS_FAILED;
214
}
215
@@ -158,7 +218,7 @@ static inline decompressor_status_t receiver_feed_decompressor(struct receiver_s
218
"STREAM RCV[x] '%s' [from [%s]:%s]: received a compressed message of %zu bytes, "
219
"which is bigger than the max compressed message "
220
"size supported of %zu. Ignoring message.",
161
- rrdhost_hostname(r->host), r->client_ip, r->client_port,
221
+ rrdhost_hostname(r->host), r->remote_ip, r->remote_port,
222
compressed_message_size, (size_t)COMPRESSION_MAX_MSG_SIZE);
223
return DECOMPRESS_FAILED;
224
}
@@ -175,7 +235,7 @@ static inline decompressor_status_t receiver_feed_decompressor(struct receiver_s
235
if (unlikely(!bytes_to_parse)) {
236
nd_log(NDLS_DAEMON, NDLP_ERR,
237
"STREAM RCV[x] '%s' [from [%s]:%s]: no bytes to decompress.",
178
- rrdhost_hostname(r->host), r->client_ip, r->client_port);
238
+ rrdhost_hostname(r->host), r->remote_ip, r->remote_port);
239
return DECOMPRESS_FAILED;
240
}
241
@@ -191,20 +251,20 @@ static inline decompressor_status_t receiver_get_decompressed(struct receiver_st
251
if (unlikely(!stream_decompressed_bytes_in_buffer(&r->thread.compressed.decompressor)))
252
return DECOMPRESS_NEED_MORE_DATA;
253
194
- size_t available = sizeof(r->reader.read_buffer) - r->reader.read_len - 1;
254
+ size_t available = sizeof(r->thread.uncompressed.read_buffer) - r->thread.uncompressed.read_len - 1;
255
if (likely(available)) {
256
size_t len = stream_decompressor_get(
197
- &r->thread.compressed.decompressor, r->reader.read_buffer + r->reader.read_len, available);
257
+ &r->thread.compressed.decompressor, r->thread.uncompressed.read_buffer + r->thread.uncompressed.read_len, available);
258
if (unlikely(!len)) {
259
internal_error(true, "decompressor returned zero length #1");
260
return DECOMPRESS_FAILED;
261
}
262
203
- r->reader.read_len += (int)len;
204
- r->reader.read_buffer[r->reader.read_len] = '\0';
263
+ r->thread.uncompressed.read_len += (int)len;
264
+ r->thread.uncompressed.read_buffer[r->thread.uncompressed.read_len] = '\0';
265
}
266
else {
207
- internal_fatal(true, "The line to read is too big! Already have %zd bytes in read_buffer.", r->reader.read_len);
267
+ internal_fatal(true, "The line to read is too big! Already have %zd bytes in read_buffer.", r->thread.uncompressed.read_len);
268
return DECOMPRESS_FAILED;
269
}
270
@@ -213,7 +273,7 @@ static inline decompressor_status_t receiver_get_decompressed(struct receiver_st
273
274
static inline ssize_t receiver_read_compressed(struct receiver_state *r) {
275
216
- internal_fatal(r->reader.read_buffer[r->reader.read_len] != '\0',
276
+ internal_fatal(r->thread.uncompressed.read_buffer[r->thread.uncompressed.read_len] != '\0',
277
"%s: read_buffer does not start with zero #2", __FUNCTION__ );
278
279
ssize_t bytes_read = read_stream(r, r->thread.compressed.buf + r->thread.compressed.used,
@@ -248,11 +308,10 @@ static inline bool receiver_should_stop(struct receiver_state *rpt) {
308
void stream_receiver_handle_op(struct stream_thread *sth, struct receiver_state *rpt, struct stream_opcode *msg) {
309
ND_LOG_STACK lgs[] = {
310
ND_LOG_FIELD_STR(NDF_NIDL_NODE, rpt->host->hostname),
251
- ND_LOG_FIELD_TXT(NDF_SRC_IP, rpt->client_ip),
252
- ND_LOG_FIELD_TXT(NDF_SRC_PORT, rpt->client_port),
311
+ ND_LOG_FIELD_TXT(NDF_SRC_IP, rpt->remote_ip),
312
+ ND_LOG_FIELD_TXT(NDF_SRC_PORT, rpt->remote_port),
313
ND_LOG_FIELD_CB(NDF_SRC_TRANSPORT, stream_receiver_log_transport, rpt),
314
ND_LOG_FIELD_CB(NDF_SRC_CAPABILITIES, stream_receiver_log_capabilities, rpt),
255
- ND_LOG_FIELD_UUID(NDF_MESSAGE_ID, &streaming_to_parent_msgid),
315
ND_LOG_FIELD_END(),
316
};
317
ND_LOG_STACK_PUSH(lgs);
@@ -267,7 +326,7 @@ void stream_receiver_handle_op(struct stream_thread *sth, struct receiver_state
326
nd_log(NDLS_DAEMON, NDLP_ERR,
327
"STREAM RCV[%zu] '%s' [from [%s]:%s]: send buffer is full (buffer size %u, max %u, used %u, available %u). "
328
"Restarting connection.",
270
- sth->id, rrdhost_hostname(rpt->host), rpt->client_ip, rpt->client_port,
329
+ sth->id, rrdhost_hostname(rpt->host), rpt->remote_ip, rpt->remote_port,
330
stats.bytes_size, stats.bytes_max_size, stats.bytes_outstanding, stats.bytes_available);
331
332
stream_receiver_remove(sth, rpt, "receiver send buffer overflow");
@@ -297,8 +356,12 @@ static ssize_t send_to_child(const char *txt, void *data, STREAM_TRAFFIC_TYPE ty
356
msg.opcode = STREAM_OPCODE_RECEIVER_BUFFER_OVERFLOW;
357
rc = -1;
358
}
300
- else if(was_empty)
301
- msg.opcode = STREAM_OPCODE_RECEIVER_POLLOUT;
359
+ else {
360
+ stream_receiver_log_payload(rpt, txt, type, false);
361
+
362
+ if(was_empty)
363
+ msg.opcode = STREAM_OPCODE_RECEIVER_POLLOUT;
364
+ }
365
366
spinlock_unlock(&rpt->thread.send_to_child.spinlock);
367
@@ -308,81 +371,6 @@ static ssize_t send_to_child(const char *txt, void *data, STREAM_TRAFFIC_TYPE ty
371
return rc;
372
}
373
311
-static void streaming_parser_init(struct receiver_state *rpt) {
312
- rpt->thread.cd = (struct plugind){
313
- .update_every = default_rrd_update_every,
314
- .unsafe = {
315
- .spinlock = SPINLOCK_INITIALIZER,
316
- .running = true,
317
- .enabled = true,
318
- },
319
- .started_t = now_realtime_sec(),
320
- };
321
-
322
- // put the client IP and port into the buffers used by plugins.d
323
- {
324
- char buf[CONFIG_MAX_NAME];
325
- snprintfz(buf, sizeof(buf), "[%s]:%s", rpt->client_ip, rpt->client_port);
326
- string_freez(rpt->thread.cd.id);
327
- rpt->thread.cd.id = string_strdupz(buf);
328
-
329
- string_freez(rpt->thread.cd.filename);
330
- rpt->thread.cd.filename = string_strdupz(buf);
331
-
332
- string_freez(rpt->thread.cd.fullfilename);
333
- rpt->thread.cd.fullfilename = string_strdupz(buf);
334
-
335
- string_freez(rpt->thread.cd.cmd);
336
- rpt->thread.cd.cmd = string_strdupz(buf);
337
- }
338
-
339
- PARSER *parser = NULL;
340
- {
341
- PARSER_USER_OBJECT user = {
342
- .enabled = plugin_is_enabled(&rpt->thread.cd),
343
- .host = rpt->host,
344
- .opaque = rpt,
345
- .cd = &rpt->thread.cd,
346
- .trust_durations = 1,
347
- .capabilities = rpt->capabilities,
348
- };
349
-
350
- parser = parser_init(&user, -1, -1, PARSER_INPUT_SPLIT, &rpt->sock);
351
- parser->send_to_plugin_data = rpt;
352
- parser->send_to_plugin_cb = send_to_child;
353
- }
354
-
355
-#ifdef ENABLE_H2O
356
- parser->h2o_ctx = rpt->h2o_ctx;
357
-#endif
358
-
359
- pluginsd_keywords_init(parser, PARSER_INIT_STREAMING);
360
-
361
- rpt->thread.compressed.start = 0;
362
- rpt->thread.compressed.used = 0;
363
- rpt->thread.compressed.enabled = stream_decompression_initialize(rpt);
364
- buffered_reader_init(&rpt->reader);
365
-
366
-#ifdef NETDATA_LOG_STREAM_RECEIVE
367
- {
368
- char filename[FILENAME_MAX + 1];
369
- snprintfz(filename, FILENAME_MAX, "/tmp/stream-receiver-%s.txt", rpt->host ? rrdhost_hostname(
370
- rpt->host) : "unknown"
371
- );
372
- parser->user.stream_log_fp = fopen(filename, "w");
373
- parser->user.stream_log_repertoire = PARSER_REP_METADATA;
374
- }
375
-#endif
376
-
377
- __atomic_store_n(&rpt->thread.parser, parser, __ATOMIC_RELAXED);
378
- stream_receiver_send_node_and_claim_id_to_child(rpt->host);
379
-
380
- rpt->thread.buffer = buffer_create(sizeof(rpt->reader.read_buffer), NULL);
381
-
382
- // help rrdset_push_metric_initialize() select the right buffer
383
- rpt->host->stream.snd.commit.receiver_tid = gettid_cached();
384
-}
385
-
374
// --------------------------------------------------------------------------------------------------------------------
375
376
static void stream_receive_log_database_gap(struct receiver_state *rpt) {
@@ -398,14 +386,14 @@ static void stream_receive_log_database_gap(struct receiver_state *rpt) {
386
if(!last_db_entry) {
387
nd_log(NDLS_DAEMON, NDLP_NOTICE,
388
"STREAM RCV '%s' [from [%s]:%s]: node connected; for the first time!",
401
- rrdhost_hostname(host), rpt->client_ip, rpt->client_port);
389
+ rrdhost_hostname(host), rpt->remote_ip, rpt->remote_port);
390
}
391
else {
392
char buf[128];
393
duration_snprintf(buf, sizeof(buf), now - last_db_entry, "s", true);
394
nd_log(NDLS_DAEMON, NDLP_NOTICE,
395
"STREAM RCV '%s' [from [%s]:%s]: node connected; last sample in the database %s ago",
408
- rrdhost_hostname(host), rpt->client_ip, rpt->client_port, buf);
396
+ rrdhost_hostname(host), rpt->remote_ip, rpt->remote_port, buf);
397
}
398
}
399
@@ -416,14 +404,23 @@ void stream_receiver_move_to_running_unsafe(struct stream_thread *sth, struct re
404
405
ND_LOG_STACK lgs[] = {
406
ND_LOG_FIELD_STR(NDF_NIDL_NODE, rpt->host->hostname),
419
- ND_LOG_FIELD_UUID(NDF_MESSAGE_ID, &streaming_to_parent_msgid),
407
+ ND_LOG_FIELD_UUID(NDF_MESSAGE_ID, &streaming_from_child_msgid),
408
ND_LOG_FIELD_END(),
409
};
410
ND_LOG_STACK_PUSH(lgs);
411
412
nd_log(NDLS_DAEMON, NDLP_DEBUG,
413
"STREAM RCV[%zu] '%s' [from [%s]:%s]: moving host from receiver queue to receiver running...",
426
- sth->id, rrdhost_hostname(rpt->host), rpt->client_ip, rpt->client_port);
414
+ sth->id, rrdhost_hostname(rpt->host), rpt->remote_ip, rpt->remote_port);
415
+
416
+ sock_setcloexec(rpt->sock.fd, true);
417
+ sock_enlarge_rcv_buf(rpt->sock.fd);
418
+ sock_enlarge_snd_buf(rpt->sock.fd);
419
+ sock_setcork(rpt->sock.fd, false);
420
+ if(sock_setnonblock(rpt->sock.fd, true) != 1)
421
+ nd_log(NDLS_DAEMON, NDLP_ERR,
422
+ "STREAM RCV '%s' [from [%s]:%s]: failed to set non-blocking mode on socket %d",
423
+ rrdhost_hostname(rpt->host), rpt->remote_ip, rpt->remote_port, rpt->sock.fd);
424
425
rpt->host->stream.rcv.status.tid = gettid_cached();
426
rpt->thread.meta.type = POLLFD_TYPE_RECEIVER;
@@ -439,22 +436,83 @@ void stream_receiver_move_to_running_unsafe(struct stream_thread *sth, struct re
436
internal_fatal(META_GET(&sth->run.meta, (Word_t)&rpt->thread.meta) != NULL, "Receiver to be added is already in the list of receivers");
437
META_SET(&sth->run.meta, (Word_t)&rpt->thread.meta, &rpt->thread.meta);
438
442
- if(sock_setnonblock(rpt->sock.fd) < 0)
443
- nd_log(NDLS_DAEMON, NDLP_ERR,
444
- "STREAM RCV '%s' [from [%s]:%s]: cannot set the non-blocking flag from socket %d",
445
- rrdhost_hostname(rpt->host), rpt->client_ip, rpt->client_port, rpt->sock.fd);
446
-
447
- if(!nd_poll_add(sth->run.ndpl, rpt->sock.fd, ND_POLL_READ, &rpt->thread.meta))
439
+ rpt->thread.wanted = ND_POLL_READ;
440
+ if(!nd_poll_add(sth->run.ndpl, rpt->sock.fd, rpt->thread.wanted, &rpt->thread.meta))
441
nd_log(NDLS_DAEMON, NDLP_ERR,
442
"STREAM RCV[%zu] '%s' [from [%s]:%s]:"
443
"Failed to add receiver socket to nd_poll()",
451
- sth->id, rrdhost_hostname(rpt->host), rpt->client_ip, rpt->client_port);
444
+ sth->id, rrdhost_hostname(rpt->host), rpt->remote_ip, rpt->remote_port);
445
+
446
+ // put the client IP and port into the buffers used by plugins.d
447
+ {
448
+ char buf[CONFIG_MAX_NAME];
449
+ snprintfz(buf, sizeof(buf), "[%s]:%s", rpt->remote_ip, rpt->remote_port);
450
+ string_freez(rpt->thread.cd.id);
451
+ rpt->thread.cd.id = string_strdupz(buf);
452
+
453
+ string_freez(rpt->thread.cd.filename);
454
+ rpt->thread.cd.filename = string_strdupz(buf);
455
+
456
+ string_freez(rpt->thread.cd.fullfilename);
457
+ rpt->thread.cd.fullfilename = string_strdupz(buf);
458
+
459
+ string_freez(rpt->thread.cd.cmd);
460
+ rpt->thread.cd.cmd = string_strdupz(buf);
461
+ }
462
+
463
+ rpt->thread.compressed.start = 0;
464
+ rpt->thread.compressed.used = 0;
465
+ rpt->thread.compressed.enabled = stream_decompression_initialize(rpt);
466
+ buffered_reader_init(&rpt->thread.uncompressed);
467
+
468
+ rpt->thread.line_buffer = buffer_create(sizeof(rpt->thread.uncompressed.read_buffer), NULL);
469
+
470
+ // help preferred_sender_buffer() select the right buffer
471
+ rpt->host->stream.snd.commit.receiver_tid = gettid_cached();
472
+
473
+ rpt->replication.last_progress_ut = now_monotonic_usec();
474
+
475
+ PARSER *parser = NULL;
476
+ {
477
+ rpt->thread.cd = (struct plugind){
478
+ .update_every = nd_profile.update_every,
479
+ .unsafe = {
480
+ .spinlock = SPINLOCK_INITIALIZER,
481
+ .running = true,
482
+ .enabled = true,
483
+ },
484
+ .started_t = now_realtime_sec(),
485
+ };
486
+
487
+ PARSER_USER_OBJECT user = {
488
+ .enabled = plugin_is_enabled(&rpt->thread.cd),
489
+ .host = rpt->host,
490
+ .opaque = rpt,
491
+ .cd = &rpt->thread.cd,
492
+ .trust_durations = 1,
493
+ .capabilities = rpt->capabilities,
494
+#ifdef NETDATA_LOG_STREAM_RECEIVER
495
+ .rpt = rpt,
496
+#endif
497
+ };
498
+
499
+ parser = parser_init(&user, -1, -1, PARSER_INPUT_SPLIT, &rpt->sock);
500
+ parser->send_to_plugin_data = rpt;
501
+ parser->send_to_plugin_cb = send_to_child;
502
+
503
+ pluginsd_keywords_init(parser, PARSER_INIT_STREAMING);
504
+
505
+ __atomic_store_n(&rpt->thread.parser, parser, __ATOMIC_RELAXED);
506
+ }
507
+
508
+#ifdef ENABLE_H2O
509
+ parser->h2o_ctx = rpt->h2o_ctx;
510
+#endif
511
512
stream_receive_log_database_gap(rpt);
454
- rrdhost_state_connected(rpt->host);
513
456
- // keep this last, since it sends commands back to the child
457
- streaming_parser_init(rpt);
514
+ // keep this last - it needs everything ready since to sends data to the child
515
+ stream_receiver_send_node_and_claim_id_to_child(rpt->host);
516
}
517
518
void stream_receiver_move_entire_queue_to_running_unsafe(struct stream_thread *sth) {
@@ -473,6 +531,17 @@ void stream_receiver_move_entire_queue_to_running_unsafe(struct stream_thread *s
531
static void stream_receiver_remove(struct stream_thread *sth, struct receiver_state *rpt, const char *why) {
532
internal_fatal(sth->tid != gettid_cached(), "Function %s() should only be used by the dispatcher thread", __FUNCTION__ );
533
534
+ ND_LOG_STACK lgs[] = {
535
+ ND_LOG_FIELD_STR(NDF_NIDL_NODE, rpt->host->hostname),
536
+ ND_LOG_FIELD_TXT(NDF_SRC_IP, rpt->remote_ip),
537
+ ND_LOG_FIELD_TXT(NDF_SRC_PORT, rpt->remote_port),
538
+ ND_LOG_FIELD_CB(NDF_SRC_TRANSPORT, stream_receiver_log_transport, rpt),
539
+ ND_LOG_FIELD_CB(NDF_SRC_CAPABILITIES, stream_receiver_log_capabilities, rpt),
540
+ ND_LOG_FIELD_UUID(NDF_MESSAGE_ID, &streaming_from_child_msgid),
541
+ ND_LOG_FIELD_END(),
542
+ };
543
+ ND_LOG_STACK_PUSH(lgs);
544
+
545
PARSER *parser = __atomic_load_n(&rpt->thread.parser, __ATOMIC_RELAXED);
546
size_t count = parser ? parser->user.data_collections_count : 0;
547
@@ -482,45 +551,36 @@ static void stream_receiver_remove(struct stream_thread *sth, struct receiver_st
551
"receiver disconnected (after %zu received messages): %s"
552
, sth->id
553
, rpt->hostname ? rpt->hostname : "-"
485
- , rpt->client_ip ? rpt->client_ip : "-"
486
- , rpt->client_port ? rpt->client_port : "-"
554
+ , rpt->remote_ip ? rpt->remote_ip : "-"
555
+ , rpt->remote_port ? rpt->remote_port : "-"
556
, count
557
, why ? why : "");
558
490
- rrdhost_state_disconnected(rpt->host);
559
+ internal_fatal(META_GET(&sth->run.meta, (Word_t)&rpt->thread.meta) == NULL,
560
+ "Receiver to be removed is not found in the list of receivers");
561
492
- internal_fatal(META_GET(&sth->run.meta, (Word_t)&rpt->thread.meta) == NULL, "Receiver to be removed is not found in the list of receivers");
562
META_DEL(&sth->run.meta, (Word_t)&rpt->thread.meta);
563
495
- if(!nd_poll_del(sth->run.ndpl, rpt->sock.fd))
564
+ rpt->thread.wanted = 0;
565
+ if(!nd_poll_del(sth->run.ndpl, rpt->sock.fd, &rpt->thread.meta))
566
nd_log(NDLS_DAEMON, NDLP_ERR, "Failed to delete receiver socket from nd_poll()");
567
568
rpt->host->stream.rcv.status.tid = 0;
569
500
- spinlock_lock(&rpt->thread.send_to_child.spinlock);
501
- rpt->thread.send_to_child.msg.session = 0;
502
- rpt->thread.send_to_child.msg.meta = NULL;
503
- stream_circular_buffer_destroy(rpt->thread.send_to_child.scb);
504
- rpt->thread.send_to_child.scb = NULL;
505
- spinlock_unlock(&rpt->thread.send_to_child.spinlock);
506
-
507
- stream_thread_node_removed(rpt->host);
508
-
509
- buffer_free(rpt->thread.buffer);
510
- rpt->thread.buffer = NULL;
511
-
570
+ // make sure send_to_plugin() will not write any data to the socket (or wait for it to finish)
571
if(parser) {
513
- parser->user.v2.stream_buffer.wb = NULL;
514
-
515
- // make sure send_to_plugin() will not write any data to the socket
572
spinlock_lock(&parser->writer.spinlock);
573
parser->fd_input = -1;
574
parser->fd_output = -1;
575
parser->sock = NULL;
576
spinlock_unlock(&parser->writer.spinlock);
577
+
578
+ parser->user.v2.stream_buffer.wb = NULL;
579
}
580
523
- // the parser stopped
581
+ stream_thread_node_removed(rpt->host);
582
+
583
+ // set a default exit reason, if not set
584
receiver_set_exit_reason(rpt, STREAM_HANDSHAKE_DISCONNECT_PARSER_EXIT, false);
585
586
// in case we are connected to netdata cloud,
@@ -535,9 +595,29 @@ static void stream_receiver_remove(struct stream_thread *sth, struct receiver_st
595
// DO NOT USE rpt after this point
596
}
597
598
+static bool stream_receiver_dequeue_senders(struct stream_thread *sth, struct receiver_state *rpt, usec_t now_ut) {
599
+ internal_fatal(sth->tid != gettid_cached(), "Function %s() should only be used by the dispatcher thread", __FUNCTION__);
600
+
601
+ // re-check if we need to send data after reading - if we do, try now
602
+ if(rpt->thread.wanted & ND_POLL_WRITE) {
603
+ worker_is_busy(WORKER_STREAM_JOB_SOCKET_SEND);
604
+ if(!stream_receiver_send_data(sth, rpt, now_ut, false))
605
+ return false;
606
+ }
607
+
608
+ if(rpt->host->sender && // the host has a sender
609
+ rpt->host->stream.snd.status.tid == gettid_cached() && // the sender is mine
610
+ (rpt->host->sender->thread.wanted & ND_POLL_WRITE)) // the sender needs to send data
611
+ if(!stream_sender_send_data(sth, rpt->host->sender, now_ut, false))
612
+ return false;
613
+
614
+ return true;
615
+}
616
+
617
static ssize_t
539
-stream_receive_and_process(struct stream_thread *sth, struct receiver_state *rpt, PARSER *parser, bool *removed) {
618
+stream_receive_and_process(struct stream_thread *sth, struct receiver_state *rpt, PARSER *parser, usec_t now_ut __maybe_unused, bool *removed) {
619
internal_fatal(sth->tid != gettid_cached(), "Function %s() should only be used by the dispatcher thread", __FUNCTION__);
620
+ *removed = false;
621
622
ssize_t rc;
623
if(rpt->thread.compressed.enabled) {
@@ -559,16 +639,16 @@ stream_receive_and_process(struct stream_thread *sth, struct receiver_state *rpt
639
if (likely(decompress_rc == DECOMPRESS_OK)) {
640
// loop through all the complete lines found in the uncompressed buffer
641
562
- while (buffered_reader_next_line(&rpt->reader, rpt->thread.buffer)) {
563
- if (unlikely(parser_action(parser, rpt->thread.buffer->buffer))) {
642
+ while (buffered_reader_next_line(&rpt->thread.uncompressed, rpt->thread.line_buffer)) {
643
+ if (unlikely(parser_action(parser, rpt->thread.line_buffer->buffer))) {
644
receiver_set_exit_reason(rpt, STREAM_HANDSHAKE_DISCONNECT_PARSER_FAILED, false);
645
stream_receiver_remove(sth, rpt, "parser action failed");
646
*removed = true;
647
return -1;
648
}
649
570
- rpt->thread.buffer->len = 0;
571
- rpt->thread.buffer->buffer[0] = '\0';
650
+ rpt->thread.line_buffer->len = 0;
651
+ rpt->thread.line_buffer->buffer[0] = '\0';
652
}
653
}
654
else if (decompress_rc == DECOMPRESS_NEED_MORE_DATA)
@@ -601,41 +681,203 @@ stream_receive_and_process(struct stream_thread *sth, struct receiver_state *rpt
681
}
682
else {
683
rc = receiver_read_uncompressed(rpt);
604
- if(rc <= 0) return rc;
684
+ if(rc <= 0)
685
+ return rc;
686
606
- while(buffered_reader_next_line(&rpt->reader, rpt->thread.buffer)) {
607
- if(unlikely(parser_action(parser, rpt->thread.buffer->buffer))) {
687
+ while(buffered_reader_next_line(&rpt->thread.uncompressed, rpt->thread.line_buffer)) {
688
+ if(unlikely(parser_action(parser, rpt->thread.line_buffer->buffer))) {
689
receiver_set_exit_reason(rpt, STREAM_HANDSHAKE_DISCONNECT_PARSER_FAILED, false);
690
stream_receiver_remove(sth, rpt, "parser action failed");
691
*removed = true;
692
return -1;
693
}
694
614
- rpt->thread.buffer->len = 0;
615
- rpt->thread.buffer->buffer[0] = '\0';
695
+ rpt->thread.line_buffer->len = 0;
696
+ rpt->thread.line_buffer->buffer[0] = '\0';
697
}
698
}
699
700
return rc;
701
}
702
703
+bool stream_receiver_send_data(struct stream_thread *sth, struct receiver_state *rpt, usec_t now_ut, bool process_opcodes_and_enable_removal) {
704
+ internal_fatal(sth->tid != gettid_cached(), "Function %s() should only be used by the dispatcher thread", __FUNCTION__ );
705
+
706
+ EVLOOP_STATUS status = EVLOOP_STATUS_CONTINUE;
707
+ while(status == EVLOOP_STATUS_CONTINUE) {
708
+ if (!spinlock_trylock(&rpt->thread.send_to_child.spinlock)) {
709
+ status = EVLOOP_STATUS_CANT_GET_LOCK;
710
+ break;
711
+ }
712
+
713
+ char *chunk;
714
+ STREAM_CIRCULAR_BUFFER *scb = rpt->thread.send_to_child.scb;
715
+ STREAM_CIRCULAR_BUFFER_STATS *stats = stream_circular_buffer_stats_unsafe(scb);
716
+ size_t outstanding = stream_circular_buffer_get_unsafe(scb, &chunk);
717
+
718
+ if(!outstanding) {
719
+ status = EVLOOP_STATUS_NO_MORE_DATA;
720
+ spinlock_unlock(&rpt->thread.send_to_child.spinlock);
721
+ continue;
722
+ }
723
+
724
+ ssize_t rc = write_stream(rpt, chunk, outstanding);
725
+ if (likely(rc > 0)) {
726
+ rpt->thread.last_traffic_ut = now_ut;
727
+ stream_circular_buffer_del_unsafe(scb, rc, now_ut);
728
+ if (!stats->bytes_outstanding) {
729
+ rpt->thread.wanted = ND_POLL_READ;
730
+ if (!nd_poll_upd(sth->run.ndpl, rpt->sock.fd, rpt->thread.wanted, &rpt->thread.meta))
731
+ nd_log(NDLS_DAEMON, NDLP_ERR,
732
+ "STREAM RCV[%zu] '%s' [from [%s]:%s]: cannot update nd_poll()",
733
+ sth->id, rrdhost_hostname(rpt->host), rpt->remote_ip, rpt->remote_port);
734
+
735
+ // recreate the circular buffer if we have to
736
+ stream_circular_buffer_recreate_timed_unsafe(rpt->thread.send_to_child.scb, now_ut, false);
737
+ status = EVLOOP_STATUS_NO_MORE_DATA;
738
+ }
739
+ }
740
+ else if (rc == 0 || errno == ECONNRESET)
741
+ status = EVLOOP_STATUS_SOCKET_CLOSED;
742
+
743
+ else if (rc < 0) {
744
+ if (errno == EWOULDBLOCK || errno == EAGAIN || errno == EINTR)
745
+ status = EVLOOP_STATUS_SOCKET_FULL;
746
+ else
747
+ status = EVLOOP_STATUS_SOCKET_ERROR;
748
+ }
749
+
750
+ spinlock_unlock(&rpt->thread.send_to_child.spinlock);
751
+
752
+ if (status == EVLOOP_STATUS_SOCKET_ERROR || status == EVLOOP_STATUS_SOCKET_CLOSED) {
753
+ const char *disconnect_reason;
754
+ STREAM_HANDSHAKE reason;
755
+
756
+ if(status == EVLOOP_STATUS_SOCKET_ERROR) {
757
+ worker_is_busy(WORKER_STREAM_JOB_DISCONNECT_SEND_ERROR);
758
+ disconnect_reason = "socket reports error while writing";
759
+ reason = STREAM_HANDSHAKE_DISCONNECT_SOCKET_WRITE_FAILED;
760
+ }
761
+ else /* if(status == EVLOOP_STATUS_SOCKET_CLOSED) */ {
762
+ worker_is_busy(WORKER_STREAM_JOB_DISCONNECT_REMOTE_CLOSED);
763
+ disconnect_reason = "socket reports EOF (closed by child)";
764
+ reason = STREAM_HANDSHAKE_DISCONNECT_SOCKET_CLOSED_BY_REMOTE_END;
765
+ }
766
+
767
+ nd_log(NDLS_DAEMON, NDLP_ERR,
768
+ "STREAM RCV[%zu] '%s' [from [%s]:%s]: %s (%zd, on fd %d) - closing receiver connection - "
769
+ "we have sent %zu bytes in %zu operations.",
770
+ sth->id, rrdhost_hostname(rpt->host), rpt->remote_ip, rpt->remote_port,
771
+ disconnect_reason, rc, rpt->sock.fd, stats->bytes_sent, stats->sends);
772
+
773
+ receiver_set_exit_reason(rpt, reason, false);
774
+
775
+ if(process_opcodes_and_enable_removal) {
776
+ // this is not executed from the opcode handling mechanism
777
+ // so we can safely remove the receiver.
778
+ stream_receiver_remove(sth, rpt, disconnect_reason);
779
+ }
780
+ else {
781
+ // protection against this case:
782
+ //
783
+ // 1. receiver gets a replication request
784
+ // 2. parser processes the request
785
+ // 3. parser decides to send back a message to the child (REPLAY_CHART)
786
+ // 4. send_to_child appends the data to the sending circular buffer
787
+ // 5. send_to_child sends opcode to enable sending
788
+ // 6. opcode bypasses the signal and runs this function inline to dispatch immediately
789
+ // 7. sending fails (child disconnected)
790
+ // 8. receiver is removed
791
+ //
792
+ // Point 2 above crashes. The parser is no longer there (freed at point 7)
793
+ // and there is no way for point 2 to know...
794
+ }
795
+ }
796
+ else if(process_opcodes_and_enable_removal &&
797
+ status == EVLOOP_STATUS_CONTINUE &&
798
+ stream_thread_process_opcodes(sth, &rpt->thread.meta))
799
+ status = EVLOOP_STATUS_OPCODE_ON_ME;
800
+ }
801
+
802
+ return EVLOOP_STATUS_STILL_ALIVE(status);
803
+}
804
+
805
+bool stream_receiver_receive_data(struct stream_thread *sth, struct receiver_state *rpt, usec_t now_ut, bool process_opcodes) {
806
+ internal_fatal(sth->tid != gettid_cached(), "Function %s() should only be used by the dispatcher thread", __FUNCTION__ );
807
+
808
+ PARSER *parser = __atomic_load_n(&rpt->thread.parser, __ATOMIC_RELAXED);
809
+ ND_LOG_STACK lgs[] = {
810
+ ND_LOG_FIELD_CB(NDF_REQUEST, line_splitter_reconstruct_line, &parser->line),
811
+ ND_LOG_FIELD_CB(NDF_NIDL_NODE, parser_reconstruct_node, parser),
812
+ ND_LOG_FIELD_CB(NDF_NIDL_INSTANCE, parser_reconstruct_instance, parser),
813
+ ND_LOG_FIELD_CB(NDF_NIDL_CONTEXT, parser_reconstruct_context, parser),
814
+ ND_LOG_FIELD_END(),
815
+ };
816
+ ND_LOG_STACK_PUSH(lgs);
817
+
818
+ EVLOOP_STATUS status = EVLOOP_STATUS_CONTINUE;
819
+ while(status == EVLOOP_STATUS_CONTINUE) {
820
+ bool removed = false;
821
+ ssize_t rc = stream_receive_and_process(sth, rpt, parser, now_ut, &removed);
822
+ if(unlikely(removed))
823
+ status = EVLOOP_STATUS_PARSER_FAILED;
824
+
825
+ else if (likely(rc > 0)) {
826
+ rpt->thread.last_traffic_ut = now_ut;
827
+
828
+ if(!stream_receiver_dequeue_senders(sth, rpt, now_ut))
829
+ status = EVLOOP_STATUS_SOCKET_ERROR;
830
+ }
831
+ else if (rc == 0 || errno == ECONNRESET) {
832
+ status = EVLOOP_STATUS_SOCKET_CLOSED;
833
+ }
834
+ else if (rc < 0) {
835
+ if ((errno == EWOULDBLOCK || errno == EAGAIN || errno == EINTR))
836
+ status = EVLOOP_STATUS_SOCKET_FULL;
837
+ else
838
+ status = EVLOOP_STATUS_SOCKET_ERROR;
839
+ }
840
+
841
+ if(status == EVLOOP_STATUS_SOCKET_ERROR || status == EVLOOP_STATUS_SOCKET_CLOSED) {
842
+ const char *disconnect_reason;
843
+ STREAM_HANDSHAKE reason;
844
+
845
+ if(status == EVLOOP_STATUS_SOCKET_ERROR) {
846
+ worker_is_busy(WORKER_STREAM_JOB_DISCONNECT_RECEIVE_ERROR);
847
+ reason = STREAM_HANDSHAKE_DISCONNECT_SOCKET_READ_FAILED;
848
+ disconnect_reason = "error during receive";
849
+ }
850
+ else /* if(status == EVLOOP_STATUS_SOCKET_CLOSED) */ {
851
+ worker_is_busy(WORKER_STREAM_JOB_DISCONNECT_REMOTE_CLOSED);
852
+ reason = STREAM_HANDSHAKE_DISCONNECT_SOCKET_CLOSED_BY_REMOTE_END;
853
+ disconnect_reason = "socket reports EOF (closed by child)";
854
+ }
855
+
856
+ nd_log(NDLS_DAEMON, NDLP_ERR,
857
+ "STREAM RCV[%zu] '%s' [from [%s]:%s]: %s (fd %d) - closing receiver connection.",
858
+ sth->id, rrdhost_hostname(rpt->host), rpt->remote_ip, rpt->remote_port, disconnect_reason, rpt->sock.fd);
859
+
860
+ receiver_set_exit_reason(rpt, reason, false);
861
+ stream_receiver_remove(sth, rpt, disconnect_reason);
862
+ }
863
+ else if(status == EVLOOP_STATUS_CONTINUE && process_opcodes && stream_thread_process_opcodes(sth, &rpt->thread.meta))
864
+ status = EVLOOP_STATUS_OPCODE_ON_ME;
865
+ }
866
+
867
+ return EVLOOP_STATUS_STILL_ALIVE(status);
868
+}
869
+
870
// process poll() events for streaming receivers
871
// returns true when the receiver is still there, false if it removed it
624
-bool stream_receive_process_poll_events(struct stream_thread *sth, struct receiver_state *rpt, nd_poll_event_t events, usec_t now_ut)
625
-{
872
+bool stream_receive_process_poll_events(struct stream_thread *sth, struct receiver_state *rpt, nd_poll_event_t events, usec_t now_ut) {
873
internal_fatal(sth->tid != gettid_cached(), "Function %s() should only be used by the dispatcher thread", __FUNCTION__);
874
628
- PARSER *parser = __atomic_load_n(&rpt->thread.parser, __ATOMIC_RELAXED);
875
ND_LOG_STACK lgs[] = {
630
- ND_LOG_FIELD_TXT(NDF_SRC_IP, rpt->client_ip),
631
- ND_LOG_FIELD_TXT(NDF_SRC_PORT, rpt->client_port),
876
+ ND_LOG_FIELD_TXT(NDF_SRC_IP, rpt->remote_ip),
877
+ ND_LOG_FIELD_TXT(NDF_SRC_PORT, rpt->remote_port),
878
ND_LOG_FIELD_TXT(NDF_NIDL_NODE, rpt->hostname),
879
ND_LOG_FIELD_CB(NDF_SRC_TRANSPORT, stream_receiver_log_transport, rpt),
880
ND_LOG_FIELD_CB(NDF_SRC_CAPABILITIES, stream_receiver_log_capabilities, rpt),
635
- ND_LOG_FIELD_CB(NDF_REQUEST, line_splitter_reconstruct_line, &parser->line),
636
- ND_LOG_FIELD_CB(NDF_NIDL_NODE, parser_reconstruct_node, parser),
637
- ND_LOG_FIELD_CB(NDF_NIDL_INSTANCE, parser_reconstruct_instance, parser),
638
- ND_LOG_FIELD_CB(NDF_NIDL_CONTEXT, parser_reconstruct_context, parser),
881
ND_LOG_FIELD_END(),
882
};
883
ND_LOG_STACK_PUSH(lgs);
@@ -664,7 +906,7 @@ bool stream_receive_process_poll_events(struct stream_thread *sth, struct receiv
906
907
nd_log(NDLS_DAEMON, NDLP_ERR,
908
"STREAM RCV[%zu] '%s' [from [%s]:%s]: %s - closing connection",
667
- sth->id, rrdhost_hostname(rpt->host), rpt->client_ip, rpt->client_port, error);
909
+ sth->id, rrdhost_hostname(rpt->host), rpt->remote_ip, rpt->remote_port, error);
910
911
receiver_set_exit_reason(rpt, STREAM_HANDSHAKE_DISCONNECT_SOCKET_ERROR, false);
912
stream_receiver_remove(sth, rpt, error);
@@ -673,111 +915,154 @@ bool stream_receive_process_poll_events(struct stream_thread *sth, struct receiv
915
916
if (events & ND_POLL_WRITE) {
917
worker_is_busy(WORKER_STREAM_JOB_SOCKET_SEND);
918
+ if(!stream_receiver_send_data(sth, rpt, now_ut, true))
919
+ return false;
920
+ }
921
677
- bool stop = false;
678
- while(!stop) {
679
- if (spinlock_trylock(&rpt->thread.send_to_child.spinlock)) {
680
- const char *disconnect_reason = NULL;
681
- STREAM_HANDSHAKE reason;
682
-
683
- char *chunk;
684
- STREAM_CIRCULAR_BUFFER *scb = rpt->thread.send_to_child.scb;
685
- STREAM_CIRCULAR_BUFFER_STATS *stats = stream_circular_buffer_stats_unsafe(scb);
686
- size_t outstanding = stream_circular_buffer_get_unsafe(scb, &chunk);
687
- ssize_t rc = write_stream(rpt, chunk, outstanding);
688
- if (likely(rc > 0)) {
689
- stream_circular_buffer_del_unsafe(scb, rc);
690
- if (!stats->bytes_outstanding) {
691
- if (!nd_poll_upd(sth->run.ndpl, rpt->sock.fd, ND_POLL_READ, &rpt->thread.meta))
692
- nd_log(NDLS_DAEMON, NDLP_ERR,
693
- "STREAM RCV[%zu] '%s' [from [%s]:%s]: cannot update nd_poll()",
694
- sth->id, rrdhost_hostname(rpt->host), rpt->client_ip, rpt->client_port);
695
-
696
- // recreate the circular buffer if we have to
697
- stream_circular_buffer_recreate_timed_unsafe(rpt->thread.send_to_child.scb, now_ut, false);
698
- stop = true;
699
- }
700
- else if(stream_thread_process_opcodes(sth, &rpt->thread.meta))
701
- stop = true;
702
- }
703
- else if (rc == 0 || errno == ECONNRESET) {
704
- disconnect_reason = "socket reports EOF (closed by child)";
705
- reason = STREAM_HANDSHAKE_DISCONNECT_SOCKET_CLOSED_BY_REMOTE_END;
706
- }
707
- else if (rc < 0) {
708
- if (errno == EWOULDBLOCK || errno == EAGAIN || errno == EINTR)
709
- // will try later
710
- stop = true;
711
- else {
712
- disconnect_reason = "socket reports error while writing";
713
- reason = STREAM_HANDSHAKE_DISCONNECT_SOCKET_WRITE_FAILED;
714
- }
715
- }
716
- spinlock_unlock(&rpt->thread.send_to_child.spinlock);
922
+ if (events & ND_POLL_READ) {
923
+ worker_is_busy(WORKER_STREAM_JOB_SOCKET_RECEIVE);
924
+ if(!stream_receiver_receive_data(sth, rpt, now_ut, true))
925
+ return false;
926
+ }
927
718
- if (disconnect_reason) {
719
- worker_is_busy(WORKER_SENDER_JOB_DISCONNECT_SEND_ERROR);
720
- nd_log(NDLS_DAEMON, NDLP_ERR,
721
- "STREAM RCV[%zu] '%s' [from [%s]:%s]: %s (%zd, on fd %d) - closing connection - "
722
- "we have sent %zu bytes in %zu operations.",
723
- sth->id, rrdhost_hostname(rpt->host), rpt->client_ip, rpt->client_port,
724
- disconnect_reason, rc, rpt->sock.fd, stats->bytes_sent, stats->sends);
725
-
726
- receiver_set_exit_reason(rpt, reason, false);
727
- stream_receiver_remove(sth, rpt, disconnect_reason);
728
- return false;
729
- }
730
- }
731
- else
732
- break;
928
+ return true;
929
+}
930
+
931
+void stream_receiver_check_all_nodes_from_poll(struct stream_thread *sth, usec_t now_ut) {
932
+ internal_fatal(sth->tid != gettid_cached(), "Function %s() should only be used by the dispatcher thread", __FUNCTION__ );
933
+
934
+ NETDATA_DOUBLE overall_buffer_ratio = 0.0;
935
+
936
+ Word_t idx = 0;
937
+ for(struct pollfd_meta *m = META_FIRST(&sth->run.meta, &idx);
938
+ m;
939
+ m = META_NEXT(&sth->run.meta, &idx)) {
940
+ if (m->type != POLLFD_TYPE_RECEIVER) continue;
941
+ struct receiver_state *rpt = m->rpt;
942
+
943
+ spinlock_lock(&rpt->thread.send_to_child.spinlock);
944
+ STREAM_CIRCULAR_BUFFER_STATS stats = *stream_circular_buffer_stats_unsafe(rpt->thread.send_to_child.scb);
945
+ spinlock_unlock(&rpt->thread.send_to_child.spinlock);
946
+
947
+ if (stats.buffer_ratio > overall_buffer_ratio)
948
+ overall_buffer_ratio = stats.buffer_ratio;
949
+
950
+ time_t timeout_s = 600;
951
+ if(unlikely(rpt->thread.last_traffic_ut + timeout_s * USEC_PER_SEC < now_ut &&
952
+ !rrdhost_receiver_replicating_charts(rpt->host))) {
953
+
954
+ ND_LOG_STACK lgs[] = {
955
+ ND_LOG_FIELD_TXT(NDF_SRC_IP, rpt->remote_ip),
956
+ ND_LOG_FIELD_TXT(NDF_SRC_PORT, rpt->remote_port),
957
+ ND_LOG_FIELD_TXT(NDF_NIDL_NODE, rpt->hostname),
958
+ ND_LOG_FIELD_CB(NDF_SRC_TRANSPORT, stream_receiver_log_transport, rpt),
959
+ ND_LOG_FIELD_CB(NDF_SRC_CAPABILITIES, stream_receiver_log_capabilities, rpt),
960
+ ND_LOG_FIELD_END(),
961
+ };
962
+ ND_LOG_STACK_PUSH(lgs);
963
+
964
+ worker_is_busy(WORKER_SENDER_JOB_DISCONNECT_TIMEOUT);
965
+
966
+ char duration[RFC3339_MAX_LENGTH];
967
+ duration_snprintf(duration, sizeof(duration), (int64_t)(now_monotonic_usec() - rpt->thread.last_traffic_ut), "us", true);
968
+
969
+ char pending[64] = "0";
970
+ if(stats.bytes_outstanding)
971
+ size_snprintf(pending, sizeof(pending), stats.bytes_outstanding, "B", false);
972
+
973
+ nd_log(NDLS_DAEMON, NDLP_ERR,
974
+ "STREAM RCV[%zu] '%s' [from %s]: there was not traffic for %ld seconds - closing connection - "
975
+ "we have sent %zu bytes in %zu operations, it is idle for %s, and we have %s pending to send "
976
+ "(buffer is used %.2f%%).",
977
+ sth->id, rrdhost_hostname(rpt->host), rpt->remote_ip, timeout_s,
978
+ stats.bytes_sent, stats.sends, duration, pending, stats.buffer_ratio);
979
+
980
+ receiver_set_exit_reason(rpt, STREAM_HANDSHAKE_DISCONNECT_SOCKET_TIMEOUT, false);
981
+ stream_receiver_remove(sth, rpt, "timeout");
982
+ continue;
983
}
984
+
985
+ rpt->thread.wanted = ND_POLL_READ | (stats.bytes_outstanding ? ND_POLL_WRITE : 0);
986
+ if(!nd_poll_upd(sth->run.ndpl, rpt->sock.fd, rpt->thread.wanted, &rpt->thread.meta))
987
+ nd_log(NDLS_DAEMON, NDLP_ERR,
988
+ "STREAM RCV[%zu] '%s' [from %s]: failed to update nd_poll().",
989
+ sth->id, rrdhost_hostname(rpt->host), rpt->remote_ip);
990
+
991
+ }
992
+}
993
+
994
+static bool stream_receiver_did_replication_progress(struct receiver_state *rpt) {
995
+ RRDHOST *host = rpt->host;
996
+
997
+ size_t my_counter_in = __atomic_load_n(&rpt->replication.last_counter_in, __ATOMIC_RELAXED);
998
+ size_t my_counter_out = __atomic_load_n(&rpt->replication.last_counter_out, __ATOMIC_RELAXED);
999
+ size_t host_counter_in = __atomic_load_n(&host->stream.rcv.status.replication.counter_in, __ATOMIC_RELAXED);
1000
+ size_t host_counter_out = __atomic_load_n(&host->stream.rcv.status.replication.counter_out, __ATOMIC_RELAXED);
1001
+ if(my_counter_in != host_counter_in || my_counter_out != host_counter_out) {
1002
+ // there has been some progress
1003
+ __atomic_store_n(&rpt->replication.last_counter_in, __atomic_load_n(&host->stream.rcv.status.replication.counter_in, __ATOMIC_RELAXED), __ATOMIC_RELAXED);
1004
+ __atomic_store_n(&rpt->replication.last_counter_out, __atomic_load_n(&host->stream.rcv.status.replication.counter_out, __ATOMIC_RELAXED), __ATOMIC_RELAXED);
1005
+ rpt->replication.last_progress_ut = now_monotonic_usec();
1006
+ return true;
1007
}
1008
736
- if (!(events & ND_POLL_READ))
1009
+ if(!my_counter_in || !my_counter_out)
1010
+ // we have not started yet
1011
return true;
1012
739
- // we can receive data from this socket
1013
+ if(__atomic_load_n(&host->stream.rcv.status.replication.backfill_pending, __ATOMIC_RELAXED))
1014
+ // we still have requests to execute
1015
+ return true;
1016
741
- worker_is_busy(WORKER_STREAM_JOB_SOCKET_RECEIVE);
742
- bool removed = false, stop = false;
743
- size_t iterations = 0;
744
- while(!removed && !stop && iterations++ < MAX_IO_ITERATIONS_PER_EVENT) {
745
- ssize_t rc = stream_receive_and_process(sth, rpt, parser, &removed);
746
- if (likely(rc > 0)) {
747
- rpt->last_msg_t = (time_t)(now_ut / USEC_PER_SEC);
1017
+ return (now_monotonic_usec() - rpt->replication.last_progress_ut < 5ULL * 60 * USEC_PER_SEC);
1018
+}
1019
749
- if(stream_thread_process_opcodes(sth, &rpt->thread.meta))
750
- stop = true;
751
- }
752
- else if (rc == 0 || errno == ECONNRESET) {
753
- worker_is_busy(WORKER_SENDER_JOB_DISCONNECT_REMOTE_CLOSED);
754
- nd_log(NDLS_DAEMON, NDLP_ERR,
755
- "STREAM RCV[%zu] '%s' [from [%s]:%s]: socket %d reports EOF (closed by child).",
756
- sth->id, rrdhost_hostname(rpt->host), rpt->client_ip, rpt->client_port, rpt->sock.fd);
757
- receiver_set_exit_reason(rpt, STREAM_HANDSHAKE_DISCONNECT_SOCKET_CLOSED_BY_REMOTE_END, false);
758
- stream_receiver_remove(sth, rpt, "socket reports EOF (closed by child)");
759
- return false;
760
- }
761
- else if (rc < 0) {
762
- if(removed)
763
- return false;
1020
+void stream_receiver_replication_check_from_poll(struct stream_thread *sth, usec_t now_ut __maybe_unused) {
1021
+ internal_fatal(sth->tid != gettid_cached(), "Function %s() should only be used by the dispatcher thread", __FUNCTION__);
1022
765
- else if ((errno == EWOULDBLOCK || errno == EAGAIN || errno == EINTR))
766
- // will try later
767
- stop = true;
768
- else {
769
- worker_is_busy(WORKER_SENDER_JOB_DISCONNECT_RECEIVE_ERROR);
770
- nd_log(NDLS_DAEMON, NDLP_ERR,
771
- "STREAM RCV[%zu] '%s' [from [%s]:%s]: error during receive (%zd, on fd %d) - closing connection.",
772
- sth->id, rrdhost_hostname(rpt->host), rpt->client_ip, rpt->client_port, rc, rpt->sock.fd);
773
- receiver_set_exit_reason(rpt, STREAM_HANDSHAKE_DISCONNECT_SOCKET_READ_FAILED, false);
774
- stream_receiver_remove(sth, rpt, "error during receive");
775
- return false;
776
- }
1023
+ Word_t idx = 0;
1024
+ for(struct pollfd_meta *m = META_FIRST(&sth->run.meta, &idx);
1025
+ m;
1026
+ m = META_NEXT(&sth->run.meta, &idx)) {
1027
+ if (m->type != POLLFD_TYPE_RECEIVER) continue;
1028
+ struct receiver_state *rpt = m->rpt;
1029
+ RRDHOST *host = rpt->host;
1030
+
1031
+
1032
+ if(stream_receiver_did_replication_progress(rpt))
1033
+ continue;
1034
+
1035
+ size_t exceptions = 0;
1036
+ RRDSET *st;
1037
+ rrdset_foreach_read(st, rpt->host) {
1038
+ RRDSET_FLAGS st_flags = rrdset_flag_get(st);
1039
+ if(st_flags & (RRDSET_FLAG_OBSOLETE | RRDSET_FLAG_RECEIVER_REPLICATION_FINISHED))
1040
+ continue;
1041
+
1042
+ const char *status = (st_flags & RRDSET_FLAG_RECEIVER_REPLICATION_IN_PROGRESS) ? "has not finished" : "has not started";
1043
+
1044
+ nd_log(NDLS_DAEMON, NDLP_WARNING,
1045
+ "STREAM RCV[%zu] '%s' [from %s]: REPLICATION EXCEPTIONS: instance '%s' %s replication yet.",
1046
+ sth->id, rrdhost_hostname(host), rpt->remote_ip,
1047
+ rrdset_id(st), status);
1048
+
1049
+ exceptions++;
1050
+ }
1051
+ rrdset_foreach_done(st);
1052
+
1053
+ if(exceptions && !stream_receiver_did_replication_progress(rpt)) {
1054
+ nd_log(NDLS_DAEMON, NDLP_WARNING,
1055
+ "STREAM RCV[%zu] '%s' [from %s]: REPLICATION EXCEPTIONS SUMMARY: node has %zu stalled replication requests. "
1056
+ "We have received %u and sent %u replication commands. "
1057
+ "Disconnecting node to restore streaming.",
1058
+ sth->id, rrdhost_hostname(rpt->host), rpt->remote_ip, exceptions,
1059
+ __atomic_load_n(&host->stream.rcv.status.replication.counter_in, __ATOMIC_RELAXED),
1060
+ __atomic_load_n(&host->stream.rcv.status.replication.counter_out, __ATOMIC_RELAXED));
1061
+
1062
+ receiver_set_exit_reason(rpt, STREAM_HANDSHAKE_REPLICATION_STALLED, false);
1063
+ stream_receiver_remove(sth, rpt, "replication reception stalled");
1064
}
1065
}
779
-
780
- return !removed;
1066
}
1067
1068
void stream_receiver_cleanup(struct stream_thread *sth) {
@@ -795,11 +1080,27 @@ void stream_receiver_cleanup(struct stream_thread *sth) {
1080
static void stream_receiver_replication_reset(RRDHOST *host) {
1081
RRDSET *st;
1082
rrdset_foreach_read(st, host) {
798
- rrdset_flag_clear(st, RRDSET_FLAG_RECEIVER_REPLICATION_IN_PROGRESS);
799
- rrdset_flag_set(st, RRDSET_FLAG_RECEIVER_REPLICATION_FINISHED);
1083
+ RRDSET_FLAGS old = rrdset_flag_set_and_clear(st, RRDSET_FLAG_RECEIVER_REPLICATION_FINISHED, RRDSET_FLAG_RECEIVER_REPLICATION_IN_PROGRESS);
1084
+ if(!(old & RRDSET_FLAG_RECEIVER_REPLICATION_FINISHED))
1085
+ rrdhost_receiver_replicating_charts_minus_one(host);
1086
+
1087
+#ifdef REPLICATION_TRACKING
1088
+ st->stream.rcv.who = REPLAY_WHO_UNKNOWN;
1089
+#endif
1090
}
1091
rrdset_foreach_done(st);
802
- rrdhost_receiver_replicating_charts_zero(host);
1092
+
1093
+ if(rrdhost_receiver_replicating_charts(host) != 0) {
1094
+ nd_log(NDLS_DAEMON, NDLP_WARNING,
1095
+ "STREAM REPLAY ERROR: receiver replication instances counter should be zero, but it is %u"
1096
+ " - resetting it to zero",
1097
+ rrdhost_receiver_replicating_charts(host));
1098
+
1099
+ rrdhost_receiver_replicating_charts_zero(host);
1100
+ }
1101
+
1102
+ __atomic_store_n(&host->stream.rcv.status.replication.counter_in, 0, __ATOMIC_RELAXED);
1103
+ __atomic_store_n(&host->stream.rcv.status.replication.counter_out, 0, __ATOMIC_RELAXED);
1104
}
1105
1106
bool rrdhost_set_receiver(RRDHOST *host, struct receiver_state *rpt) {
@@ -809,7 +1110,10 @@ bool rrdhost_set_receiver(RRDHOST *host, struct receiver_state *rpt) {
1110
rrdhost_receiver_lock(host);
1111
1112
if (!host->receiver) {
1113
+ object_state_activate(&host->state_id);
1114
+
1115
rrdhost_flag_clear(host, RRDHOST_FLAG_ORPHAN);
1116
+ rrdhost_set_health_evloop_iteration(host);
1117
1118
host->stream.rcv.status.connections++;
1119
streaming_receiver_connected();
@@ -829,7 +1133,7 @@ bool rrdhost_set_receiver(RRDHOST *host, struct receiver_state *rpt) {
1133
nd_log(NDLS_DAEMON, NDLP_DEBUG,
1134
"STREAM RCV '%s' [from [%s]:%s]: "
1135
"Postponing health checks for %" PRId64 " seconds, because it was just connected.",
832
- rrdhost_hostname(host), rpt->client_ip, rpt->client_port,
1136
+ rrdhost_hostname(host), rpt->remote_ip, rpt->remote_port,
1137
(int64_t) rpt->config.health.delay);
1138
}
1139
}
@@ -875,12 +1179,15 @@ void rrdhost_clear_receiver(struct receiver_state *rpt) {
1179
1180
rrdhost_receiver_unlock(host);
1181
{
1182
+ // this will wait until all workers finish
1183
+ object_state_deactivate(&host->state_id);
1184
+
1185
// run all these without having the receiver lock
1186
1187
+ rrdhost_set_health_evloop_iteration(host);
1188
ml_host_stop(host);
1189
stream_path_child_disconnected(host);
1190
stream_sender_signal_to_stop_and_wait(host, STREAM_HANDSHAKE_DISCONNECT_RECEIVER_LEFT, false);
883
- stream_receiver_replication_reset(host);
1191
rrdcontext_host_child_disconnected(host);
1192
1193
if (rpt->config.health.enabled)
@@ -892,6 +1199,7 @@ void rrdhost_clear_receiver(struct receiver_state *rpt) {
1199
1200
// now we have the lock again
1201
1202
+ stream_receiver_replication_reset(host);
1203
streaming_receiver_disconnected();
1204
1205
__atomic_store_n(&host->receiver->exit.shutdown, false, __ATOMIC_RELAXED);
@@ -940,7 +1248,7 @@ bool stream_receiver_signal_to_stop_and_wait(RRDHOST *host, STREAM_HANDSHAKE rea
1248
netdata_log_error("STREAM RCV[x] '%s' [from [%s]:%s]: "
1249
"streaming thread takes too long to stop, giving up..."
1250
, rrdhost_hostname(host)
943
- , host->receiver->client_ip, host->receiver->client_port);
1251
+ , host->receiver->remote_ip, host->receiver->remote_port);
1252
else
1253
ret = true;
1254
src/streaming/stream-replication-receiver.c
new
+261
@@ -0,0 +1,261 @@
1
+// SPDX-License-Identifier: GPL-3.0-or-later
2
+
3
+#include "stream-replication-receiver.h"
4
+#include "stream-receiver-internals.h"
5
+
6
+struct replication_request_details {
7
+ struct {
8
+ send_command callback;
9
+ struct parser *parser;
10
+ } caller;
11
+
12
+ RRDHOST *host;
13
+ RRDSET *st;
14
+
15
+ struct {
16
+ time_t first_entry_t; // the first entry time the child has
17
+ time_t last_entry_t; // the last entry time the child has
18
+ time_t wall_clock_time; // the current time of the child
19
+ bool fixed_last_entry; // when set we set the last entry to wall clock time
20
+ } child_db;
21
+
22
+ struct {
23
+ time_t first_entry_t; // the first entry time we have
24
+ time_t last_entry_t; // the last entry time we have
25
+ time_t wall_clock_time; // the current local world clock time
26
+ } local_db;
27
+
28
+ struct {
29
+ time_t from; // the starting time of the entire gap we have
30
+ time_t to; // the ending time of the entire gap we have
31
+ } gap;
32
+
33
+ struct {
34
+ time_t after; // the start time we requested previously from this child
35
+ time_t before; // the end time we requested previously from this child
36
+ } last_request;
37
+
38
+ struct {
39
+ time_t after; // the start time of this replication request - the child will add 1 second
40
+ time_t before; // the end time of this replication request
41
+ bool start_streaming; // true when we want the child to send anything remaining and start streaming - the child will overwrite 'before'
42
+ } wanted;
43
+};
44
+
45
+static void replicate_log_request(struct replication_request_details *r, const char *msg) {
46
+#ifdef NETDATA_INTERNAL_CHECKS
47
+ internal_error(true,
48
+#else
49
+ nd_log_limit_static_global_var(erl, 1, 0);
50
+ nd_log_limit(&erl, NDLS_DAEMON, NDLP_NOTICE,
51
+#endif
52
+ "STREAM SND REPLAY ERROR: 'host:%s/chart:%s' child sent: "
53
+ "db from %ld to %ld%s, wall clock time %ld, "
54
+ "last request from %ld to %ld, "
55
+ "issue: %s - "
56
+ "sending replication request from %ld to %ld, start streaming %s",
57
+ rrdhost_hostname(r->st->rrdhost), rrdset_id(r->st),
58
+ r->child_db.first_entry_t,
59
+ r->child_db.last_entry_t, r->child_db.fixed_last_entry ? " (fixed)" : "",
60
+ r->child_db.wall_clock_time,
61
+ r->last_request.after,
62
+ r->last_request.before,
63
+ msg,
64
+ r->wanted.after,
65
+ r->wanted.before,
66
+ r->wanted.start_streaming ? "true" : "false");
67
+}
68
+
69
+static bool send_replay_chart_cmd(struct replication_request_details *r, const char *msg, bool log) {
70
+ RRDSET *st = r->st;
71
+
72
+ if(log)
73
+ replicate_log_request(r, msg);
74
+
75
+ if(st->rrdhost->receiver && (!st->rrdhost->receiver->replication.first_time_s || r->wanted.after < st->rrdhost->receiver->replication.first_time_s))
76
+ st->rrdhost->receiver->replication.first_time_s = r->wanted.after;
77
+
78
+#ifdef NETDATA_LOG_REPLICATION_REQUESTS
79
+ st->replay.log_next_data_collection = true;
80
+
81
+ char wanted_after_buf[LOG_DATE_LENGTH + 1] = "", wanted_before_buf[LOG_DATE_LENGTH + 1] = "";
82
+
83
+ if(r->wanted.after)
84
+ log_date(wanted_after_buf, LOG_DATE_LENGTH, r->wanted.after);
85
+
86
+ if(r->wanted.before)
87
+ log_date(wanted_before_buf, LOG_DATE_LENGTH, r->wanted.before);
88
+
89
+ internal_error(true,
90
+ "STREAM SND REPLAY: 'host:%s/chart:%s' sending replication request %ld [%s] to %ld [%s], start streaming '%s': %s: "
91
+ "last[%ld - %ld] child[%ld - %ld, now %ld %s] local[%ld - %ld, now %ld] gap[%ld - %ld %s] %s"
92
+ , rrdhost_hostname(r->host), rrdset_id(r->st)
93
+ , r->wanted.after, wanted_after_buf
94
+ , r->wanted.before, wanted_before_buf
95
+ , r->wanted.start_streaming ? "YES" : "NO"
96
+ , msg
97
+ , r->last_request.after, r->last_request.before
98
+ , r->child_db.first_entry_t, r->child_db.last_entry_t
99
+ , r->child_db.wall_clock_time, (r->child_db.wall_clock_time == r->local_db.wall_clock_time) ? "SAME" : (r->child_db.wall_clock_time < r->local_db.wall_clock_time) ? "BEHIND" : "AHEAD"
100
+ , r->local_db.first_entry_t, r->local_db.last_entry_t
101
+ , r->local_db.wall_clock_time
102
+ , r->gap.from, r->gap.to
103
+ , (r->gap.from == r->wanted.after) ? "FULL" : "PARTIAL"
104
+ , (st->replay.after != 0 || st->replay.before != 0) ? "OVERLAPPING" : ""
105
+ );
106
+
107
+ st->replay.start_streaming = r->wanted.start_streaming;
108
+ st->replay.after = r->wanted.after;
109
+ st->replay.before = r->wanted.before;
110
+#endif // NETDATA_LOG_REPLICATION_REQUESTS
111
+
112
+ char buffer[2048 + 1];
113
+ snprintfz(buffer, sizeof(buffer) - 1, PLUGINSD_KEYWORD_REPLAY_CHART " \"%s\" \"%s\" %llu %llu\n",
114
+ rrdset_id(st), r->wanted.start_streaming ? "true" : "false",
115
+ (unsigned long long)r->wanted.after, (unsigned long long)r->wanted.before);
116
+
117
+ ssize_t ret = r->caller.callback(buffer, r->caller.parser, STREAM_TRAFFIC_TYPE_REPLICATION);
118
+ if (ret < 0) {
119
+ netdata_log_error("STREAM SND REPLAY ERROR: 'host:%s/chart:%s' failed to send replication request to child (error %zd)",
120
+ rrdhost_hostname(r->host), rrdset_id(r->st), ret);
121
+ return false;
122
+ }
123
+
124
+ __atomic_add_fetch(&st->rrdhost->stream.rcv.status.replication.counter_out, 1, __ATOMIC_RELAXED);
125
+
126
+#ifdef REPLICATION_TRACKING
127
+ st->stream.rcv.who = REPLAY_WHO_THEM;
128
+#endif
129
+
130
+ return true;
131
+}
132
+
133
+bool replicate_chart_request(send_command callback, struct parser *parser, RRDHOST *host, RRDSET *st,
134
+ time_t child_first_entry, time_t child_last_entry, time_t child_wall_clock_time,
135
+ time_t prev_first_entry_wanted, time_t prev_last_entry_wanted)
136
+{
137
+ struct replication_request_details r = {
138
+ .caller = {
139
+ .callback = callback,
140
+ .parser = parser,
141
+ },
142
+
143
+ .host = host,
144
+ .st = st,
145
+
146
+ .child_db = {
147
+ .first_entry_t = child_first_entry,
148
+ .last_entry_t = child_last_entry,
149
+ .wall_clock_time = child_wall_clock_time,
150
+ .fixed_last_entry = false,
151
+ },
152
+
153
+ .local_db = {
154
+ .first_entry_t = 0,
155
+ .last_entry_t = 0,
156
+ .wall_clock_time = now_realtime_sec(),
157
+ },
158
+
159
+ .last_request = {
160
+ .after = prev_first_entry_wanted,
161
+ .before = prev_last_entry_wanted,
162
+ },
163
+
164
+ .wanted = {
165
+ .after = 0,
166
+ .before = 0,
167
+ .start_streaming = true,
168
+ },
169
+ };
170
+
171
+ if(r.child_db.last_entry_t > r.child_db.wall_clock_time) {
172
+ replicate_log_request(&r, "child's db last entry > child's wall clock time");
173
+ r.child_db.last_entry_t = r.child_db.wall_clock_time;
174
+ r.child_db.fixed_last_entry = true;
175
+ }
176
+
177
+ rrdset_get_retention_of_tier_for_collected_chart(r.st, &r.local_db.first_entry_t, &r.local_db.last_entry_t, r.local_db.wall_clock_time, 0);
178
+
179
+ // let's find the GAP we have
180
+ if(!r.last_request.after || !r.last_request.before) {
181
+ // there is no previous request
182
+
183
+ if(r.local_db.last_entry_t)
184
+ // we have some data, let's continue from the last point we have
185
+ r.gap.from = r.local_db.last_entry_t;
186
+ else
187
+ // we don't have any data, the gap is the max timeframe we are allowed to replicate
188
+ r.gap.from = r.local_db.wall_clock_time - r.host->stream.replication.period;
189
+
190
+ }
191
+ else {
192
+ // we had sent a request - let's continue at the point we left it
193
+ // for this we don't take into account the actual data in our db
194
+ // because the child may also have gaps, and we need to get over it
195
+ r.gap.from = r.last_request.before;
196
+ }
197
+
198
+ // we want all the data up to now
199
+ r.gap.to = r.local_db.wall_clock_time;
200
+
201
+ // The gap is now r.gap.from -> r.gap.to
202
+
203
+ if (unlikely(!rrdhost_option_check(host, RRDHOST_OPTION_REPLICATION)))
204
+ return send_replay_chart_cmd(&r, "sending empty replication request, replication is disabled", false);
205
+
206
+ if (unlikely(!rrdset_number_of_dimensions(st)))
207
+ return send_replay_chart_cmd(&r, "sending empty replication request, chart has no dimensions", false);
208
+
209
+ if (unlikely(!r.child_db.first_entry_t || !r.child_db.last_entry_t))
210
+ return send_replay_chart_cmd(&r, "sending empty replication request, child has no stored data", false);
211
+
212
+ if (unlikely(r.child_db.first_entry_t < 0 || r.child_db.last_entry_t < 0))
213
+ return send_replay_chart_cmd(&r, "sending empty replication request, child db timestamps are invalid", true);
214
+
215
+ if (unlikely(r.child_db.first_entry_t > r.child_db.wall_clock_time))
216
+ return send_replay_chart_cmd(&r, "sending empty replication request, child db first entry is after its wall clock time", true);
217
+
218
+ if (unlikely(r.child_db.first_entry_t > r.child_db.last_entry_t))
219
+ return send_replay_chart_cmd(&r, "sending empty replication request, child timings are invalid (first entry > last entry)", true);
220
+
221
+ if (unlikely(r.local_db.last_entry_t > r.child_db.last_entry_t))
222
+ return send_replay_chart_cmd(&r, "sending empty replication request, local last entry is later than the child one", false);
223
+
224
+ // let's find what the child can provide to fill that gap
225
+
226
+ if(r.child_db.first_entry_t > r.gap.from)
227
+ // the child does not have all the data - let's get what it has
228
+ r.wanted.after = r.child_db.first_entry_t;
229
+ else
230
+ // ok, the child can fill the entire gap we have
231
+ r.wanted.after = r.gap.from;
232
+
233
+ if(r.gap.to - r.wanted.after > host->stream.replication.step)
234
+ // the duration is too big for one request - let's take the first step
235
+ r.wanted.before = r.wanted.after + host->stream.replication.step;
236
+ else
237
+ // wow, we can do it in one request
238
+ r.wanted.before = r.gap.to;
239
+
240
+ // don't ask from the child more than it has
241
+ if(r.wanted.before > r.child_db.last_entry_t)
242
+ r.wanted.before = r.child_db.last_entry_t;
243
+
244
+ if(r.wanted.after > r.wanted.before) {
245
+ r.wanted.after = 0;
246
+ r.wanted.before = 0;
247
+ r.wanted.start_streaming = true;
248
+ return send_replay_chart_cmd(&r, "sending empty replication request, because wanted 'after' computed bigger than wanted 'before'", true);
249
+ }
250
+
251
+ // the child should start streaming immediately if the wanted duration is small, or we reached the last entry of the child
252
+ r.wanted.start_streaming = (r.local_db.wall_clock_time - r.wanted.after <= host->stream.replication.step ||
253
+ r.wanted.before >= r.child_db.last_entry_t ||
254
+ r.wanted.before >= r.child_db.wall_clock_time ||
255
+ r.wanted.before >= r.local_db.wall_clock_time);
256
+
257
+ // the wanted timeframe is now r.wanted.after -> r.wanted.before
258
+ // send it
259
+ return send_replay_chart_cmd(&r, "OK", false);
260
+}
261
+
src/streaming/stream-replication-receiver.h
new
+27
@@ -0,0 +1,27 @@
1
+// SPDX-License-Identifier: GPL-3.0-or-later
2
+
3
+#ifndef NETDATA_STREAM_REPLICATION_RECEIVER_H
4
+#define NETDATA_STREAM_REPLICATION_RECEIVER_H
5
+
6
+#include "libnetdata/libnetdata.h"
7
+#include "stream-traffic-types.h"
8
+
9
+#ifdef __cplusplus
10
+extern "C" {
11
+#endif
12
+
13
+struct parser;
14
+struct rrdhost;
15
+struct rrdset;
16
+
17
+typedef ssize_t (*send_command)(const char *txt, struct parser *parser, STREAM_TRAFFIC_TYPE type);
18
+bool replicate_chart_request(send_command callback, struct parser *parser,
19
+ struct rrdhost *rh, struct rrdset *rs,
20
+ time_t child_first_entry, time_t child_last_entry, time_t child_wall_clock_time,
21
+ time_t response_first_start_time, time_t response_last_end_time);
22
+
23
+#ifdef __cplusplus
24
+}
25
+#endif
26
+
27
+#endif //NETDATA_STREAM_REPLICATION_RECEIVER_H
src/streaming/stream-replication-sender.c
renamed
+172
-330
@@ -1,9 +1,7 @@
1
// SPDX-License-Identifier: GPL-3.0-or-later
2
3
-#include "stream-receiver-internals.h"
3
#include "stream-sender-internals.h"
5
-#include "replication.h"
6
-#include "Judy.h"
4
+#include "stream-replication-sender.h"
5
6
#define MAX_REPLICATION_MESSAGE_PERCENT_SENDER_BUFFER 25ULL
7
#define MAX_SENDER_BUFFER_PERCENTAGE_ALLOWED 50ULL
@@ -33,7 +31,7 @@
31
#define SECONDS_TO_RESET_POINT_IN_TIME 10
32
33
#define MAX_REPLICATION_THREADS 256
36
-#define REQUESTS_AHEAD_PER_THREAD 1 // 1 = enable synchronous queries
34
+#define REQUESTS_AHEAD_PER_THREAD 0 // 0 = dynamic, 1 = enable synchronous queries, > 1 static
35
36
static struct replication_query_statistics replication_queries = {
37
.spinlock = SPINLOCK_INITIALIZER,
@@ -52,7 +50,7 @@ struct replication_query_statistics replication_get_query_statistics(void) {
50
51
static size_t replication_buffers_allocated = 0;
52
55
-size_t replication_allocated_buffers(void) {
53
+size_t replication_sender_allocated_buffers(void) {
54
return __atomic_load_n(&replication_buffers_allocated, __ATOMIC_RELAXED);
55
}
56
@@ -189,8 +187,7 @@ static struct replication_query *replication_query_prepare(
187
d->rda = dictionary_acquired_item_dup(rd_dfe.dict, rd_dfe.item);
188
d->rd = rd;
189
192
- STORAGE_PRIORITY priority = q->query.locked_data_collection ? STORAGE_PRIORITY_HIGH : STORAGE_PRIORITY_LOW;
193
- if(synchronous) priority = STORAGE_PRIORITY_SYNCHRONOUS_FIRST;
190
+ STORAGE_PRIORITY priority = (synchronous) ? STORAGE_PRIORITY_SYNCHRONOUS_FIRST : STORAGE_PRIORITY_LOW;
191
192
stream_control_replication_query_started();
193
storage_engine_query_init(q->backend, rd->tiers[0].smh, &d->handle,
@@ -622,14 +619,15 @@ static struct replication_query *replication_response_prepare(
619
wall_clock_time, capabilities, synchronous);
620
}
621
625
-void replication_response_cancel_and_finalize(struct replication_query *q) {
622
+static void replication_response_cancel_and_finalize(struct replication_query *q) {
623
if(!q) return;
624
replication_query_finalize(NULL, q, false);
625
}
626
627
static bool sender_is_still_connected_for_this_request(struct replication_request *rq);
628
+static void replication_replied_add(void);
629
632
-bool replication_response_execute_and_finalize(struct replication_query *q, size_t max_msg_size, bool workers) {
630
+bool replication_response_execute_finalize_and_send(struct replication_query *q, size_t max_msg_size, bool workers) {
631
bool with_slots = (q->query.capabilities & STREAM_CAP_SLOTS) ? true : false;
632
NUMBER_ENCODING integer_encoding = (q->query.capabilities & STREAM_CAP_IEEE754) ? NUMBER_ENCODING_BASE64 : NUMBER_ENCODING_DECIMAL;
633
struct replication_request *rq = q->rq;
@@ -693,15 +691,20 @@ bool replication_response_execute_and_finalize(struct replication_query *q, size
691
if(workers) worker_is_busy(WORKER_JOB_BUFFER_COMMIT);
692
sender_commit(host->sender, wb, STREAM_TRAFFIC_TYPE_REPLICATION);
693
if(workers) worker_is_busy(WORKER_JOB_CLEANUP);
694
+ __atomic_add_fetch(&host->stream.snd.status.replication.counter_out, 1, __ATOMIC_RELAXED);
695
+ replication_replied_add();
696
697
if(enable_streaming) {
698
+#ifdef REPLICATION_TRACKING
699
+ st->stream.snd.who = REPLAY_WHO_FINISHED;
700
+#endif
701
+
702
if(sender_is_still_connected_for_this_request(rq)) {
703
// enable normal streaming if we have to
704
// but only if the sender buffer has not been flushed since we started
705
702
- if(rrdset_flag_check(st, RRDSET_FLAG_SENDER_REPLICATION_IN_PROGRESS)) {
703
- rrdset_flag_clear(st, RRDSET_FLAG_SENDER_REPLICATION_IN_PROGRESS);
704
- rrdset_flag_set(st, RRDSET_FLAG_SENDER_REPLICATION_FINISHED);
706
+ RRDSET_FLAGS old = rrdset_flag_set_and_clear(st, RRDSET_FLAG_SENDER_REPLICATION_FINISHED, RRDSET_FLAG_SENDER_REPLICATION_IN_PROGRESS);
707
+ if(!(old & RRDSET_FLAG_SENDER_REPLICATION_FINISHED)) {
708
rrdhost_sender_replicating_charts_minus_one(st->rrdhost);
709
710
if(!finished_with_gap)
@@ -713,12 +716,18 @@ bool replication_response_execute_and_finalize(struct replication_query *q, size
716
#endif
717
}
718
else
716
- internal_error(true,
717
- "STREAM SND REPLAY ERROR: 'host:%s/chart:%s' "
718
- "received start streaming command, but the chart is not in progress replicating",
719
- rrdhost_hostname(st->rrdhost), rrdset_id(st));
719
+ internal_error(
720
+ true,
721
+ "STREAM SND REPLAY ERROR: 'host:%s/chart:%s' "
722
+ "received start streaming command, but the chart was not in progress replicating",
723
+ rrdhost_hostname(st->rrdhost), rrdset_id(st));
724
}
725
}
726
+ else {
727
+#ifdef REPLICATION_TRACKING
728
+ st->stream.snd.who = REPLAY_WHO_THEM;
729
+#endif
730
+ }
731
732
if(locked_data_collection)
733
spinlock_unlock(&st->data_collection_lock);
@@ -726,259 +735,6 @@ bool replication_response_execute_and_finalize(struct replication_query *q, size
735
return enable_streaming;
736
}
737
729
-// ----------------------------------------------------------------------------
730
-// sending replication requests
731
-
732
-struct replication_request_details {
733
- struct {
734
- send_command callback;
735
- struct parser *parser;
736
- } caller;
737
-
738
- RRDHOST *host;
739
- RRDSET *st;
740
-
741
- struct {
742
- time_t first_entry_t; // the first entry time the child has
743
- time_t last_entry_t; // the last entry time the child has
744
- time_t wall_clock_time; // the current time of the child
745
- bool fixed_last_entry; // when set we set the last entry to wall clock time
746
- } child_db;
747
-
748
- struct {
749
- time_t first_entry_t; // the first entry time we have
750
- time_t last_entry_t; // the last entry time we have
751
- time_t wall_clock_time; // the current local world clock time
752
- } local_db;
753
-
754
- struct {
755
- time_t from; // the starting time of the entire gap we have
756
- time_t to; // the ending time of the entire gap we have
757
- } gap;
758
-
759
- struct {
760
- time_t after; // the start time we requested previously from this child
761
- time_t before; // the end time we requested previously from this child
762
- } last_request;
763
-
764
- struct {
765
- time_t after; // the start time of this replication request - the child will add 1 second
766
- time_t before; // the end time of this replication request
767
- bool start_streaming; // true when we want the child to send anything remaining and start streaming - the child will overwrite 'before'
768
- } wanted;
769
-};
770
-
771
-static void replicate_log_request(struct replication_request_details *r, const char *msg) {
772
-#ifdef NETDATA_INTERNAL_CHECKS
773
- internal_error(true,
774
-#else
775
- nd_log_limit_static_global_var(erl, 1, 0);
776
- nd_log_limit(&erl, NDLS_DAEMON, NDLP_NOTICE,
777
-#endif
778
- "STREAM SND REPLAY ERROR: 'host:%s/chart:%s' child sent: "
779
- "db from %ld to %ld%s, wall clock time %ld, "
780
- "last request from %ld to %ld, "
781
- "issue: %s - "
782
- "sending replication request from %ld to %ld, start streaming %s",
783
- rrdhost_hostname(r->st->rrdhost), rrdset_id(r->st),
784
- r->child_db.first_entry_t,
785
- r->child_db.last_entry_t, r->child_db.fixed_last_entry ? " (fixed)" : "",
786
- r->child_db.wall_clock_time,
787
- r->last_request.after,
788
- r->last_request.before,
789
- msg,
790
- r->wanted.after,
791
- r->wanted.before,
792
- r->wanted.start_streaming ? "true" : "false");
793
-}
794
-
795
-static bool send_replay_chart_cmd(struct replication_request_details *r, const char *msg, bool log) {
796
- RRDSET *st = r->st;
797
-
798
- if(log)
799
- replicate_log_request(r, msg);
800
-
801
- if(st->rrdhost->receiver && (!st->rrdhost->receiver->replication_first_time_t || r->wanted.after < st->rrdhost->receiver->replication_first_time_t))
802
- st->rrdhost->receiver->replication_first_time_t = r->wanted.after;
803
-
804
-#ifdef NETDATA_LOG_REPLICATION_REQUESTS
805
- st->replay.log_next_data_collection = true;
806
-
807
- char wanted_after_buf[LOG_DATE_LENGTH + 1] = "", wanted_before_buf[LOG_DATE_LENGTH + 1] = "";
808
-
809
- if(r->wanted.after)
810
- log_date(wanted_after_buf, LOG_DATE_LENGTH, r->wanted.after);
811
-
812
- if(r->wanted.before)
813
- log_date(wanted_before_buf, LOG_DATE_LENGTH, r->wanted.before);
814
-
815
- internal_error(true,
816
- "STREAM SND REPLAY: 'host:%s/chart:%s' sending replication request %ld [%s] to %ld [%s], start streaming '%s': %s: "
817
- "last[%ld - %ld] child[%ld - %ld, now %ld %s] local[%ld - %ld, now %ld] gap[%ld - %ld %s] %s"
818
- , rrdhost_hostname(r->host), rrdset_id(r->st)
819
- , r->wanted.after, wanted_after_buf
820
- , r->wanted.before, wanted_before_buf
821
- , r->wanted.start_streaming ? "YES" : "NO"
822
- , msg
823
- , r->last_request.after, r->last_request.before
824
- , r->child_db.first_entry_t, r->child_db.last_entry_t
825
- , r->child_db.wall_clock_time, (r->child_db.wall_clock_time == r->local_db.wall_clock_time) ? "SAME" : (r->child_db.wall_clock_time < r->local_db.wall_clock_time) ? "BEHIND" : "AHEAD"
826
- , r->local_db.first_entry_t, r->local_db.last_entry_t
827
- , r->local_db.wall_clock_time
828
- , r->gap.from, r->gap.to
829
- , (r->gap.from == r->wanted.after) ? "FULL" : "PARTIAL"
830
- , (st->replay.after != 0 || st->replay.before != 0) ? "OVERLAPPING" : ""
831
- );
832
-
833
- st->replay.start_streaming = r->wanted.start_streaming;
834
- st->replay.after = r->wanted.after;
835
- st->replay.before = r->wanted.before;
836
-#endif // NETDATA_LOG_REPLICATION_REQUESTS
837
-
838
- char buffer[2048 + 1];
839
- snprintfz(buffer, sizeof(buffer) - 1, PLUGINSD_KEYWORD_REPLAY_CHART " \"%s\" \"%s\" %llu %llu\n",
840
- rrdset_id(st), r->wanted.start_streaming ? "true" : "false",
841
- (unsigned long long)r->wanted.after, (unsigned long long)r->wanted.before);
842
-
843
- ssize_t ret = r->caller.callback(buffer, r->caller.parser, STREAM_TRAFFIC_TYPE_REPLICATION);
844
- if (ret < 0) {
845
- netdata_log_error("STREAM SND REPLAY ERROR: 'host:%s/chart:%s' failed to send replication request to child (error %zd)",
846
- rrdhost_hostname(r->host), rrdset_id(r->st), ret);
847
- return false;
848
- }
849
-
850
- return true;
851
-}
852
-
853
-bool replicate_chart_request(send_command callback, struct parser *parser, RRDHOST *host, RRDSET *st,
854
- time_t child_first_entry, time_t child_last_entry, time_t child_wall_clock_time,
855
- time_t prev_first_entry_wanted, time_t prev_last_entry_wanted)
856
-{
857
- struct replication_request_details r = {
858
- .caller = {
859
- .callback = callback,
860
- .parser = parser,
861
- },
862
-
863
- .host = host,
864
- .st = st,
865
-
866
- .child_db = {
867
- .first_entry_t = child_first_entry,
868
- .last_entry_t = child_last_entry,
869
- .wall_clock_time = child_wall_clock_time,
870
- .fixed_last_entry = false,
871
- },
872
-
873
- .local_db = {
874
- .first_entry_t = 0,
875
- .last_entry_t = 0,
876
- .wall_clock_time = now_realtime_sec(),
877
- },
878
-
879
- .last_request = {
880
- .after = prev_first_entry_wanted,
881
- .before = prev_last_entry_wanted,
882
- },
883
-
884
- .wanted = {
885
- .after = 0,
886
- .before = 0,
887
- .start_streaming = true,
888
- },
889
- };
890
-
891
- if(r.child_db.last_entry_t > r.child_db.wall_clock_time) {
892
- replicate_log_request(&r, "child's db last entry > child's wall clock time");
893
- r.child_db.last_entry_t = r.child_db.wall_clock_time;
894
- r.child_db.fixed_last_entry = true;
895
- }
896
-
897
- rrdset_get_retention_of_tier_for_collected_chart(r.st, &r.local_db.first_entry_t, &r.local_db.last_entry_t, r.local_db.wall_clock_time, 0);
898
-
899
- // let's find the GAP we have
900
- if(!r.last_request.after || !r.last_request.before) {
901
- // there is no previous request
902
-
903
- if(r.local_db.last_entry_t)
904
- // we have some data, let's continue from the last point we have
905
- r.gap.from = r.local_db.last_entry_t;
906
- else
907
- // we don't have any data, the gap is the max timeframe we are allowed to replicate
908
- r.gap.from = r.local_db.wall_clock_time - r.host->stream.replication.period;
909
-
910
- }
911
- else {
912
- // we had sent a request - let's continue at the point we left it
913
- // for this we don't take into account the actual data in our db
914
- // because the child may also have gaps, and we need to get over it
915
- r.gap.from = r.last_request.before;
916
- }
917
-
918
- // we want all the data up to now
919
- r.gap.to = r.local_db.wall_clock_time;
920
-
921
- // The gap is now r.gap.from -> r.gap.to
922
-
923
- if (unlikely(!rrdhost_option_check(host, RRDHOST_OPTION_REPLICATION)))
924
- return send_replay_chart_cmd(&r, "empty replication request, replication is disabled", false);
925
-
926
- if (unlikely(!rrdset_number_of_dimensions(st)))
927
- return send_replay_chart_cmd(&r, "empty replication request, chart has no dimensions", false);
928
-
929
- if (unlikely(!r.child_db.first_entry_t || !r.child_db.last_entry_t))
930
- return send_replay_chart_cmd(&r, "empty replication request, child has no stored data", false);
931
-
932
- if (unlikely(r.child_db.first_entry_t < 0 || r.child_db.last_entry_t < 0))
933
- return send_replay_chart_cmd(&r, "empty replication request, child db timestamps are invalid", true);
934
-
935
- if (unlikely(r.child_db.first_entry_t > r.child_db.wall_clock_time))
936
- return send_replay_chart_cmd(&r, "empty replication request, child db first entry is after its wall clock time", true);
937
-
938
- if (unlikely(r.child_db.first_entry_t > r.child_db.last_entry_t))
939
- return send_replay_chart_cmd(&r, "empty replication request, child timings are invalid (first entry > last entry)", true);
940
-
941
- if (unlikely(r.local_db.last_entry_t > r.child_db.last_entry_t))
942
- return send_replay_chart_cmd(&r, "empty replication request, local last entry is later than the child one", false);
943
-
944
- // let's find what the child can provide to fill that gap
945
-
946
- if(r.child_db.first_entry_t > r.gap.from)
947
- // the child does not have all the data - let's get what it has
948
- r.wanted.after = r.child_db.first_entry_t;
949
- else
950
- // ok, the child can fill the entire gap we have
951
- r.wanted.after = r.gap.from;
952
-
953
- if(r.gap.to - r.wanted.after > host->stream.replication.step)
954
- // the duration is too big for one request - let's take the first step
955
- r.wanted.before = r.wanted.after + host->stream.replication.step;
956
- else
957
- // wow, we can do it in one request
958
- r.wanted.before = r.gap.to;
959
-
960
- // don't ask from the child more than it has
961
- if(r.wanted.before > r.child_db.last_entry_t)
962
- r.wanted.before = r.child_db.last_entry_t;
963
-
964
- if(r.wanted.after > r.wanted.before) {
965
- r.wanted.after = 0;
966
- r.wanted.before = 0;
967
- r.wanted.start_streaming = true;
968
- return send_replay_chart_cmd(&r, "empty replication request, wanted 'after' computed bigger than wanted 'before'", true);
969
- }
970
-
971
- // the child should start streaming immediately if the wanted duration is small, or we reached the last entry of the child
972
- r.wanted.start_streaming = (r.local_db.wall_clock_time - r.wanted.after <= host->stream.replication.step ||
973
- r.wanted.before >= r.child_db.last_entry_t ||
974
- r.wanted.before >= r.child_db.wall_clock_time ||
975
- r.wanted.before >= r.local_db.wall_clock_time);
976
-
977
- // the wanted timeframe is now r.wanted.after -> r.wanted.before
978
- // send it
979
- return send_replay_chart_cmd(&r, "OK", false);
980
-}
981
-
738
// ----------------------------------------------------------------------------
739
// replication thread
740
@@ -990,7 +746,7 @@ struct replication_request {
746
time_t after; // the start time of the query (maybe zero) key for sorting (JudyL)
747
time_t before; // the end time of the query (maybe zero)
748
993
- usec_t sender_circular_buffer_since_ut; // the timestamp of the sender, at the time we indexed this request
749
+ usec_t sender_circular_buffer_last_flush_ut; // the timestamp of the sender, at the time we indexed this request
750
Word_t unique_id; // auto-increment, later requests have bigger
751
752
bool start_streaming; // true, when the parent wants to send the rest of the data (before is overwritten) and enable normal streaming
@@ -1040,13 +796,24 @@ static struct replication_thread {
796
797
struct {
798
Word_t unique_id; // the last unique id we gave to a request (auto-increment, starting from 1)
799
+ size_t received; // the number of replication requests received
800
size_t executed; // the number of replication requests executed
801
+ size_t replied;
802
+ size_t error_not_found; // the number of replication requests ignored because the chart was not found
803
+ size_t error_duplicate; // the number of replication requests found duplicate (same chart)
804
+ size_t error_flushed; // the number of replication requests deleted due to disconnections
805
size_t latest_first_time; // the 'after' timestamp of the last request we executed
806
size_t memory; // the total memory allocated by replication
807
} atomic; // access should be with atomic operations
808
809
struct {
1049
- size_t last_executed; // caching of the atomic.executed to report number of requests executed since last time
810
+ // same as the atomic versions, for finding the delta over time
811
+ size_t last_received;
812
+ size_t last_executed;
813
+ size_t last_replied;
814
+ size_t last_error_flushed;
815
+ size_t last_error_duplicate;
816
+ size_t last_error_not_found;
817
818
ND_THREAD **threads_ptrs;
819
size_t threads;
@@ -1085,7 +852,7 @@ static struct replication_thread {
852
},
853
};
854
1088
-size_t replication_allocated_memory(void) {
855
+size_t replication_sender_allocated_memory(void) {
856
return __atomic_load_n(&replication_globals.atomic.memory, __ATOMIC_RELAXED);
857
}
858
@@ -1378,6 +1145,8 @@ static bool replication_request_conflict_callback(const DICTIONARY_ITEM *item __
1145
struct replication_request *rq = old_value; (void)rq;
1146
struct replication_request *rq_new = new_value;
1147
1148
+ __atomic_add_fetch(&replication_globals.atomic.error_duplicate, 1, __ATOMIC_RELAXED);
1149
+
1150
replication_recursive_lock();
1151
1152
if(!rq->indexed_in_judy && rq->not_indexed_buffer_full && !rq->not_indexed_preprocessing) {
@@ -1385,7 +1154,7 @@ static bool replication_request_conflict_callback(const DICTIONARY_ITEM *item __
1154
internal_error(
1155
true,
1156
"STREAM SND '%s' [to %s]: REPLAY: 'host:%s/chart:%s' replacing duplicate replication command received (existing from %llu to %llu [%s], new from %llu to %llu [%s])",
1388
- rrdhost_hostname(s->host), s->connected_to, rrdhost_hostname(s->host), dictionary_acquired_item_name(item),
1157
+ rrdhost_hostname(s->host), s->remote_ip, rrdhost_hostname(s->host), dictionary_acquired_item_name(item),
1158
(unsigned long long)rq->after, (unsigned long long)rq->before, rq->start_streaming ? "true" : "false",
1159
(unsigned long long)rq_new->after, (unsigned long long)rq_new->before, rq_new->start_streaming ? "true" : "false");
1160
@@ -1398,7 +1167,7 @@ static bool replication_request_conflict_callback(const DICTIONARY_ITEM *item __
1167
internal_error(
1168
true,
1169
"STREAM SND '%s' [to %s]: REPLAY: 'host:%s/chart:%s' adding duplicate replication command received (existing from %llu to %llu [%s], new from %llu to %llu [%s])",
1401
- rrdhost_hostname(s->host), s->connected_to, rrdhost_hostname(s->host), dictionary_acquired_item_name(item),
1170
+ rrdhost_hostname(s->host), s->remote_ip, rrdhost_hostname(s->host), dictionary_acquired_item_name(item),
1171
(unsigned long long)rq->after, (unsigned long long)rq->before, rq->start_streaming ? "true" : "false",
1172
(unsigned long long)rq_new->after, (unsigned long long)rq_new->before, rq_new->start_streaming ? "true" : "false");
1173
}
@@ -1406,7 +1175,7 @@ static bool replication_request_conflict_callback(const DICTIONARY_ITEM *item __
1175
internal_error(
1176
true,
1177
"STREAM SND '%s' [to %s]: REPLAY: 'host:%s/chart:%s' ignoring duplicate replication command received (existing from %llu to %llu [%s], new from %llu to %llu [%s])",
1409
- rrdhost_hostname(s->host), s->connected_to, rrdhost_hostname(s->host),
1178
+ rrdhost_hostname(s->host), s->remote_ip, rrdhost_hostname(s->host),
1179
dictionary_acquired_item_name(item),
1180
(unsigned long long) rq->after, (unsigned long long) rq->before, rq->start_streaming ? "true" : "false",
1181
(unsigned long long) rq_new->after, (unsigned long long) rq_new->before, rq_new->start_streaming ? "true" : "false");
@@ -1437,7 +1206,7 @@ static void replication_request_delete_callback(const DICTIONARY_ITEM *item __ma
1206
}
1207
1208
static bool sender_is_still_connected_for_this_request(struct replication_request *rq) {
1440
- return rq->sender_circular_buffer_since_ut == stream_circular_buffer_get_since_ut(rq->sender->scb);
1209
+ return rq->sender_circular_buffer_last_flush_ut == stream_circular_buffer_last_flush_ut(rq->sender->scb);
1210
}
1211
1212
static bool replication_execute_request(struct replication_request *rq, bool workers) {
@@ -1446,13 +1215,13 @@ static bool replication_execute_request(struct replication_request *rq, bool wor
1215
if(!rq->st) {
1216
if(likely(workers)) worker_is_busy(WORKER_JOB_FIND_CHART);
1217
rq->st = rrdset_find(rq->sender->host, string2str(rq->chart_id));
1449
- }
1450
-
1451
- if(!rq->st) {
1452
- internal_error(true, "STREAM SND REPLAY ERROR: 'host:%s/chart:%s' not found",
1453
- rrdhost_hostname(rq->sender->host), string2str(rq->chart_id));
1454
-
1455
- goto cleanup;
1218
+ if(!rq->st) {
1219
+ __atomic_add_fetch(&replication_globals.atomic.error_not_found, 1, __ATOMIC_RELAXED);
1220
+ nd_log(NDLS_DAEMON, NDLP_ERR,
1221
+ "STREAM SND REPLAY ERROR: 'host:%s/chart:%s' not found",
1222
+ rrdhost_hostname(rq->sender->host), string2str(rq->chart_id));
1223
+ goto cleanup;
1224
+ }
1225
}
1226
1227
if(!rq->q) {
@@ -1468,22 +1237,17 @@ static bool replication_execute_request(struct replication_request *rq, bool wor
1237
if(likely(workers)) worker_is_busy(WORKER_JOB_QUERYING);
1238
1239
// send the replication data
1240
+ size_t max_msg_size = (size_t)((unsigned long long)stream_circular_buffer_get_max_size(rq->sender->scb) * MAX_REPLICATION_MESSAGE_PERCENT_SENDER_BUFFER / 100ULL);
1241
rq->q->rq = rq;
1472
- replication_response_execute_and_finalize(
1473
- rq->q,
1474
- (size_t)((unsigned long long)stream_circular_buffer_get_max_size(rq->sender->scb) * MAX_REPLICATION_MESSAGE_PERCENT_SENDER_BUFFER / 100ULL), workers);
1475
-
1242
+ replication_response_execute_finalize_and_send(rq->q, max_msg_size, workers);
1243
rq->q = NULL;
1244
1245
__atomic_add_fetch(&replication_globals.atomic.executed, 1, __ATOMIC_RELAXED);
1479
-
1246
ret = true;
1247
1248
cleanup:
1483
- if(rq->q) {
1484
- replication_response_cancel_and_finalize(rq->q);
1485
- rq->q = NULL;
1486
- }
1249
+ replication_response_cancel_and_finalize(rq->q);
1250
+ rq->q = NULL;
1251
1252
string_freez(rq->chart_id);
1253
worker_is_idle();
@@ -1493,14 +1257,14 @@ cleanup:
1257
// ----------------------------------------------------------------------------
1258
// public API
1259
1496
-void replication_add_request(struct sender_state *sender, const char *chart_id, time_t after, time_t before, bool start_streaming) {
1260
+void replication_sender_request_add(struct sender_state *sender, const char *chart_id, time_t after, time_t before, bool start_streaming) {
1261
struct replication_request rq = {
1262
.sender = sender,
1263
.chart_id = string_strdupz(chart_id),
1264
.after = after,
1265
.before = before,
1266
.start_streaming = start_streaming,
1503
- .sender_circular_buffer_since_ut = stream_circular_buffer_get_since_ut(sender->scb),
1267
+ .sender_circular_buffer_last_flush_ut = stream_circular_buffer_last_flush_ut(sender->scb),
1268
.indexed_in_judy = false,
1269
.not_indexed_buffer_full = false,
1270
.not_indexed_preprocessing = false,
@@ -1510,10 +1274,12 @@ void replication_add_request(struct sender_state *sender, const char *chart_id,
1274
sender->replication.oldest_request_after_t = rq.after;
1275
1276
dictionary_set(sender->replication.requests, chart_id, &rq, sizeof(struct replication_request));
1277
+ __atomic_add_fetch(&replication_globals.atomic.received, 1, __ATOMIC_RELAXED);
1278
}
1279
1280
void replication_sender_delete_pending_requests(struct sender_state *sender) {
1281
// allow the dictionary destructor to go faster on locks
1282
+ __atomic_add_fetch(&replication_globals.atomic.error_flushed, dictionary_entries(sender->replication.requests), __ATOMIC_RELAXED);
1283
dictionary_flush(sender->replication.requests);
1284
sender->replication.oldest_request_after_t = 0;
1285
}
@@ -1527,14 +1293,18 @@ void replication_sender_init(struct sender_state *sender) {
1293
dictionary_register_delete_callback(sender->replication.requests, replication_request_delete_callback, sender);
1294
}
1295
1530
-void replication_cleanup_sender(struct sender_state *sender) {
1296
+void replication_sender_cleanup(struct sender_state *sender) {
1297
// allow the dictionary destructor to go faster on locks
1298
replication_recursive_lock();
1299
dictionary_destroy(sender->replication.requests);
1300
replication_recursive_unlock();
1301
}
1302
1537
-void replication_recalculate_buffer_used_ratio_unsafe(struct sender_state *s) {
1303
+static void replication_replied_add(void) {
1304
+ __atomic_add_fetch(&replication_globals.atomic.replied, 1, __ATOMIC_RELAXED);
1305
+}
1306
+
1307
+void replication_sender_recalculate_buffer_used_ratio_unsafe(struct sender_state *s) {
1308
size_t percentage = stream_sender_get_buffer_used_percent(s->scb);
1309
1310
if(unlikely(percentage > MAX_SENDER_BUFFER_PERCENTAGE_ALLOWED && !stream_sender_replication_buffer_full_get(s))) {
@@ -1574,22 +1344,30 @@ void replication_recalculate_buffer_used_ratio_unsafe(struct sender_state *s) {
1344
1345
static size_t verify_host_charts_are_streaming_now(RRDHOST *host) {
1346
internal_error(
1577
- host->sender &&
1578
- !stream_sender_pending_replication_requests(host->sender) &&
1579
- dictionary_entries(host->sender->replication.requests) != 0,
1580
- "STREAM SND REPLAY SUMMARY: 'host:%s' reports %zu pending replication requests, "
1581
- "but its chart replication index says there are %zu charts pending replication",
1582
- rrdhost_hostname(host),
1347
+ host->sender &&
1348
+ !stream_sender_pending_replication_requests(host->sender) &&
1349
+ dictionary_entries(host->sender->replication.requests) != 0,
1350
+ "STREAM SND REPLAY SUMMARY: 'host:%s' reports %zu pending replication requests, "
1351
+ "but its chart replication index says there are %zu charts pending replication",
1352
+ rrdhost_hostname(host),
1353
stream_sender_pending_replication_requests(host->sender),
1584
- dictionary_entries(host->sender->replication.requests)
1585
- );
1354
+ dictionary_entries(host->sender->replication.requests)
1355
+ );
1356
1357
size_t ok = 0;
1358
size_t errors = 0;
1359
+ size_t ignored = 0;
1360
1361
RRDSET *st;
1362
rrdset_foreach_read(st, host) {
1592
- RRDSET_FLAGS flags = rrdset_flag_check(st, RRDSET_FLAG_SENDER_REPLICATION_IN_PROGRESS | RRDSET_FLAG_SENDER_REPLICATION_FINISHED);
1363
+ RRDSET_FLAGS flags = rrdset_flag_check(
1364
+ st, RRDSET_FLAG_OBSOLETE | RRDSET_FLAG_UPSTREAM_IGNORE |
1365
+ RRDSET_FLAG_SENDER_REPLICATION_IN_PROGRESS | RRDSET_FLAG_SENDER_REPLICATION_FINISHED);
1366
+
1367
+ if(flags & (RRDSET_FLAG_OBSOLETE | RRDSET_FLAG_UPSTREAM_IGNORE)) {
1368
+ ignored++;
1369
+ continue;
1370
+ }
1371
1372
bool is_error = false;
1373
@@ -1619,8 +1397,10 @@ static size_t verify_host_charts_are_streaming_now(RRDHOST *host) {
1397
rrdset_foreach_done(st);
1398
1399
internal_error(errors,
1622
- "STREAM SND REPLAY SUMMARY: 'host:%s' finished replicating %zu charts, but %zu charts are still in progress although replication finished",
1623
- rrdhost_hostname(host), ok, errors);
1400
+ "STREAM SND REPLAY SUMMARY: 'host:%s' finished replicating %zu charts, "
1401
+ "but %zu charts are still in progress although replication finished "
1402
+ "(%zu charts are not streamed - obsolete or excluded)",
1403
+ rrdhost_hostname(host), ok, errors, ignored);
1404
1405
return errors;
1406
}
@@ -1628,16 +1408,65 @@ static size_t verify_host_charts_are_streaming_now(RRDHOST *host) {
1408
static void verify_all_hosts_charts_are_streaming_now(void) {
1409
worker_is_busy(WORKER_JOB_CHECK_CONSISTENCY);
1410
1631
- size_t errors = 0;
1411
+ size_t charts_flagged_pending = 0, entries_in_dictionaries = 0;
1412
RRDHOST *host;
1633
- dfe_start_read(rrdhost_root_index, host)
1634
- errors += verify_host_charts_are_streaming_now(host);
1413
+ dfe_start_read(rrdhost_root_index, host) {
1414
+ charts_flagged_pending += verify_host_charts_are_streaming_now(host);
1415
+
1416
+ if(host->sender)
1417
+ entries_in_dictionaries += dictionary_entries(host->sender->replication.requests);
1418
+ }
1419
dfe_done(host);
1420
1421
+ size_t flushed = __atomic_load_n(&replication_globals.atomic.error_flushed, __ATOMIC_RELAXED);
1422
+ size_t duplicate = __atomic_load_n(&replication_globals.atomic.error_duplicate, __ATOMIC_RELAXED);
1423
+ size_t not_found = __atomic_load_n(&replication_globals.atomic.error_not_found, __ATOMIC_RELAXED);
1424
+ size_t received = __atomic_load_n(&replication_globals.atomic.received, __ATOMIC_RELAXED);
1425
size_t executed = __atomic_load_n(&replication_globals.atomic.executed, __ATOMIC_RELAXED);
1638
- netdata_log_info("REPLICATION SUMMARY: finished, executed %zu replication requests, %zu charts pending replication",
1639
- executed - replication_globals.main_thread.last_executed, errors);
1426
+ size_t replied = __atomic_load_n(&replication_globals.atomic.replied, __ATOMIC_RELAXED);
1427
+
1428
+ CLEAN_BUFFER *wb = buffer_create(0, NULL);
1429
+
1430
+ if(entries_in_dictionaries) {
1431
+ if(buffer_strlen(wb)) buffer_strcat(wb, ", ");
1432
+ buffer_sprintf(wb, "%zu requests pending",
1433
+ entries_in_dictionaries);
1434
+ }
1435
+ if(charts_flagged_pending) {
1436
+ if(buffer_strlen(wb)) buffer_strcat(wb, ", ");
1437
+ buffer_sprintf(wb, "%zu instances waiting parent",
1438
+ charts_flagged_pending);
1439
+ }
1440
+ if(not_found - replication_globals.main_thread.last_error_not_found) {
1441
+ if(buffer_strlen(wb)) buffer_strcat(wb, ", ");
1442
+ buffer_sprintf(wb, "%zu ignored-not-found",
1443
+ not_found - replication_globals.main_thread.last_error_not_found);
1444
+ }
1445
+ if(duplicate - replication_globals.main_thread.last_error_duplicate) {
1446
+ if(buffer_strlen(wb)) buffer_strcat(wb, ", ");
1447
+ buffer_sprintf(wb, "%zu ignored-merged",
1448
+ duplicate - replication_globals.main_thread.last_error_duplicate);
1449
+ }
1450
+ if(flushed - replication_globals.main_thread.last_error_flushed) {
1451
+ if(buffer_strlen(wb)) buffer_strcat(wb, ", ");
1452
+ buffer_sprintf(wb, "%zu were flushed",
1453
+ flushed - replication_globals.main_thread.last_error_flushed);
1454
+ }
1455
+
1456
+ nd_log(NDLS_DAEMON, NDLP_NOTICE,
1457
+ "REPLICATION SEND SUMMARY: all senders finished replication. "
1458
+ "Received %zu, executed %zu and replied to %zu requests. %s",
1459
+ received - replication_globals.main_thread.last_received,
1460
+ executed - replication_globals.main_thread.last_executed,
1461
+ replied - replication_globals.main_thread.last_replied,
1462
+ buffer_tostring(wb));
1463
+
1464
+ replication_globals.main_thread.last_error_flushed = flushed;
1465
+ replication_globals.main_thread.last_error_duplicate = duplicate;
1466
+ replication_globals.main_thread.last_error_not_found = not_found;
1467
+ replication_globals.main_thread.last_received = received;
1468
replication_globals.main_thread.last_executed = executed;
1469
+ replication_globals.main_thread.last_replied = replied;
1470
}
1471
1472
static void replication_initialize_workers(bool master) {
@@ -1722,7 +1551,18 @@ static int replication_pipeline_execute_next(void) {
1551
struct replication_request *rq;
1552
1553
if(unlikely(!rtp.rqs)) {
1554
+#if REQUESTS_AHEAD_PER_THREAD == 0
1555
+ rtp.max_requests_ahead = (int)netdata_conf_cpus() / 2;
1556
+
1557
+ if (rtp.max_requests_ahead > libuv_worker_threads * 2)
1558
+ rtp.max_requests_ahead = libuv_worker_threads * 2;
1559
+
1560
+ if (rtp.max_requests_ahead < 5)
1561
+ rtp.max_requests_ahead = 5;
1562
+#else
1563
rtp.max_requests_ahead = REQUESTS_AHEAD_PER_THREAD;
1564
+#endif
1565
+
1566
rtp.rqs = callocz(rtp.max_requests_ahead, sizeof(struct replication_request));
1567
__atomic_add_fetch(&replication_buffers_allocated, rtp.max_requests_ahead * sizeof(struct replication_request), __ATOMIC_RELAXED);
1568
}
@@ -1742,20 +1582,22 @@ static int replication_pipeline_execute_next(void) {
1582
rq = &rtp.rqs[rtp.rqs_last_prepared];
1583
1584
if(rq->found) {
1745
- if (!rq->st) {
1746
- worker_is_busy(WORKER_JOB_FIND_CHART);
1747
- rq->st = rrdset_find(rq->sender->host, string2str(rq->chart_id));
1748
- }
1585
+ if(!rq->start_streaming) {
1586
+ if (!rq->st) {
1587
+ worker_is_busy(WORKER_JOB_FIND_CHART);
1588
+ rq->st = rrdset_find(rq->sender->host, string2str(rq->chart_id));
1589
+ }
1590
1750
- if (rq->st && !rq->q) {
1751
- worker_is_busy(WORKER_JOB_PREPARE_QUERY);
1752
- rq->q = replication_response_prepare(
1753
- rq->st,
1754
- rq->start_streaming,
1755
- rq->after,
1756
- rq->before,
1757
- rq->sender->capabilities,
1758
- rtp.max_requests_ahead == 1);
1591
+ if (rq->st && !rq->q) {
1592
+ worker_is_busy(WORKER_JOB_PREPARE_QUERY);
1593
+ rq->q = replication_response_prepare(
1594
+ rq->st,
1595
+ rq->start_streaming,
1596
+ rq->after,
1597
+ rq->before,
1598
+ rq->sender->capabilities,
1599
+ rtp.max_requests_ahead == 1);
1600
+ }
1601
}
1602
1603
rq->executed = false;
@@ -1773,7 +1615,7 @@ static int replication_pipeline_execute_next(void) {
1615
if(rq->found) {
1616
internal_fatal(rq->executed, "REPLAY FATAL: query has already been executed!");
1617
1776
- if (rq->sender_circular_buffer_since_ut != stream_circular_buffer_get_since_ut(rq->sender->scb)) {
1618
+ if (rq->sender_circular_buffer_last_flush_ut != stream_circular_buffer_last_flush_ut(rq->sender->scb)) {
1619
// the sender has reconnected since this request was queued,
1620
// we can safely throw it away, since the parent will resend it
1621
replication_response_cancel_and_finalize(rq->q);
@@ -1842,7 +1684,7 @@ static void *replication_worker_thread(void *ptr __maybe_unused) {
1684
}
1685
1686
if (unlikely(replication_pipeline_execute_next() == REQUEST_QUEUE_EMPTY)) {
1845
- sender_commit_thread_buffer_free();
1687
+ sender_thread_buffer_free();
1688
worker_is_busy(WORKER_JOB_WAIT);
1689
worker_is_idle();
1690
sleep_usec(1 * USEC_PER_SEC);
@@ -1889,7 +1731,7 @@ void *replication_thread_main(void *ptr) {
1731
1732
replication_initialize_workers(true);
1733
1892
- size_t threads = stream_conf_is_parent(false) ? (netdata_conf_cpus() / 2) : 1;
1734
+ size_t threads = netdata_conf_is_parent() ? (netdata_conf_cpus() / 3) : 1;
1735
if (threads < 1) threads = 1;
1736
else if (threads > MAX_REPLICATION_THREADS) threads = MAX_REPLICATION_THREADS;
1737
@@ -1941,7 +1783,7 @@ void *replication_thread_main(void *ptr) {
1783
1784
// statistics
1785
usec_t now_mono_ut = now_monotonic_usec();
1944
- if(unlikely(now_mono_ut - last_now_mono_ut > default_rrd_update_every * USEC_PER_SEC)) {
1786
+ if(unlikely(now_mono_ut - last_now_mono_ut > nd_profile.update_every * USEC_PER_SEC)) {
1787
last_now_mono_ut = now_mono_ut;
1788
1789
worker_is_busy(WORKER_JOB_STATISTICS);
@@ -2009,7 +1851,7 @@ void *replication_thread_main(void *ptr) {
1851
if(slow) {
1852
// no work to be done, wait for a request to come in
1853
timeout = 1000 * USEC_PER_MS;
2012
- sender_commit_thread_buffer_free();
1854
+ sender_thread_buffer_free();
1855
}
1856
1857
else if(replication_globals.unsafe.pending > 0) {
src/streaming/stream-replication-sender.h
new
+38
@@ -0,0 +1,38 @@
1
+// SPDX-License-Identifier: GPL-3.0-or-later
2
+
3
+#ifndef REPLICATION_H
4
+#define REPLICATION_H
5
+
6
+#include "daemon/common.h"
7
+#include "stream-circular-buffer.h"
8
+
9
+#ifdef __cplusplus
10
+extern "C" {
11
+#endif
12
+
13
+struct parser;
14
+
15
+struct replication_query_statistics {
16
+ SPINLOCK spinlock;
17
+ size_t queries_started;
18
+ size_t queries_finished;
19
+ size_t points_read;
20
+ size_t points_generated;
21
+};
22
+
23
+struct replication_query_statistics replication_get_query_statistics(void);
24
+
25
+void replication_sender_init(struct sender_state *sender);
26
+void replication_sender_cleanup(struct sender_state *sender);
27
+void replication_sender_delete_pending_requests(struct sender_state *sender);
28
+void replication_sender_request_add(struct sender_state *sender, const char *chart_id, time_t after, time_t before, bool start_streaming);
29
+void replication_sender_recalculate_buffer_used_ratio_unsafe(struct sender_state *s);
30
+
31
+size_t replication_sender_allocated_memory(void);
32
+size_t replication_sender_allocated_buffers(void);
33
+
34
+#ifdef __cplusplus
35
+}
36
+#endif
37
+
38
+#endif /* REPLICATION_H */
src/streaming/stream-replication-tracking.c
new
+58
@@ -0,0 +1,58 @@
1
+// SPDX-License-Identifier: GPL-3.0-or-later
2
+
3
+#include "stream-replication-tracking.h"
4
+#include "database/rrd.h"
5
+#include "stream-receiver-internals.h"
6
+
7
+#ifdef REPLICATION_TRACKING
8
+
9
+void replication_tracking_counters(struct rrdhost *host, struct replay_who_counters *c) {
10
+ if(!rrdhost_flag_check(host, RRDHOST_FLAG_COLLECTOR_ONLINE))
11
+ return;
12
+
13
+ bool is_host_local = host == localhost || rrdhost_option_check(host, RRDHOST_OPTION_VIRTUAL_HOST);
14
+ bool is_host_sending = rrdhost_flag_check(host, RRDHOST_FLAG_STREAM_SENDER_READY_4_METRICS);
15
+
16
+ RRDSET *st;
17
+ rrdset_foreach_read(st, host) {
18
+ RRDSET_FLAGS st_flags = rrdset_flag_get(st);
19
+
20
+ if(st_flags & RRDSET_FLAG_OBSOLETE)
21
+ continue;
22
+
23
+ if(!is_host_local && !(st_flags & RRDSET_FLAG_ANOMALY_DETECTION)) {
24
+ REPLAY_WHO rcv = st->stream.rcv.who;
25
+ if (rcv <= 0 || rcv >= REPLAY_WHO_MAX)
26
+ rcv = REPLAY_WHO_UNKNOWN;
27
+ c->rcv[rcv]++;
28
+
29
+#ifdef NETDATA_LOG_STREAM_RECEIVER
30
+ if(rcv == REPLAY_WHO_ME || rcv == REPLAY_WHO_THEM) {
31
+ char buf[1024];
32
+ snprintfz(buf, sizeof(buf), "### REPLICATION RECEIVE waits on %s for chart '%s'\n",
33
+ rcv == REPLAY_WHO_ME ? "me" : "them", rrdset_id(st));
34
+ stream_receiver_log_payload(host->receiver, buf, STREAM_TRAFFIC_TYPE_METADATA, rcv == REPLAY_WHO_THEM);
35
+ }
36
+#endif
37
+ }
38
+
39
+ if(is_host_sending && (st_flags & RRDSET_FLAG_UPSTREAM_SEND) && !(st_flags & RRDSET_FLAG_UPSTREAM_IGNORE)) {
40
+ REPLAY_WHO snd = st->stream.snd.who;
41
+ if (snd <= 0 || snd >= REPLAY_WHO_MAX)
42
+ snd = REPLAY_WHO_UNKNOWN;
43
+ c->snd[snd]++;
44
+
45
+#ifdef NETDATA_LOG_STREAM_SENDER
46
+ if(snd == REPLAY_WHO_ME || snd == REPLAY_WHO_THEM) {
47
+ char buf[1024];
48
+ snprintfz(buf, sizeof(buf), "### REPLICATION SEND waits on %s for chart '%s'\n",
49
+ snd == REPLAY_WHO_ME ? "me" : "them", rrdset_id(st));
50
+ stream_receiver_log_payload(host->receiver, buf, STREAM_TRAFFIC_TYPE_METADATA, snd == REPLAY_WHO_THEM);
51
+ }
52
+#endif
53
+ }
54
+ }
55
+ rrdset_foreach_done(st);
56
+}
57
+
58
+#endif
src/streaming/stream-replication-tracking.h
new
+32
@@ -0,0 +1,32 @@
1
+// SPDX-License-Identifier: GPL-3.0-or-later
2
+
3
+#ifndef NETDATA_STREAM_REPLICATION_TRACKING_H
4
+#define NETDATA_STREAM_REPLICATION_TRACKING_H
5
+
6
+#include "libnetdata/libnetdata.h"
7
+
8
+// #define REPLICATION_TRACKING 1
9
+
10
+#ifdef REPLICATION_TRACKING
11
+
12
+typedef enum __attribute__((packed)) {
13
+ REPLAY_WHO_UNKNOWN = 0, // default value
14
+ REPLAY_WHO_ME, // I have to respond
15
+ REPLAY_WHO_THEM, // they have to respond
16
+ REPLAY_WHO_FINISHED, // replication finished
17
+
18
+ // terminator
19
+ REPLAY_WHO_MAX,
20
+} REPLAY_WHO;
21
+
22
+struct replay_who_counters {
23
+ size_t rcv[REPLAY_WHO_MAX];
24
+ size_t snd[REPLAY_WHO_MAX];
25
+};
26
+
27
+struct rrdhost;
28
+void replication_tracking_counters(struct rrdhost *host, struct replay_who_counters *c);
29
+
30
+#endif
31
+
32
+#endif //NETDATA_STREAM_REPLICATION_TRACKING_H
src/streaming/stream-sender-api.c
+5
-2
@@ -1,7 +1,7 @@
1
// SPDX-License-Identifier: GPL-3.0-or-later
2
3
#include "stream-sender-internals.h"
4
-#include "replication.h"
4
+#include "stream-replication-sender.h"
5
6
bool stream_sender_has_capabilities(struct rrdhost *host, STREAM_CAPABILITIES capabilities) {
7
return host && stream_has_capability(host->sender, capabilities);
@@ -34,6 +34,7 @@ void stream_sender_structures_init(RRDHOST *host, bool stream, STRING *parents,
34
host->sender->connector.id = -1;
35
host->sender->host = host;
36
host->sender->scb = stream_circular_buffer_create();
37
+ host->sender->wait_queue = waiting_queue_create();
38
host->sender->capabilities = stream_our_capabilities(host, true);
39
40
nd_sock_init(&host->sender->sock, netdata_ssl_streaming_sender_ctx, netdata_ssl_validate_certificate_sender);
@@ -64,9 +65,11 @@ void stream_sender_structures_free(struct rrdhost *host) {
65
stream_sender_signal_to_stop_and_wait(host, STREAM_HANDSHAKE_DISCONNECT_HOST_CLEANUP, true);
66
stream_circular_buffer_destroy(host->sender->scb);
67
host->sender->scb = NULL;
68
+ waiting_queue_destroy(host->sender->wait_queue);
69
+ host->sender->wait_queue = NULL;
70
stream_compressor_destroy(&host->sender->compressor);
71
69
- replication_cleanup_sender(host->sender);
72
+ replication_sender_cleanup(host->sender);
73
74
__atomic_sub_fetch(&netdata_buffers_statistics.rrdhost_senders, sizeof(*host->sender), __ATOMIC_RELAXED);
75
src/streaming/stream-sender-commit.c
+29
-18
@@ -1,7 +1,7 @@
1
// SPDX-License-Identifier: GPL-3.0-or-later
2
3
#include "stream-thread.h"
4
-#include "replication.h"
4
+#include "stream-replication-sender.h"
5
6
static __thread struct sender_buffer commit___thread = { 0 };
7
@@ -14,7 +14,7 @@ void sender_buffer_destroy(struct sender_buffer *commit) {
14
commit->last_function = NULL;
15
}
16
17
-void sender_commit_thread_buffer_free(void) {
17
+void sender_thread_buffer_free(void) {
18
sender_buffer_destroy(&commit___thread);
19
}
20
@@ -23,14 +23,14 @@ BUFFER *sender_commit_start_with_trace(struct sender_state *s __maybe_unused, st
23
if(unlikely(commit->used))
24
fatal("STREAM SND '%s' [to %s]: thread buffer is used multiple times concurrently (%u). "
25
"It is already being used by '%s()', and now is called by '%s()'",
26
- rrdhost_hostname(s->host), s->connected_to,
26
+ rrdhost_hostname(s->host), s->remote_ip,
27
(unsigned)commit->used,
28
commit->last_function ? commit->last_function : "(null)",
29
func ? func : "(null)");
30
31
if(unlikely(commit->receiver_tid && commit->receiver_tid != gettid_cached()))
32
fatal("STREAM SND '%s' [to %s]: thread buffer is reserved for tid %d, but it used by thread %d function '%s()'.",
33
- rrdhost_hostname(s->host), s->connected_to,
33
+ rrdhost_hostname(s->host), s->remote_ip,
34
commit->receiver_tid, gettid_cached(), func ? func : "(null)");
35
36
if(unlikely(commit->wb &&
@@ -68,6 +68,9 @@ void sender_buffer_commit(struct sender_state *s, BUFFER *wb, struct sender_buff
68
if (unlikely(!src || !src_len))
69
return;
70
71
+ waiting_queue_wait(s->wait_queue, (s->host->stream.rcv.status.tid == gettid_cached()
72
+ || s->host->stream.snd.status.tid == gettid_cached()) ?
73
+ WAITING_QUEUE_PRIO_HIGH : WAITING_QUEUE_PRIO_NORMAL);
74
stream_sender_lock(s);
75
76
// copy the sequence number of sender buffer recreates, while having our lock
@@ -77,9 +80,12 @@ void sender_buffer_commit(struct sender_state *s, BUFFER *wb, struct sender_buff
80
81
if (!s->thread.msg.session) {
82
// the dispatcher is not there anymore - ignore these data
80
- stream_sender_unlock(s);
83
+
84
if(commit)
85
sender_buffer_destroy(commit);
86
+
87
+ stream_sender_unlock(s);
88
+ waiting_queue_done(s->wait_queue);
89
return;
90
}
91
@@ -88,11 +94,14 @@ void sender_buffer_commit(struct sender_state *s, BUFFER *wb, struct sender_buff
94
// adaptive sizing of the circular buffer
95
nd_log(NDLS_DAEMON, NDLP_NOTICE,
96
"STREAM SND '%s' [to %s]: Increased max buffer size to %u (message size %zu).",
91
- rrdhost_hostname(s->host), s->connected_to, stats->bytes_max_size, src_len + 1);
97
+ rrdhost_hostname(s->host), s->remote_ip, stats->bytes_max_size, src_len + 1);
98
}
99
100
stream_sender_log_payload(s, wb, type, false);
101
102
+ // if there are data already in the buffer, we don't need to send an opcode
103
+ bool enable_sending = stats->bytes_outstanding == 0;
104
+
105
if (s->compressor.initialized) {
106
// compressed traffic
107
if(rrdhost_is_this_a_stream_thread(s->host))
@@ -127,7 +136,7 @@ void sender_buffer_commit(struct sender_state *s, BUFFER *wb, struct sender_buff
136
if (!dst_len) {
137
nd_log(NDLS_DAEMON, NDLP_ERR,
138
"STREAM SND '%s' [to %s]: COMPRESSION failed. Resetting compressor and re-trying",
130
- rrdhost_hostname(s->host), s->connected_to);
139
+ rrdhost_hostname(s->host), s->remote_ip);
140
141
stream_compression_initialize(s);
142
dst_len = stream_compress(&s->compressor, src, size_to_compress, &dst);
@@ -144,7 +153,7 @@ void sender_buffer_commit(struct sender_state *s, BUFFER *wb, struct sender_buff
153
fatal(
154
"STREAM SND '%s' [to %s]: invalid signature, original payload %zu bytes, "
155
"compressed payload length %zu bytes, but signature says payload is %zu bytes",
147
- rrdhost_hostname(s->host), s->connected_to,
156
+ rrdhost_hostname(s->host), s->remote_ip,
157
size_to_compress, dst_len, decoded_dst_len);
158
#endif
159
@@ -166,13 +175,13 @@ void sender_buffer_commit(struct sender_state *s, BUFFER *wb, struct sender_buff
175
goto overflow_with_lock;
176
}
177
169
- bool enable_sending = stats->bytes_outstanding == 0;
170
- replication_recalculate_buffer_used_ratio_unsafe(s);
178
+ replication_sender_recalculate_buffer_used_ratio_unsafe(s);
179
180
if (enable_sending)
181
msg = s->thread.msg;
182
183
stream_sender_unlock(s);
184
+ waiting_queue_done(s->wait_queue);
185
186
if (enable_sending) {
187
msg.opcode = STREAM_OPCODE_SENDER_POLLOUT;
@@ -184,14 +193,15 @@ void sender_buffer_commit(struct sender_state *s, BUFFER *wb, struct sender_buff
193
overflow_with_lock: {
194
msg = s->thread.msg;
195
stream_sender_unlock(s);
196
+ waiting_queue_done(s->wait_queue);
197
msg.opcode = STREAM_OPCODE_SENDER_BUFFER_OVERFLOW;
198
stream_sender_send_opcode(s, msg);
199
nd_log_limit_static_global_var(erl, 1, 0);
200
nd_log_limit(&erl, NDLS_DAEMON, NDLP_ERR,
191
- "STREAM SND '%s' [to %s]: buffer overflow (buffer size %u, max size %u, used %u, available %u). "
201
+ "STREAM SND '%s' [to %s]: buffer overflow (buffer size %u, max size %u, available %u). "
202
"Restarting connection.",
193
- rrdhost_hostname(s->host), s->connected_to,
194
- stats->bytes_size, stats->bytes_max_size, stats->bytes_outstanding, stats->bytes_available);
203
+ rrdhost_hostname(s->host), s->remote_ip,
204
+ stats->bytes_size, stats->bytes_max_size, stats->bytes_available);
205
return;
206
}
207
@@ -199,26 +209,27 @@ compression_failed_with_lock: {
209
stream_compression_deactivate(s);
210
msg = s->thread.msg;
211
stream_sender_unlock(s);
212
+ waiting_queue_done(s->wait_queue);
213
msg.opcode = STREAM_OPCODE_SENDER_RECONNECT_WITHOUT_COMPRESSION;
214
stream_sender_send_opcode(s, msg);
215
nd_log_limit_static_global_var(erl, 1, 0);
216
nd_log_limit(&erl, NDLS_DAEMON, NDLP_ERR,
217
"STREAM SND '%s' [to %s]: COMPRESSION failed (twice). "
218
"Deactivating compression and restarting connection.",
208
- rrdhost_hostname(s->host), s->connected_to);
219
+ rrdhost_hostname(s->host), s->remote_ip);
220
}
221
}
222
212
-void sender_thread_commit(struct sender_state *s, BUFFER *wb, STREAM_TRAFFIC_TYPE type, const char *func) {
213
- struct sender_buffer *commit = (wb == commit___thread.wb) ? & commit___thread : &s->host->stream.snd.commit;
223
+void sender_thread_commit_with_trace(struct sender_state *s, BUFFER *wb, STREAM_TRAFFIC_TYPE type, const char *func) {
224
+ struct sender_buffer *commit = (wb == commit___thread.wb) ? &commit___thread : &s->host->stream.snd.commit;
225
226
if (unlikely(wb != commit->wb))
227
fatal("STREAM SND '%s' [to %s]: function '%s()' is trying to commit an unknown commit buffer.",
217
- rrdhost_hostname(s->host), s->connected_to, func);
228
+ rrdhost_hostname(s->host), s->remote_ip, func);
229
230
if (unlikely(!commit->used))
231
fatal("STREAM SND '%s' [to %s]: function '%s()' is committing a sender buffer twice.",
221
- rrdhost_hostname(s->host), s->connected_to, func);
232
+ rrdhost_hostname(s->host), s->remote_ip, func);
233
234
commit->used = false;
235
commit->last_function = NULL;
src/streaming/stream-sender-commit.h
+11
-4
@@ -22,18 +22,25 @@ void sender_buffer_destroy(struct sender_buffer *commit);
22
23
// thread buffer for sending data upstream (to a parent)
24
25
+void sender_thread_buffer_free(void);
26
BUFFER *sender_thread_buffer_with_trace(struct sender_state *s, const char *func);
27
#define sender_thread_buffer(s) sender_thread_buffer_with_trace(s, __FUNCTION__)
28
29
+// commit the global host buffer
30
+// this is the preferred buffer for stream threads (unified receiver / sender threads)
31
+// these threads require a buffer that can remain intact while switching hosts
32
BUFFER *sender_host_buffer_with_trace(struct rrdhost *host, const char *func);
33
#define sender_host_buffer(host) sender_host_buffer_with_trace(host, __FUNCTION__)
34
31
-void sender_thread_commit(struct sender_state *s, BUFFER *wb, STREAM_TRAFFIC_TYPE type, const char *func);
32
-#define sender_commit(s, wb, type) sender_thread_commit(s, wb, type, __FUNCTION__)
35
+// commit a buffer acquired with sender_thread_buffer()
36
+// this is the preferred buffer for dedicated workers sending a lot of messages (like replication)
37
+// these threads need to maintain enough allocation for repeated use of the buffer
38
+void sender_thread_commit_with_trace(struct sender_state *s, BUFFER *wb, STREAM_TRAFFIC_TYPE type, const char *func);
39
+#define sender_commit(s, wb, type) sender_thread_commit_with_trace(s, wb, type, __FUNCTION__)
40
41
+// commit any buffer
42
+// this is the preferred buffer for occasional senders, as it avoids constant buffer allocations
43
void sender_buffer_commit(struct sender_state *s, BUFFER *wb, struct sender_buffer *commit, STREAM_TRAFFIC_TYPE type);
44
#define sender_commit_clean_buffer(s, wb, type) sender_buffer_commit(s, wb, NULL, type)
45
37
-void sender_commit_thread_buffer_free(void);
38
-
46
#endif //NETDATA_STREAM_SENDER_COMMIT_H
src/streaming/stream-sender-execute.c
+17
-12
@@ -1,7 +1,7 @@
1
// SPDX-License-Identifier: GPL-3.0-or-later
2
3
#include "stream-thread.h"
4
-#include "replication.h"
4
+#include "stream-replication-sender.h"
5
6
struct inflight_stream_function {
7
struct sender_state *sender;
@@ -27,7 +27,7 @@ static void stream_execute_function_callback(BUFFER *func_wb, int code, void *da
27
sender_commit_clean_buffer(s, wb, STREAM_TRAFFIC_TYPE_FUNCTIONS);
28
29
internal_error(true, "STREAM SND '%s' [to %s]: FUNCTION transaction %s sending back response (%zu bytes, %"PRIu64" usec).",
30
- rrdhost_hostname(s->host), s->connected_to,
30
+ rrdhost_hostname(s->host), s->remote_ip,
31
string2str(tmp->transaction),
32
buffer_strlen(func_wb),
33
now_realtime_usec() - tmp->received_ut);
@@ -58,7 +58,7 @@ static void execute_commands_function(struct sender_state *s, const char *comman
58
59
if(!transaction || !*transaction || !timeout_s || !*timeout_s || !function || !*function) {
60
netdata_log_error("STREAM SND '%s' [to %s]: %s execution command is incomplete (transaction = '%s', timeout = '%s', function = '%s'). Ignoring it.",
61
- rrdhost_hostname(s->host), s->connected_to,
61
+ rrdhost_hostname(s->host), s->remote_ip,
62
command,
63
transaction?transaction:"(unset)",
64
timeout_s?timeout_s:"(unset)",
@@ -112,7 +112,7 @@ static void execute_deferred_json(struct sender_state *s, void *data) {
112
else
113
nd_log(NDLS_DAEMON, NDLP_ERR,
114
"STREAM SND '%s' [to %s]: unknown JSON keyword '%s' with payload: %s",
115
- rrdhost_hostname(s->host), s->connected_to,
115
+ rrdhost_hostname(s->host), s->remote_ip,
116
keyword, buffer_tostring(s->defer.payload));
117
}
118
@@ -268,6 +268,8 @@ void stream_sender_execute_commands(struct sender_state *s) {
268
stream_sender_log_payload(s, s->log.received, STREAM_TRAFFIC_TYPE_REPLICATION, true);
269
#endif
270
271
+ __atomic_add_fetch(&s->host->stream.snd.status.replication.counter_in, 1, __ATOMIC_RELAXED);
272
+
273
// do not log replication commands received - way too many!
274
// nd_log(NDLS_ACCESS, NDLP_DEBUG, NULL);
275
@@ -277,9 +279,9 @@ void stream_sender_execute_commands(struct sender_state *s) {
279
const char *before = get_word(s->rbuf.line.words, s->rbuf.line.num_words, 4);
280
281
if (!chart_id || !start_streaming || !after || !before) {
280
- netdata_log_error("STREAM SND '%s' [to %s] %s command is incomplete"
282
+ netdata_log_error("STREAM REPLAY ERROR '%s' [send to %s] %s command is incomplete"
283
" (chart=%s, start_streaming=%s, after=%s, before=%s)",
282
- rrdhost_hostname(s->host), s->connected_to,
284
+ rrdhost_hostname(s->host), s->remote_ip,
285
command,
286
chart_id ? chart_id : "(unset)",
287
start_streaming ? start_streaming : "(unset)",
@@ -287,11 +289,14 @@ void stream_sender_execute_commands(struct sender_state *s) {
289
before ? before : "(unset)");
290
}
291
else {
290
- replication_add_request(s, chart_id,
291
- strtoll(after, NULL, 0),
292
- strtoll(before, NULL, 0),
293
- !strcmp(start_streaming, "true")
294
- );
292
+#ifdef REPLICATION_TRACKING
293
+ RRDSET *st = rrdset_find(s->host, chart_id);
294
+ if(st)
295
+ st->stream.snd.who = REPLAY_WHO_ME;
296
+#endif
297
+
298
+ replication_sender_request_add(
299
+ s, chart_id, strtoll(after, NULL, 0), strtoll(before, NULL, 0), !strcmp(start_streaming, "true"));
300
}
301
}
302
else if(command && strcmp(command, PLUGINSD_KEYWORD_NODE_ID) == 0) {
@@ -314,7 +319,7 @@ void stream_sender_execute_commands(struct sender_state *s) {
319
}
320
else {
321
netdata_log_error("STREAM SND '%s' [to %s] received unknown command over connection: %s",
317
- rrdhost_hostname(s->host), s->connected_to, s->rbuf.line.words[0]?s->rbuf.line.words[0]:"(unset)");
322
+ rrdhost_hostname(s->host), s->remote_ip, s->rbuf.line.words[0]?s->rbuf.line.words[0]:"(unset)");
323
}
324
325
line_splitter_reset(&s->rbuf.line);
src/streaming/stream-sender-internals.h
+8
-4
@@ -40,8 +40,6 @@ struct sender_state {
40
ND_SOCK sock;
41
42
struct {
43
- bool draining_input; // used exclusively by the stream thread
44
-
43
struct stream_opcode msg; // the template for sending a message to the dispatcher - protected by sender_lock()
44
45
// this is a property of stream_sender_send_msg_to_dispatcher()
@@ -49,8 +47,8 @@ struct sender_state {
47
// DO NOT READ OR WRITE ANYWHERE
48
uint32_t msg_slot; // ensures a opcode queue that can never get full
49
50
+ nd_poll_event_t wanted;
51
usec_t last_traffic_ut;
53
-
52
struct pollfd_meta meta;
53
} thread;
54
@@ -58,9 +56,10 @@ struct sender_state {
56
int8_t id; // the connector id - protected by sender_lock()
57
} connector;
58
61
- char connected_to[CONNECTED_TO_SIZE + 1]; // We don't know which proxy we connect to, passed back from socket.c
59
+ char remote_ip[CONNECTED_TO_SIZE + 1]; // We don't know which proxy we connect to, passed back from socket.c
60
time_t last_state_since_t; // the timestamp of the last state (online/offline) change
61
62
+ WAITING_QUEUE *wait_queue;
63
STREAM_CIRCULAR_BUFFER *scb;
64
65
struct {
@@ -74,6 +73,7 @@ struct sender_state {
73
#ifdef NETDATA_LOG_STREAM_SENDER
74
struct {
75
SPINLOCK spinlock;
76
+ struct timespec first_call;
77
BUFFER *received;
78
FILE *fp;
79
} log;
@@ -85,6 +85,10 @@ struct sender_state {
85
} exit;
86
87
struct {
88
+ uint32_t last_counter_in; // copy from the host, to detect progress
89
+ uint32_t last_counter_out; // copy from the host, to detect progress
90
+ usec_t last_progress_ut; // last time we found some progress (monotonic)
91
+
92
DICTIONARY *requests; // de-duplication of replication requests, per chart
93
time_t oldest_request_after_t; // the timestamp of the oldest replication request
94
time_t latest_completed_before_t; // the timestamp of the latest replication request
src/streaming/stream-sender.c
+363
-173
@@ -2,7 +2,7 @@
2
3
#include "stream-thread.h"
4
#include "stream-sender-internals.h"
5
-#include "replication.h"
5
+#include "stream-replication-sender.h"
6
7
static void stream_sender_move_running_to_connector_or_remove(struct stream_thread *sth, struct sender_state *s, STREAM_HANDSHAKE reason, bool reconnect);
8
@@ -18,25 +18,50 @@ void stream_sender_log_payload(struct sender_state *s, BUFFER *payload, STREAM_T
18
filename, FILENAME_MAX, "/tmp/stream-sender-%s.txt", s->host ? rrdhost_hostname(s->host) : "unknown");
19
20
s->log.fp = fopen(filename, "w");
21
- }
21
23
- if(inbound) {
24
- fprintf(
25
- s->log.fp,
26
- "\n--- RECEIVE MESSAGE START: %s => %s ----\n"
27
- "%s"
28
- "--- RECEIVE MESSAGE END ----------------------------------------\n",
29
- s->connected_to, rrdhost_hostname(s->host), buffer_tostring(payload));
22
+ // Align first_call to wall clock time
23
+ clock_gettime(CLOCK_REALTIME, &s->log.first_call);
24
+ s->log.first_call.tv_nsec = 0; // Align to the start of the second
25
}
31
- else {
32
- fprintf(
33
- s->log.fp,
34
- "\n--- SEND MESSAGE START: %s => %s ----\n"
35
- "%s"
36
- "--- SEND MESSAGE END ----------------------------------------\n",
37
- rrdhost_hostname(s->host), s->connected_to, buffer_tostring(payload));
26
+
27
+ if (s->log.fp) {
28
+ struct timespec now;
29
+ clock_gettime(CLOCK_REALTIME, &now);
30
+
31
+ time_t elapsed_sec = now.tv_sec - s->log.first_call.tv_sec;
32
+ long elapsed_nsec = now.tv_nsec - s->log.first_call.tv_nsec;
33
+
34
+ if (elapsed_nsec < 0) {
35
+ elapsed_sec--;
36
+ elapsed_nsec += 1000000000;
37
+ }
38
+
39
+ uint16_t days = elapsed_sec / 86400;
40
+ uint8_t hours = (elapsed_sec % 86400) / 3600;
41
+ uint8_t minutes = (elapsed_sec % 3600) / 60;
42
+ uint8_t seconds = elapsed_sec % 60;
43
+ uint16_t milliseconds = elapsed_nsec / 1000000;
44
+
45
+ char prefix[30];
46
+ snprintf(prefix, sizeof(prefix), "%03ud.%02u:%02u:%02u.%03u ",
47
+ days, hours, minutes, seconds, milliseconds);
48
+
49
+ const char *line_start = buffer_tostring(payload);
50
+ const char *line_end;
51
+
52
+ while (line_start && *line_start) {
53
+ line_end = strchr(line_start, '\n');
54
+ if (line_end) {
55
+ fprintf(s->log.fp, "%s%s%.*s\n", prefix, inbound ? "> " : "< ", (int)(line_end - line_start), line_start);
56
+ line_start = line_end + 1;
57
+ } else {
58
+ fprintf(s->log.fp, "%s%s%s\n", prefix, inbound ? "> " : "< ", line_start);
59
+ break;
60
+ }
61
+ }
62
}
63
64
+ // fflush(s->log.fp);
65
spinlock_unlock(&s->log.spinlock);
66
}
67
#endif
@@ -50,8 +75,13 @@ static void stream_sender_charts_and_replication_reset(struct sender_state *s) {
75
// reset the state of all charts
76
RRDSET *st;
77
rrdset_foreach_read(st, s->host) {
53
- rrdset_flag_clear(st, RRDSET_FLAG_SENDER_REPLICATION_IN_PROGRESS);
54
- rrdset_flag_set(st, RRDSET_FLAG_SENDER_REPLICATION_FINISHED);
78
+ RRDSET_FLAGS old = rrdset_flag_set_and_clear(st, RRDSET_FLAG_SENDER_REPLICATION_FINISHED, RRDSET_FLAG_SENDER_REPLICATION_IN_PROGRESS);
79
+ if(!(old & RRDSET_FLAG_SENDER_REPLICATION_FINISHED))
80
+ rrdhost_sender_replicating_charts_minus_one(st->rrdhost);
81
+
82
+#ifdef REPLICATION_TRACKING
83
+ st->stream.snd.who = REPLAY_WHO_UNKNOWN;
84
+#endif
85
86
st->stream.snd.resync_time_s = 0;
87
@@ -64,8 +94,19 @@ static void stream_sender_charts_and_replication_reset(struct sender_state *s) {
94
}
95
rrdset_foreach_done(st);
96
67
- rrdhost_sender_replicating_charts_zero(s->host);
97
+ if(rrdhost_sender_replicating_charts(s->host) != 0) {
98
+ nd_log(NDLS_DAEMON, NDLP_WARNING,
99
+ "STREAM REPLAY ERROR: sender replicating instances counter should be zero, but it is %u"
100
+ " - resetting it to zero",
101
+ rrdhost_sender_replicating_charts(s->host));
102
+
103
+ rrdhost_sender_replicating_charts_zero(s->host);
104
+ }
105
+
106
stream_sender_replicating_charts_zero(s);
107
+
108
+ __atomic_store_n(&s->host->stream.snd.status.replication.counter_in, 0, __ATOMIC_RELAXED);
109
+ __atomic_store_n(&s->host->stream.snd.status.replication.counter_out, 0, __ATOMIC_RELAXED);
110
}
111
112
// --------------------------------------------------------------------------------------------------------------------
@@ -169,7 +210,6 @@ void stream_sender_handle_op(struct stream_thread *sth, struct sender_state *s,
210
ND_LOG_FIELD_CB(NDF_DST_PORT, stream_sender_log_dst_port, s),
211
ND_LOG_FIELD_CB(NDF_DST_TRANSPORT, stream_sender_log_transport, s),
212
ND_LOG_FIELD_CB(NDF_DST_CAPABILITIES, stream_sender_log_capabilities, s),
172
- ND_LOG_FIELD_UUID(NDF_MESSAGE_ID, &streaming_to_parent_msgid),
213
ND_LOG_FIELD_END(),
214
};
215
ND_LOG_STACK_PUSH(lgs);
@@ -184,7 +224,7 @@ void stream_sender_handle_op(struct stream_thread *sth, struct sender_state *s,
224
nd_log(NDLS_DAEMON, NDLP_ERR,
225
"STREAM SND[%zu] '%s' [to %s]: send buffer is full (buffer size %u, max %u, used %u, available %u). "
226
"Restarting connection.",
187
- sth->id, rrdhost_hostname(s->host), s->connected_to,
227
+ sth->id, rrdhost_hostname(s->host), s->remote_ip,
228
stats.bytes_size, stats.bytes_max_size, stats.bytes_outstanding, stats.bytes_available);
229
230
stream_sender_move_running_to_connector_or_remove(
@@ -204,7 +244,7 @@ void stream_sender_handle_op(struct stream_thread *sth, struct sender_state *s,
244
errno_clear();
245
nd_log(NDLS_DAEMON, NDLP_ERR,
246
"STREAM SND[%zu] '%s' [to %s]: restarting connection without compression.",
207
- sth->id, rrdhost_hostname(s->host), s->connected_to);
247
+ sth->id, rrdhost_hostname(s->host), s->remote_ip);
248
249
stream_sender_move_running_to_connector_or_remove(
250
sth, s, STREAM_HANDSHAKE_DISCONNECT_NOT_SUFFICIENT_SENDER_COMPRESSION_FAILED, true);
@@ -239,14 +279,27 @@ void stream_sender_move_queue_to_running_unsafe(struct stream_thread *sth) {
279
280
ND_LOG_STACK lgs[] = {
281
ND_LOG_FIELD_STR(NDF_NIDL_NODE, s->host->hostname),
242
- ND_LOG_FIELD_UUID(NDF_MESSAGE_ID, &streaming_to_parent_msgid),
282
+ ND_LOG_FIELD_CB(NDF_DST_IP, stream_sender_log_dst_ip, s),
283
+ ND_LOG_FIELD_CB(NDF_DST_PORT, stream_sender_log_dst_port, s),
284
+ ND_LOG_FIELD_CB(NDF_DST_TRANSPORT, stream_sender_log_transport, s),
285
+ ND_LOG_FIELD_CB(NDF_DST_CAPABILITIES, stream_sender_log_capabilities, s),
286
ND_LOG_FIELD_END(),
287
};
288
ND_LOG_STACK_PUSH(lgs);
289
290
nd_log(NDLS_DAEMON, NDLP_DEBUG,
291
"STREAM SND[%zu] '%s' [to %s]: moving host from dispatcher queue to dispatcher running...",
249
- sth->id, rrdhost_hostname(s->host), s->connected_to);
292
+ sth->id, rrdhost_hostname(s->host), s->remote_ip);
293
+
294
+ if(sock_setnonblock(s->sock.fd, true) != 1)
295
+ nd_log(NDLS_DAEMON, NDLP_DEBUG,
296
+ "STREAM SND[%zu] '%s' [to %s]: failed to set non-blocking mode on socket %d",
297
+ sth->id, rrdhost_hostname(s->host), s->remote_ip, s->sock.fd);
298
+
299
+ sock_setcloexec(s->sock.fd, true);
300
+ sock_enlarge_rcv_buf(s->sock.fd);
301
+ sock_enlarge_snd_buf(s->sock.fd);
302
+ sock_setcork(s->sock.fd, false);
303
304
stream_sender_lock(s);
305
s->thread.meta.type = POLLFD_TYPE_SENDER;
@@ -260,17 +313,20 @@ void stream_sender_move_queue_to_running_unsafe(struct stream_thread *sth) {
313
s->host->stream.snd.status.connections++;
314
s->last_state_since_t = now_realtime_sec();
315
316
+ s->replication.last_progress_ut = now_monotonic_usec();
317
+
318
stream_circular_buffer_flush_unsafe(s->scb, stream_send.buffer_max_size);
264
- replication_recalculate_buffer_used_ratio_unsafe(s);
319
+ replication_sender_recalculate_buffer_used_ratio_unsafe(s);
320
stream_sender_unlock(s);
321
322
internal_fatal(META_GET(&sth->run.meta, (Word_t)&s->thread.meta) != NULL, "Sender already exists in meta list");
323
META_SET(&sth->run.meta, (Word_t)&s->thread.meta, &s->thread.meta);
324
270
- if(!nd_poll_add(sth->run.ndpl, s->sock.fd, ND_POLL_READ, &s->thread.meta))
325
+ s->thread.wanted = ND_POLL_READ;
326
+ if(!nd_poll_add(sth->run.ndpl, s->sock.fd, s->thread.wanted, &s->thread.meta))
327
nd_log(NDLS_DAEMON, NDLP_ERR,
328
"STREAM SND[%zu] '%s' [to %s]: failed to add sender socket to nd_poll()",
273
- sth->id, rrdhost_hostname(s->host), s->connected_to);
329
+ sth->id, rrdhost_hostname(s->host), s->remote_ip);
330
331
stream_sender_on_ready_to_dispatch(s);
332
}
@@ -280,10 +336,6 @@ void stream_sender_remove(struct sender_state *s) {
336
// THIS FUNCTION IS USED BY THE CONNECTOR TOO
337
// when it gives up on a certain node
338
283
- nd_log(NDLS_DAEMON, NDLP_NOTICE,
284
- "STREAM SND '%s' [to %s]: streaming sender removed host: %s",
285
- rrdhost_hostname(s->host), s->connected_to, stream_handshake_error_to_string(s->exit.reason));
286
-
339
stream_sender_lock(s);
340
341
__atomic_store_n(&s->exit.shutdown, false, __ATOMIC_RELAXED);
@@ -311,16 +363,39 @@ void stream_sender_remove(struct sender_state *s) {
363
#endif
364
}
365
366
+static void stream_sender_log_disconnection(struct stream_thread *sth, struct sender_state *s, STREAM_HANDSHAKE reason) {
367
+ ND_LOG_STACK lgs[] = {
368
+ ND_LOG_FIELD_UUID(NDF_MESSAGE_ID, &streaming_to_parent_msgid),
369
+ ND_LOG_FIELD_END(),
370
+ };
371
+ ND_LOG_STACK_PUSH(lgs);
372
+
373
+ nd_log(NDLS_DAEMON, NDLP_NOTICE,
374
+ "STREAM SND[%zu] '%s' [to %s]: sender disconnected from parent, reason: %s",
375
+ sth->id, rrdhost_hostname(s->host), s->remote_ip, stream_handshake_error_to_string(reason));
376
+}
377
+
378
static void stream_sender_move_running_to_connector_or_remove(struct stream_thread *sth, struct sender_state *s, STREAM_HANDSHAKE reason, bool reconnect) {
379
internal_fatal(sth->tid != gettid_cached(), "Function %s() should only be used by the dispatcher thread", __FUNCTION__ );
380
381
+ ND_LOG_STACK lgs[] = {
382
+ ND_LOG_FIELD_STR(NDF_NIDL_NODE, s->host->hostname),
383
+ ND_LOG_FIELD_CB(NDF_DST_IP, stream_sender_log_dst_ip, s),
384
+ ND_LOG_FIELD_CB(NDF_DST_PORT, stream_sender_log_dst_port, s),
385
+ ND_LOG_FIELD_CB(NDF_DST_TRANSPORT, stream_sender_log_transport, s),
386
+ ND_LOG_FIELD_CB(NDF_DST_CAPABILITIES, stream_sender_log_capabilities, s),
387
+ ND_LOG_FIELD_END(),
388
+ };
389
+ ND_LOG_STACK_PUSH(lgs);
390
+
391
internal_fatal(META_GET(&sth->run.meta, (Word_t)&s->thread.meta) == NULL, "Sender to be removed is not in the list of senders");
392
META_DEL(&sth->run.meta, (Word_t)&s->thread.meta);
393
320
- if(!nd_poll_del(sth->run.ndpl, s->sock.fd))
394
+ s->thread.wanted = 0;
395
+ if(!nd_poll_del(sth->run.ndpl, s->sock.fd, &s->thread.meta))
396
nd_log(NDLS_DAEMON, NDLP_ERR,
397
"STREAM SND[%zu] '%s' [to %s]: failed to delete sender socket from nd_poll()",
323
- sth->id, rrdhost_hostname(s->host), s->connected_to);
398
+ sth->id, rrdhost_hostname(s->host), s->remote_ip);
399
400
// clear this flag asap, to stop other threads from pushing metrics for this node
401
rrdhost_flag_clear(s->host, RRDHOST_FLAG_STREAM_SENDER_CONNECTED | RRDHOST_FLAG_STREAM_SENDER_READY_4_METRICS);
@@ -334,9 +409,7 @@ static void stream_sender_move_running_to_connector_or_remove(struct stream_thre
409
s->host->stream.snd.status.tid = 0;
410
stream_sender_unlock(s);
411
337
- nd_log(NDLS_DAEMON, NDLP_NOTICE,
338
- "STREAM SND[%zu] '%s' [to %s]: sender disconnected from parent, reason: %s",
339
- sth->id, rrdhost_hostname(s->host), s->connected_to, stream_handshake_error_to_string(reason));
412
+ stream_sender_log_disconnection(sth, s, reason);
413
414
nd_sock_close(&s->sock);
415
@@ -378,16 +451,14 @@ void stream_sender_check_all_nodes_from_poll(struct stream_thread *sth, usec_t n
451
if(unlikely(stats.bytes_outstanding &&
452
s->thread.last_traffic_ut + stream_send.parents.timeout_s * USEC_PER_SEC < now_ut &&
453
!stream_sender_pending_replication_requests(s) &&
381
- !stream_sender_replicating_charts(s)
382
- )) {
454
+ !stream_sender_replicating_charts(s))) {
455
456
ND_LOG_STACK lgs[] = {
457
ND_LOG_FIELD_STR(NDF_NIDL_NODE, s->host->hostname),
458
ND_LOG_FIELD_CB(NDF_DST_IP, stream_sender_log_dst_ip, s),
459
ND_LOG_FIELD_CB(NDF_DST_PORT, stream_sender_log_dst_port, s),
460
ND_LOG_FIELD_CB(NDF_DST_TRANSPORT, stream_sender_log_transport, s),
389
- ND_LOG_FIELD_CB(NDF_SRC_CAPABILITIES, stream_sender_log_capabilities, s),
390
- ND_LOG_FIELD_UUID(NDF_MESSAGE_ID, &streaming_to_parent_msgid),
461
+ ND_LOG_FIELD_CB(NDF_DST_CAPABILITIES, stream_sender_log_capabilities, s),
462
ND_LOG_FIELD_END(),
463
};
464
ND_LOG_STACK_PUSH(lgs);
@@ -402,12 +473,12 @@ void stream_sender_check_all_nodes_from_poll(struct stream_thread *sth, usec_t n
473
size_snprintf(pending, sizeof(pending), stats.bytes_outstanding, "B", false);
474
475
nd_log(NDLS_DAEMON, NDLP_ERR,
405
- "STREAM SND[%zu] '%s' [to %s]: could not send data for %ld seconds - closing connection - "
476
+ "STREAM SND[%zu] '%s' [to %s]: there was not traffic for %ld seconds - closing connection - "
477
"we have sent %zu bytes in %zu operations, it is idle for %s, and we have %s pending to send "
478
"(buffer is used %.2f%%).",
408
- sth->id, rrdhost_hostname(s->host), s->connected_to, stream_send.parents.timeout_s,
479
+ sth->id, rrdhost_hostname(s->host), s->remote_ip, stream_send.parents.timeout_s,
480
stats.bytes_sent, stats.sends,
410
- duration, pending, stats.buffer_ratio);
481
+ duration, pending, stats.buffer_ratio);
482
483
stream_sender_move_running_to_connector_or_remove(sth, s, STREAM_HANDSHAKE_DISCONNECT_SOCKET_TIMEOUT, true);
484
continue;
@@ -416,10 +487,11 @@ void stream_sender_check_all_nodes_from_poll(struct stream_thread *sth, usec_t n
487
bytes_compressed += stats.bytes_added;
488
bytes_uncompressed += stats.bytes_uncompressed;
489
419
- if(!nd_poll_upd(sth->run.ndpl, s->sock.fd, ND_POLL_READ | (stats.bytes_outstanding ? ND_POLL_WRITE : 0), &s->thread.meta))
490
+ s->thread.wanted = ND_POLL_READ | (stats.bytes_outstanding ? ND_POLL_WRITE : 0);
491
+ if(!nd_poll_upd(sth->run.ndpl, s->sock.fd, s->thread.wanted, &s->thread.meta))
492
nd_log(NDLS_DAEMON, NDLP_ERR,
493
"STREAM SND[%zu] '%s' [to %s]: failed to update nd_poll().",
422
- sth->id, rrdhost_hostname(s->host), s->connected_to);
494
+ sth->id, rrdhost_hostname(s->host), s->remote_ip);
495
}
496
497
if (bytes_compressed && bytes_uncompressed) {
@@ -432,6 +504,243 @@ void stream_sender_check_all_nodes_from_poll(struct stream_thread *sth, usec_t n
504
worker_set_metric(WORKER_SENDER_JOB_BUFFER_RATIO, overall_buffer_ratio);
505
}
506
507
+static bool stream_sender_did_replication_progress(struct sender_state *s) {
508
+ RRDHOST *host = s->host;
509
+
510
+ size_t my_counter_in = __atomic_load_n(&s->replication.last_counter_in, __ATOMIC_RELAXED);
511
+ size_t my_counter_out = __atomic_load_n(&s->replication.last_counter_out, __ATOMIC_RELAXED);
512
+ size_t host_counter_in = __atomic_load_n(&host->stream.snd.status.replication.counter_in, __ATOMIC_RELAXED);
513
+ size_t host_counter_out = __atomic_load_n(&host->stream.snd.status.replication.counter_out, __ATOMIC_RELAXED);
514
+ if(my_counter_in != host_counter_in || my_counter_out != host_counter_out) {
515
+ // there has been some progress
516
+ __atomic_store_n(&s->replication.last_counter_in, __atomic_load_n(&host->stream.snd.status.replication.counter_in, __ATOMIC_RELAXED), __ATOMIC_RELAXED);
517
+ __atomic_store_n(&s->replication.last_counter_out, __atomic_load_n(&host->stream.snd.status.replication.counter_out, __ATOMIC_RELAXED), __ATOMIC_RELAXED);
518
+ s->replication.last_progress_ut = now_monotonic_usec();
519
+ return true;
520
+ }
521
+
522
+ if(!my_counter_in || !my_counter_out)
523
+ // we have not started yet
524
+ return true;
525
+
526
+ if(dictionary_entries(s->replication.requests))
527
+ // we still have requests to execute
528
+ return true;
529
+
530
+ return (now_monotonic_usec() - s->replication.last_progress_ut < 5ULL * 60 * USEC_PER_SEC);
531
+}
532
+
533
+void stream_sender_replication_check_from_poll(struct stream_thread *sth, usec_t now_ut __maybe_unused) {
534
+ internal_fatal(sth->tid != gettid_cached(), "Function %s() should only be used by the dispatcher thread", __FUNCTION__);
535
+
536
+ Word_t idx = 0;
537
+ for(struct pollfd_meta *m = META_FIRST(&sth->run.meta, &idx);
538
+ m;
539
+ m = META_NEXT(&sth->run.meta, &idx)) {
540
+ if (m->type != POLLFD_TYPE_SENDER) continue;
541
+ struct sender_state *s = m->s;
542
+ RRDHOST *host = s->host;
543
+
544
+ if(stream_sender_did_replication_progress(s))
545
+ continue;
546
+
547
+ ND_LOG_STACK lgs[] = {
548
+ ND_LOG_FIELD_STR(NDF_NIDL_NODE, host->hostname),
549
+ ND_LOG_FIELD_CB(NDF_DST_IP, stream_sender_log_dst_ip, s),
550
+ ND_LOG_FIELD_CB(NDF_DST_PORT, stream_sender_log_dst_port, s),
551
+ ND_LOG_FIELD_CB(NDF_DST_TRANSPORT, stream_sender_log_transport, s),
552
+ ND_LOG_FIELD_CB(NDF_DST_CAPABILITIES, stream_sender_log_capabilities, s),
553
+ ND_LOG_FIELD_END(),
554
+ };
555
+ ND_LOG_STACK_PUSH(lgs);
556
+
557
+ size_t exceptions = 0;
558
+ RRDSET *st;
559
+ rrdset_foreach_read(st, host) {
560
+ RRDSET_FLAGS st_flags = rrdset_flag_get(st);
561
+ if(st_flags & (RRDSET_FLAG_OBSOLETE | RRDSET_FLAG_UPSTREAM_IGNORE | RRDSET_FLAG_SENDER_REPLICATION_FINISHED))
562
+ continue;
563
+
564
+ const char *status = (st_flags & RRDSET_FLAG_SENDER_REPLICATION_IN_PROGRESS) ? "has not finished" : "has not started";
565
+
566
+ nd_log(NDLS_DAEMON, NDLP_WARNING,
567
+ "STREAM SND[%zu] '%s' [to %s]: REPLICATION STALLED: instance '%s' %s replication yet.",
568
+ sth->id, rrdhost_hostname(host), s->remote_ip,
569
+ rrdset_id(st), status);
570
+
571
+ exceptions++;
572
+ }
573
+ rrdset_foreach_done(st);
574
+
575
+ if(exceptions && !stream_sender_did_replication_progress(s)) {
576
+ nd_log(NDLS_DAEMON, NDLP_ERR,
577
+ "STREAM SND[%zu] '%s' [to %s]: REPLICATION EXCEPTIONS SUMMARY: node has %zu stalled replication requests."
578
+ "We have received %u and sent %u replication commands. "
579
+ "Disconnecting node to restore streaming.",
580
+ sth->id, rrdhost_hostname(s->host), s->remote_ip, exceptions,
581
+ __atomic_load_n(&host->stream.snd.status.replication.counter_in, __ATOMIC_RELAXED),
582
+ __atomic_load_n(&host->stream.snd.status.replication.counter_out, __ATOMIC_RELAXED));
583
+
584
+ stream_sender_move_running_to_connector_or_remove(sth, s, STREAM_HANDSHAKE_REPLICATION_STALLED, true);
585
+ }
586
+ }
587
+}
588
+
589
+bool stream_sender_send_data(struct stream_thread *sth, struct sender_state *s, usec_t now_ut, bool process_opcodes_and_enable_removal) {
590
+ internal_fatal(sth->tid != gettid_cached(), "Function %s() should only be used by the dispatcher thread", __FUNCTION__ );
591
+
592
+ EVLOOP_STATUS status = EVLOOP_STATUS_CONTINUE;
593
+ while(status == EVLOOP_STATUS_CONTINUE) {
594
+ if(!stream_sender_trylock(s)) {
595
+ sth->snd.send_misses++;
596
+ status = EVLOOP_STATUS_CANT_GET_LOCK;
597
+ break;
598
+ }
599
+
600
+ STREAM_CIRCULAR_BUFFER_STATS *stats = stream_circular_buffer_stats_unsafe(s->scb);
601
+ char *chunk;
602
+ size_t outstanding = stream_circular_buffer_get_unsafe(s->scb, &chunk);
603
+
604
+ if(!outstanding) {
605
+ status = EVLOOP_STATUS_NO_MORE_DATA;
606
+ stream_sender_unlock(s);
607
+ continue;
608
+ }
609
+
610
+ ssize_t rc = nd_sock_send_nowait(&s->sock, chunk, outstanding);
611
+ if (likely(rc > 0)) {
612
+ stream_circular_buffer_del_unsafe(s->scb, rc, now_ut);
613
+ replication_sender_recalculate_buffer_used_ratio_unsafe(s);
614
+ s->thread.last_traffic_ut = now_ut;
615
+ sth->snd.bytes_sent += rc;
616
+
617
+ if (!stats->bytes_outstanding) {
618
+ // we sent them all - remove ND_POLL_WRITE
619
+ s->thread.wanted = ND_POLL_READ;
620
+ if (!nd_poll_upd(sth->run.ndpl, s->sock.fd, s->thread.wanted, &s->thread.meta))
621
+ nd_log(NDLS_DAEMON, NDLP_ERR,
622
+ "STREAM SND[%zu] '%s' [to %s]: failed to update nd_poll().",
623
+ sth->id, rrdhost_hostname(s->host), s->remote_ip);
624
+
625
+ // recreate the circular buffer if we have to
626
+ stream_circular_buffer_recreate_timed_unsafe(s->scb, now_ut, false);
627
+ status = EVLOOP_STATUS_NO_MORE_DATA;
628
+ }
629
+ }
630
+ else if (rc == 0 || errno == ECONNRESET)
631
+ status = EVLOOP_STATUS_SOCKET_CLOSED;
632
+
633
+ else if (rc < 0) {
634
+ if(errno == EWOULDBLOCK || errno == EAGAIN || errno == EINTR)
635
+ status = EVLOOP_STATUS_SOCKET_FULL;
636
+ else
637
+ status = EVLOOP_STATUS_SOCKET_ERROR;
638
+ }
639
+ stream_sender_unlock(s);
640
+
641
+ if (status == EVLOOP_STATUS_SOCKET_ERROR || status == EVLOOP_STATUS_SOCKET_CLOSED) {
642
+ const char *disconnect_reason = NULL;
643
+ STREAM_HANDSHAKE reason;
644
+
645
+ if(status == EVLOOP_STATUS_SOCKET_ERROR) {
646
+ worker_is_busy(WORKER_STREAM_JOB_DISCONNECT_SEND_ERROR);
647
+ disconnect_reason = "socket reports error while writing";
648
+ reason = STREAM_HANDSHAKE_DISCONNECT_SOCKET_WRITE_FAILED;
649
+ }
650
+ else /* if(status == EVLOOP_STATUS_SOCKET_CLOSED) */ {
651
+ worker_is_busy(WORKER_STREAM_JOB_DISCONNECT_REMOTE_CLOSED);
652
+ disconnect_reason = "socket reports EOF (closed by parent)";
653
+ reason = STREAM_HANDSHAKE_DISCONNECT_SOCKET_CLOSED_BY_REMOTE_END;
654
+ }
655
+
656
+ nd_log(NDLS_DAEMON, NDLP_ERR,
657
+ "STREAM SND[%zu] '%s' [to %s]: %s (%zd, on fd %d) - restarting connection - "
658
+ "we have sent %zu bytes in %zu operations.",
659
+ sth->id, rrdhost_hostname(s->host), s->remote_ip, disconnect_reason, rc, s->sock.fd,
660
+ stats->bytes_sent, stats->sends);
661
+
662
+ if(process_opcodes_and_enable_removal) {
663
+ // this is not executed from the opcode handling mechanism
664
+ // so we can safely remove the sender
665
+ stream_sender_move_running_to_connector_or_remove(sth, s, reason, true);
666
+ }
667
+ else {
668
+ // protection against this case:
669
+ //
670
+ // 1. sender gets a function request
671
+ // 2. sender executes the request
672
+ // 3. response is immediately available
673
+ // 4. sender_commit() appends the data to the sending circular buffer
674
+ // 5. sender_commit() sends opcode to enable sending
675
+ // 6. opcode bypasses the signal and runs this function inline to dispatch immediately
676
+ // 7. sending fails (remote disconnected)
677
+ // 8. sender is removed
678
+ //
679
+ // Point 2 above crashes. The sender is no longer there (freed at point 8)
680
+ // and there is no way for point 2 to know...
681
+ }
682
+ }
683
+ else if(process_opcodes_and_enable_removal &&
684
+ status == EVLOOP_STATUS_CONTINUE &&
685
+ stream_thread_process_opcodes(sth, &s->thread.meta))
686
+ status = EVLOOP_STATUS_OPCODE_ON_ME;
687
+ }
688
+
689
+ return EVLOOP_STATUS_STILL_ALIVE(status);
690
+}
691
+
692
+bool stream_sender_receive_data(struct stream_thread *sth, struct sender_state *s, usec_t now_ut, bool process_opcodes) {
693
+ EVLOOP_STATUS status = EVLOOP_STATUS_CONTINUE;
694
+ while(status == EVLOOP_STATUS_CONTINUE) {
695
+ ssize_t rc = nd_sock_revc_nowait(&s->sock, s->rbuf.b + s->rbuf.read_len, sizeof(s->rbuf.b) - s->rbuf.read_len - 1);
696
+ if (likely(rc > 0)) {
697
+ s->rbuf.read_len += rc;
698
+
699
+ s->thread.last_traffic_ut = now_ut;
700
+ sth->snd.bytes_received += rc;
701
+
702
+ worker_is_busy(WORKER_SENDER_JOB_EXECUTE);
703
+ stream_sender_execute_commands(s);
704
+ }
705
+ else if (rc == 0 || errno == ECONNRESET)
706
+ status = EVLOOP_STATUS_SOCKET_CLOSED;
707
+
708
+ else if (rc < 0) {
709
+ if(errno == EWOULDBLOCK || errno == EAGAIN || errno == EINTR)
710
+ status = EVLOOP_STATUS_SOCKET_FULL;
711
+ else
712
+ status = EVLOOP_STATUS_SOCKET_ERROR;
713
+ }
714
+
715
+ if(status == EVLOOP_STATUS_SOCKET_ERROR || status == EVLOOP_STATUS_SOCKET_CLOSED) {
716
+ const char *disconnect_reason;
717
+ STREAM_HANDSHAKE reason;
718
+
719
+ if(status == EVLOOP_STATUS_SOCKET_ERROR) {
720
+ worker_is_busy(WORKER_STREAM_JOB_DISCONNECT_RECEIVE_ERROR);
721
+ reason = STREAM_HANDSHAKE_DISCONNECT_SOCKET_READ_FAILED;
722
+ disconnect_reason = "error during receive";
723
+ }
724
+ else /* if(status == EVLOOP_STATUS_SOCKET_CLOSED) */ {
725
+ worker_is_busy(WORKER_STREAM_JOB_DISCONNECT_REMOTE_CLOSED);
726
+ reason = STREAM_HANDSHAKE_DISCONNECT_SOCKET_CLOSED_BY_REMOTE_END;
727
+ disconnect_reason = "socket reports EOF (closed by parent)";
728
+ }
729
+
730
+ nd_log(NDLS_DAEMON, NDLP_ERR,
731
+ "STREAM SND[%zu] '%s' [to %s]: %s (fd %d) - restarting sender connection.",
732
+ sth->id, rrdhost_hostname(s->host), s->remote_ip, disconnect_reason, s->sock.fd);
733
+
734
+ stream_sender_move_running_to_connector_or_remove(
735
+ sth, s, reason, true);
736
+ }
737
+ else if(status == EVLOOP_STATUS_CONTINUE && process_opcodes && stream_thread_process_opcodes(sth, &s->thread.meta))
738
+ status = EVLOOP_STATUS_OPCODE_ON_ME;
739
+ }
740
+
741
+ return EVLOOP_STATUS_STILL_ALIVE(status);
742
+}
743
+
744
// process poll() events for streaming senders
745
// returns true when the sender is still there, false if it removed it
746
bool stream_sender_process_poll_events(struct stream_thread *sth, struct sender_state *s, nd_poll_event_t events, usec_t now_ut) {
@@ -442,8 +751,7 @@ bool stream_sender_process_poll_events(struct stream_thread *sth, struct sender_
751
ND_LOG_FIELD_CB(NDF_DST_IP, stream_sender_log_dst_ip, s),
752
ND_LOG_FIELD_CB(NDF_DST_PORT, stream_sender_log_dst_port, s),
753
ND_LOG_FIELD_CB(NDF_DST_TRANSPORT, stream_sender_log_transport, s),
445
- ND_LOG_FIELD_CB(NDF_SRC_CAPABILITIES, stream_sender_log_capabilities, s),
446
- ND_LOG_FIELD_UUID(NDF_MESSAGE_ID, &streaming_to_parent_msgid),
754
+ ND_LOG_FIELD_CB(NDF_DST_CAPABILITIES, stream_sender_log_capabilities, s),
755
ND_LOG_FIELD_END(),
756
};
757
ND_LOG_STACK_PUSH(lgs);
@@ -471,129 +779,22 @@ bool stream_sender_process_poll_events(struct stream_thread *sth, struct sender_
779
780
nd_log(NDLS_DAEMON, NDLP_ERR,
781
"STREAM SND[%zu] '%s' [to %s]: %s restarting connection - %zu bytes transmitted in %zu operations.",
474
- sth->id, rrdhost_hostname(s->host), s->connected_to, error, stats.bytes_sent, stats.sends);
782
+ sth->id, rrdhost_hostname(s->host), s->remote_ip, error, stats.bytes_sent, stats.sends);
783
784
stream_sender_move_running_to_connector_or_remove(sth, s, STREAM_HANDSHAKE_DISCONNECT_SOCKET_ERROR, true);
785
return false;
786
}
787
480
- if(events & ND_POLL_WRITE) {
481
- // we can send data on this socket
482
-
483
- bool stop = false;
484
- while(!stop) {
485
- if(stream_sender_trylock(s)) {
486
- worker_is_busy(WORKER_STREAM_JOB_SOCKET_SEND);
487
-
488
- const char *disconnect_reason = NULL;
489
- STREAM_HANDSHAKE reason;
490
-
491
- STREAM_CIRCULAR_BUFFER_STATS *stats = stream_circular_buffer_stats_unsafe(s->scb);
492
- char *chunk;
493
- size_t outstanding = stream_circular_buffer_get_unsafe(s->scb, &chunk);
494
- ssize_t rc = nd_sock_send_nowait(&s->sock, chunk, outstanding);
495
- if (likely(rc > 0)) {
496
- stream_circular_buffer_del_unsafe(s->scb, rc);
497
- replication_recalculate_buffer_used_ratio_unsafe(s);
498
- s->thread.last_traffic_ut = now_ut;
499
- sth->snd.bytes_sent += rc;
500
-
501
- if (!stats->bytes_outstanding) {
502
- // we sent them all - remove ND_POLL_WRITE
503
- if (!nd_poll_upd(sth->run.ndpl, s->sock.fd, ND_POLL_READ, &s->thread.meta))
504
- nd_log(NDLS_DAEMON, NDLP_ERR,
505
- "STREAM SND[%zu] '%s' [to %s]: failed to update nd_poll().",
506
- sth->id, rrdhost_hostname(s->host), s->connected_to);
507
-
508
- // recreate the circular buffer if we have to
509
- stream_circular_buffer_recreate_timed_unsafe(s->scb, now_ut, false);
510
- stop = true;
511
- }
512
- else if(stream_thread_process_opcodes(sth, &s->thread.meta))
513
- stop = true;
514
- }
515
- else if (rc == 0 || errno == ECONNRESET) {
516
- disconnect_reason = "socket reports EOF (closed by parent)";
517
- reason = STREAM_HANDSHAKE_DISCONNECT_SOCKET_CLOSED_BY_REMOTE_END;
518
- }
519
- else if (rc < 0) {
520
- if(errno == EWOULDBLOCK || errno == EAGAIN || errno == EINTR) {
521
- // will try later
522
- stop = true;
523
- }
524
- else {
525
- disconnect_reason = "socket reports error while writing";
526
- reason = STREAM_HANDSHAKE_DISCONNECT_SOCKET_WRITE_FAILED;
527
- }
528
- }
529
- stream_sender_unlock(s);
530
-
531
- if (disconnect_reason) {
532
- worker_is_busy(WORKER_SENDER_JOB_DISCONNECT_SEND_ERROR);
533
- nd_log(NDLS_DAEMON, NDLP_ERR,
534
- "STREAM SND[%zu] '%s' [to %s]: %s (%zd, on fd %d) - restarting connection - "
535
- "we have sent %zu bytes in %zu operations.",
536
- sth->id, rrdhost_hostname(s->host), s->connected_to, disconnect_reason, rc, s->sock.fd,
537
- stats->bytes_sent, stats->sends);
538
-
539
- stream_sender_move_running_to_connector_or_remove(sth, s, reason, true);
540
- return false;
541
- }
542
- }
543
- else {
544
- sth->snd.send_misses++;
545
- break;
546
- }
547
- }
788
+ if(events & ND_POLL_READ) {
789
+ worker_is_busy(WORKER_STREAM_JOB_SOCKET_RECEIVE);
790
+ if(!stream_sender_receive_data(sth, s, now_ut, true))
791
+ return false;
792
}
793
550
- if(!(events & ND_POLL_READ))
551
- return true;
552
-
553
- // we can receive data from this socket
554
-
555
- worker_is_busy(WORKER_STREAM_JOB_SOCKET_RECEIVE);
556
- bool stop = false;
557
- size_t iterations = 0;
558
- while(!stop && iterations++ < MAX_IO_ITERATIONS_PER_EVENT) {
559
- // we have to drain the socket!
560
-
561
- ssize_t rc = nd_sock_revc_nowait(&s->sock, s->rbuf.b + s->rbuf.read_len, sizeof(s->rbuf.b) - s->rbuf.read_len - 1);
562
- if (likely(rc > 0)) {
563
- s->rbuf.read_len += rc;
564
-
565
- s->thread.last_traffic_ut = now_ut;
566
- sth->snd.bytes_received += rc;
567
-
568
- worker_is_busy(WORKER_SENDER_JOB_EXECUTE);
569
- stream_sender_execute_commands(s);
570
-
571
- if(stream_thread_process_opcodes(sth, &s->thread.meta))
572
- stop = true;
573
- }
574
- else if (rc == 0 || errno == ECONNRESET) {
575
- worker_is_busy(WORKER_SENDER_JOB_DISCONNECT_REMOTE_CLOSED);
576
- nd_log(NDLS_DAEMON, NDLP_ERR,
577
- "STREAM SND[%zu] '%s' [to %s]: socket %d reports EOF (closed by parent).",
578
- sth->id, rrdhost_hostname(s->host), s->connected_to, s->sock.fd);
579
- stream_sender_move_running_to_connector_or_remove(
580
- sth, s, STREAM_HANDSHAKE_DISCONNECT_SOCKET_CLOSED_BY_REMOTE_END, true);
794
+ if(events & ND_POLL_WRITE) {
795
+ worker_is_busy(WORKER_STREAM_JOB_SOCKET_SEND);
796
+ if(!stream_sender_send_data(sth, s, now_ut, true))
797
return false;
582
- }
583
- else if (rc < 0) {
584
- if(errno == EWOULDBLOCK || errno == EAGAIN || errno == EINTR)
585
- // will try later
586
- stop = true;
587
- else {
588
- worker_is_busy(WORKER_SENDER_JOB_DISCONNECT_RECEIVE_ERROR);
589
- nd_log(NDLS_DAEMON, NDLP_ERR,
590
- "STREAM SND[%zu] '%s' [to %s]: error during receive (%zd, on fd %d) - restarting connection.",
591
- sth->id, rrdhost_hostname(s->host), s->connected_to, rc, s->sock.fd);
592
- stream_sender_move_running_to_connector_or_remove(
593
- sth, s, STREAM_HANDSHAKE_DISCONNECT_SOCKET_READ_FAILED, true);
594
- return false;
595
- }
596
- }
798
}
799
800
return true;
@@ -608,17 +809,6 @@ void stream_sender_cleanup(struct stream_thread *sth) {
809
if(m->type != POLLFD_TYPE_SENDER) continue;
810
struct sender_state *s = m->s;
811
611
- ND_LOG_STACK lgs[] = {
612
- ND_LOG_FIELD_STR(NDF_NIDL_NODE, s->host->hostname),
613
- ND_LOG_FIELD_CB(NDF_DST_IP, stream_sender_log_dst_ip, s),
614
- ND_LOG_FIELD_CB(NDF_DST_PORT, stream_sender_log_dst_port, s),
615
- ND_LOG_FIELD_CB(NDF_DST_TRANSPORT, stream_sender_log_transport, s),
616
- ND_LOG_FIELD_CB(NDF_SRC_CAPABILITIES, stream_sender_log_capabilities, s),
617
- ND_LOG_FIELD_UUID(NDF_MESSAGE_ID, &streaming_to_parent_msgid),
618
- ND_LOG_FIELD_END(),
619
- };
620
- ND_LOG_STACK_PUSH(lgs);
621
-
812
s->exit.reason = STREAM_HANDSHAKE_DISCONNECT_SHUTDOWN;
813
s->exit.shutdown = true;
814
stream_sender_move_running_to_connector_or_remove(sth, s, STREAM_HANDSHAKE_DISCONNECT_SHUTDOWN, false);
src/streaming/stream-thread.c
+50
-26
@@ -29,12 +29,18 @@ static void stream_thread_handle_op(struct stream_thread *sth, struct stream_opc
29
{
30
if(m->type == POLLFD_TYPE_SENDER) {
31
if(msg->opcode & STREAM_OPCODE_SENDER_POLLOUT) {
32
- if(!nd_poll_upd(sth->run.ndpl, m->s->sock.fd, ND_POLL_READ|ND_POLL_WRITE, m)) {
32
+ m->s->thread.wanted = ND_POLL_READ | ND_POLL_WRITE;
33
+ if(!nd_poll_upd(sth->run.ndpl, m->s->sock.fd, m->s->thread.wanted, m)) {
34
nd_log_limit_static_global_var(erl, 1, 0);
35
nd_log_limit(&erl, NDLS_DAEMON, NDLP_DEBUG,
36
"STREAM SND[%zu] '%s' [to %s]: cannot enable output on sender socket %d.",
36
- sth->id, rrdhost_hostname(m->s->host), m->s->connected_to, m->s->sock.fd);
37
+ sth->id, rrdhost_hostname(m->s->host), m->s->remote_ip, m->s->sock.fd);
38
}
39
+
40
+ if(!stream_sender_send_data(sth, m->s, now_monotonic_usec(), false))
41
+ // sender has been removed
42
+ return;
43
+
44
msg->opcode &= ~(STREAM_OPCODE_SENDER_POLLOUT);
45
}
46
@@ -43,12 +49,18 @@ static void stream_thread_handle_op(struct stream_thread *sth, struct stream_opc
49
}
50
else if(m->type == POLLFD_TYPE_RECEIVER) {
51
if (msg->opcode & STREAM_OPCODE_RECEIVER_POLLOUT) {
46
- if (!nd_poll_upd(sth->run.ndpl, m->rpt->sock.fd, ND_POLL_READ | ND_POLL_WRITE, m)) {
52
+ m->rpt->thread.wanted = ND_POLL_READ | ND_POLL_WRITE;
53
+ if (!nd_poll_upd(sth->run.ndpl, m->rpt->sock.fd, m->rpt->thread.wanted, m)) {
54
nd_log_limit_static_global_var(erl, 1, 0);
48
- nd_log_limit(&erl, NDLS_DAEMON, NDLP_DEBUG,
55
+ nd_log_limit(&erl, NDLS_DAEMON, NDLP_ERR,
56
"STREAM RCV[%zu] '%s' [from [%s]:%s]: cannot enable output on receiver socket %d.",
50
- sth->id, rrdhost_hostname(m->rpt->host), m->rpt->client_ip, m->rpt->client_port, m->rpt->sock.fd);
57
+ sth->id, rrdhost_hostname(m->rpt->host), m->rpt->remote_ip, m->rpt->remote_port, m->rpt->sock.fd);
58
}
59
+
60
+ if(!stream_receiver_send_data(sth, m->rpt, now_monotonic_usec(), false))
61
+ // receiver has been removed
62
+ return;
63
+
64
msg->opcode &= ~(STREAM_OPCODE_RECEIVER_POLLOUT);
65
}
66
@@ -84,19 +96,19 @@ void stream_receiver_send_opcode(struct receiver_state *rpt, struct stream_opcod
96
if(msg.meta != &rpt->thread.meta) {
97
nd_log(NDLS_DAEMON, NDLP_ERR,
98
"STREAM RCV '%s' [from [%s]:%s]: the receiver in the opcode the message does not match this receiver. "
87
- "Ignoring opcode.", rrdhost_hostname(rpt->host), rpt->client_ip, rpt->client_port);
99
+ "Ignoring opcode.", rrdhost_hostname(rpt->host), rpt->remote_ip, rpt->remote_port);
100
return;
101
}
102
struct stream_thread *sth = stream_thread_by_slot_id(msg.thread_slot);
103
if(!sth) {
104
nd_log(NDLS_DAEMON, NDLP_ERR,
105
"STREAM RCV '%s' [from [%s]:%s]: the opcode (%u) message cannot be verified. Ignoring it.",
94
- rrdhost_hostname(rpt->host), rpt->client_ip, rpt->client_port, msg.opcode);
106
+ rrdhost_hostname(rpt->host), rpt->remote_ip, rpt->remote_port, msg.opcode);
107
return;
108
}
109
110
// check if we can execute the message now
99
- if(sth->tid == gettid_cached() && (!rpt->thread.draining_input || msg.opcode == STREAM_OPCODE_RECEIVER_POLLOUT)) {
111
+ if(sth->tid == gettid_cached() && msg.opcode == STREAM_OPCODE_RECEIVER_POLLOUT) {
112
// we are running at the stream thread, and the request is about enabling POLLOUT,
113
// we can do this synchronously.
114
// IMPORTANT: DO NOT HANDLE FAILURES THAT REMOVE THE RECEIVER OR THE SENDER THIS WAY
@@ -138,7 +150,7 @@ void stream_receiver_send_opcode(struct receiver_state *rpt, struct stream_opcod
150
#endif
151
152
fatal("STREAM RCV '%s' [from [%s]:%s]: The streaming opcode queue is full, but this should never happen...",
141
- rrdhost_hostname(rpt->host), rpt->client_ip, rpt->client_port);
153
+ rrdhost_hostname(rpt->host), rpt->remote_ip, rpt->remote_port);
154
}
155
156
// let's use a new slot
@@ -164,7 +176,7 @@ void stream_sender_send_opcode(struct sender_state *s, struct stream_opcode msg)
176
if(msg.meta != &s->thread.meta) {
177
nd_log(NDLS_DAEMON, NDLP_ERR,
178
"STREAM SND '%s' [to %s]: the opcode message does not match this sender. "
167
- "Ignoring opcode.", rrdhost_hostname(s->host), s->connected_to);
179
+ "Ignoring opcode.", rrdhost_hostname(s->host), s->remote_ip);
180
return;
181
}
182
@@ -172,12 +184,12 @@ void stream_sender_send_opcode(struct sender_state *s, struct stream_opcode msg)
184
if(!sth) {
185
nd_log(NDLS_DAEMON, NDLP_ERR,
186
"STREAM SND[x] '%s' [to %s] the opcode (%u) message cannot be verified. Ignoring it.",
175
- rrdhost_hostname(s->host), s->connected_to, msg.opcode);
187
+ rrdhost_hostname(s->host), s->remote_ip, msg.opcode);
188
return;
189
}
190
191
// check if we can execute the message now
180
- if(sth->tid == gettid_cached() && (!s->thread.draining_input || msg.opcode == STREAM_OPCODE_SENDER_POLLOUT)) {
192
+ if(sth->tid == gettid_cached() && msg.opcode == STREAM_OPCODE_SENDER_POLLOUT) {
193
// we are running at the stream thread, and the request is about enabling POLLOUT,
194
// we can do this synchronously.
195
// IMPORTANT: DO NOT HANDLE FAILURES THAT REMOVE THE RECEIVER OR THE SENDER THIS WAY
@@ -219,7 +231,7 @@ void stream_sender_send_opcode(struct sender_state *s, struct stream_opcode msg)
231
#endif
232
233
fatal("STREAM SND '%s' [to %s]: The streaming opcode queue is full, but this should never happen...",
222
- rrdhost_hostname(s->host), s->connected_to);
234
+ rrdhost_hostname(s->host), s->remote_ip);
235
}
236
237
// let's use a new slot
@@ -326,18 +338,19 @@ static bool stream_thread_process_poll_slot(struct stream_thread *sth, nd_poll_r
338
switch(m->type) {
339
case POLLFD_TYPE_SENDER: {
340
struct sender_state *s = m->s;
329
- s->thread.draining_input = true;
330
- if(stream_sender_process_poll_events(sth, s, ev->events, now_ut))
331
- s->thread.draining_input = false;
332
- *replay_entries += dictionary_entries(s->replication.requests);
341
+ if(stream_sender_process_poll_events(sth, s, ev->events, now_ut)) {
342
+ // the sender is still there
343
+ *replay_entries += dictionary_entries(s->replication.requests);
344
+ }
345
break;
346
}
347
348
case POLLFD_TYPE_RECEIVER: {
349
struct receiver_state *rpt = m->rpt;
338
- rpt->thread.draining_input = true;
339
- if(stream_receive_process_poll_events(sth, rpt, ev->events, now_ut))
340
- rpt->thread.draining_input = false;
350
+ if(stream_receive_process_poll_events(sth, rpt, ev->events, now_ut)) {
351
+ // the receiver is still there
352
+ ;
353
+ }
354
break;
355
}
356
@@ -394,9 +407,9 @@ void *stream_thread(void *ptr) {
407
worker_register_job_name(WORKER_SENDER_JOB_DISCONNECT_OVERFLOW, "disconnect overflow");
408
worker_register_job_name(WORKER_SENDER_JOB_DISCONNECT_TIMEOUT, "disconnect timeout");
409
worker_register_job_name(WORKER_SENDER_JOB_DISCONNECT_SOCKET_ERROR, "disconnect socket error");
397
- worker_register_job_name(WORKER_SENDER_JOB_DISCONNECT_REMOTE_CLOSED, "disconnect remote closed");
398
- worker_register_job_name(WORKER_SENDER_JOB_DISCONNECT_RECEIVE_ERROR, "disconnect receive error");
399
- worker_register_job_name(WORKER_SENDER_JOB_DISCONNECT_SEND_ERROR, "disconnect send error");
410
+ worker_register_job_name(WORKER_STREAM_JOB_DISCONNECT_REMOTE_CLOSED, "disconnect remote closed");
411
+ worker_register_job_name(WORKER_STREAM_JOB_DISCONNECT_RECEIVE_ERROR, "disconnect receive error");
412
+ worker_register_job_name(WORKER_STREAM_JOB_DISCONNECT_SEND_ERROR, "disconnect send error");
413
worker_register_job_name(WORKER_SENDER_JOB_DISCONNECT_COMPRESSION_ERROR, "disconnect compression error");
414
worker_register_job_name(WORKER_SENDER_JOB_DISCONNECT_RECEIVER_LEFT, "disconnect receiver left");
415
worker_register_job_name(WORKER_SENDER_JOB_DISCONNECT_HOST_CLEANUP, "disconnect host cleanup");
@@ -471,8 +484,8 @@ void *stream_thread(void *ptr) {
484
sth->pipe.size = set_pipe_size(sth->pipe.fds[PIPE_READ], 65536 * sizeof(*sth->pipe.buffer)) / sizeof(*sth->pipe.buffer);
485
sth->pipe.buffer = mallocz(sth->pipe.size * sizeof(*sth->pipe.buffer));
486
474
- usec_t last_check_all_nodes_ut = 0;
475
- usec_t last_dequeue_ut = 0;
487
+ usec_t last_check_replication_ut, last_check_all_nodes_ut, last_dequeue_ut;
488
+ last_check_replication_ut = last_check_all_nodes_ut = last_dequeue_ut = now_monotonic_usec();
489
490
sth->run.pipe = (struct pollfd_meta){
491
.type = POLLFD_TYPE_PIPE,
@@ -515,12 +528,14 @@ void *stream_thread(void *ptr) {
528
last_dequeue_ut = now_ut;
529
}
530
518
- if(now_ut - last_check_all_nodes_ut >= USEC_PER_SEC) {
531
+ if(now_ut - last_check_all_nodes_ut >= nd_profile.update_every * USEC_PER_SEC) {
532
worker_is_busy(WORKER_STREAM_JOB_LIST);
533
534
// periodically check the entire list of nodes
535
// this detects unresponsive parents too (timeout)
536
stream_sender_check_all_nodes_from_poll(sth, now_ut);
537
+ stream_receiver_check_all_nodes_from_poll(sth, now_ut);
538
+
539
worker_set_metric(WORKER_SENDER_JOB_MESSAGES, (NETDATA_DOUBLE)(sth->messages.processed));
540
worker_set_metric(WORKER_STREAM_METRIC_NODES, (NETDATA_DOUBLE)sth->nodes_count);
541
@@ -537,6 +552,15 @@ void *stream_thread(void *ptr) {
552
last_check_all_nodes_ut = now_ut;
553
}
554
555
+ if(now_ut - last_check_replication_ut >= 10 * 60 * USEC_PER_SEC) {
556
+ worker_is_busy(WORKER_STREAM_JOB_LIST);
557
+
558
+ stream_sender_replication_check_from_poll(sth, now_ut);
559
+ stream_receiver_replication_check_from_poll(sth, now_ut);
560
+
561
+ last_check_replication_ut = now_ut;
562
+ }
563
+
564
worker_is_idle();
565
566
nd_poll_result_t ev;
src/streaming/stream-thread.h
+22
-4
@@ -64,9 +64,9 @@ struct stream_opcode {
64
#define WORKER_SENDER_JOB_DISCONNECT_OVERFLOW 16
65
#define WORKER_SENDER_JOB_DISCONNECT_TIMEOUT 17
66
#define WORKER_SENDER_JOB_DISCONNECT_SOCKET_ERROR 18
67
-#define WORKER_SENDER_JOB_DISCONNECT_REMOTE_CLOSED 19
68
-#define WORKER_SENDER_JOB_DISCONNECT_RECEIVE_ERROR 20
69
-#define WORKER_SENDER_JOB_DISCONNECT_SEND_ERROR 21
67
+#define WORKER_STREAM_JOB_DISCONNECT_REMOTE_CLOSED 19
68
+#define WORKER_STREAM_JOB_DISCONNECT_RECEIVE_ERROR 20
69
+#define WORKER_STREAM_JOB_DISCONNECT_SEND_ERROR 21
70
#define WORKER_SENDER_JOB_DISCONNECT_COMPRESSION_ERROR 22
71
#define WORKER_SENDER_JOB_DISCONNECT_RECEIVER_LEFT 23
72
#define WORKER_SENDER_JOB_DISCONNECT_HOST_CLEANUP 24
@@ -95,7 +95,18 @@ struct stream_opcode {
95
#define STREAM_MAX_THREADS 2048
96
#define THREAD_TAG_STREAM "STREAM"
97
98
-#define MAX_IO_ITERATIONS_PER_EVENT 65536 // drain the input, take it all
98
+typedef enum {
99
+ EVLOOP_STATUS_CONTINUE,
100
+ EVLOOP_STATUS_SOCKET_FULL,
101
+ EVLOOP_STATUS_SOCKET_CLOSED,
102
+ EVLOOP_STATUS_SOCKET_ERROR,
103
+ EVLOOP_STATUS_NO_MORE_DATA,
104
+ EVLOOP_STATUS_OPCODE_ON_ME,
105
+ EVLOOP_STATUS_CANT_GET_LOCK,
106
+ EVLOOP_STATUS_PARSER_FAILED,
107
+} EVLOOP_STATUS;
108
+
109
+#define EVLOOP_STATUS_STILL_ALIVE(status) (status == EVLOOP_STATUS_CONTINUE || status == EVLOOP_STATUS_NO_MORE_DATA || status == EVLOOP_STATUS_SOCKET_FULL || status == EVLOOP_STATUS_CANT_GET_LOCK)
110
111
typedef enum {
112
POLLFD_TYPE_EMPTY,
@@ -193,6 +204,7 @@ extern struct stream_thread_globals stream_thread_globals;
204
void stream_sender_move_queue_to_running_unsafe(struct stream_thread *sth);
205
void stream_receiver_move_entire_queue_to_running_unsafe(struct stream_thread *sth);
206
void stream_sender_check_all_nodes_from_poll(struct stream_thread *sth, usec_t now_ut);
207
+void stream_sender_replication_check_from_poll(struct stream_thread *sth, usec_t now_ut);
208
209
void stream_receiver_add_to_queue(struct receiver_state *rpt);
210
void stream_sender_add_to_connector_queue(struct rrdhost *host);
@@ -214,6 +226,12 @@ bool stream_thread_process_opcodes(struct stream_thread *sth, struct pollfd_meta
226
227
void stream_receiver_move_to_running_unsafe(struct stream_thread *sth, struct receiver_state *rpt);
228
229
+bool stream_sender_receive_data(struct stream_thread *sth, struct sender_state *s, usec_t now_ut, bool process_opcodes);
230
+bool stream_sender_send_data(struct stream_thread *sth, struct sender_state *s, usec_t now_ut, bool process_opcodes_and_enable_removal);
231
+
232
+bool stream_receiver_receive_data(struct stream_thread *sth, struct receiver_state *rpt, usec_t now_ut, bool process_opcodes);
233
+bool stream_receiver_send_data(struct stream_thread *sth, struct receiver_state *rpt, usec_t now_ut, bool process_opcodes_and_enable_removal);
234
+
235
#include "stream-sender-internals.h"
236
#include "stream-receiver-internals.h"
237
#include "plugins.d/pluginsd_parser.h"
src/web/api/formatters/json_wrapper.c
+4
-4
@@ -16,7 +16,7 @@ static void jsonwrap_query_metric_plan(BUFFER *wb, QUERY_METRIC *qm) {
16
buffer_json_array_close(wb);
17
18
buffer_json_member_add_array(wb, "tiers");
19
- for (size_t tier = 0; tier < storage_tiers; tier++) {
19
+ for (size_t tier = 0; tier < nd_profile.storage_tiers; tier++) {
20
buffer_json_add_array_item_object(wb);
21
buffer_json_member_add_uint64(wb, "tr", tier);
22
buffer_json_member_add_time_t(wb, "fe", qm->tiers[tier].db_first_time_s);
@@ -921,7 +921,7 @@ void rrdr_json_wrapper_begin(RRDR *r, BUFFER *wb) {
921
buffer_json_member_add_string(wb, "format", rrdr_format_to_string(format));
922
923
buffer_json_member_add_array(wb, "db_points_per_tier");
924
- for(size_t tier = 0; tier < storage_tiers ; tier++)
924
+ for(size_t tier = 0; tier < nd_profile.storage_tiers; tier++)
925
buffer_json_add_array_item_uint64(wb, qt->db.tiers[tier].points);
926
buffer_json_array_close(wb);
927
@@ -1500,7 +1500,7 @@ void rrdr_json_wrapper_end2(RRDR *r, BUFFER *wb) {
1500
1501
buffer_json_member_add_object(wb, "db");
1502
{
1503
- buffer_json_member_add_uint64(wb, "tiers", storage_tiers);
1503
+ buffer_json_member_add_uint64(wb, "tiers", nd_profile.storage_tiers);
1504
buffer_json_member_add_time_t(wb, "update_every", qt->db.minimum_latest_update_every_s);
1505
buffer_json_member_add_time_t(wb, "first_entry", qt->db.first_time_s);
1506
buffer_json_member_add_time_t(wb, "last_entry", qt->db.last_time_s);
@@ -1515,7 +1515,7 @@ void rrdr_json_wrapper_end2(RRDR *r, BUFFER *wb) {
1515
buffer_json_object_close(wb); // dimensions
1516
1517
buffer_json_member_add_array(wb, "per_tier");
1518
- for(size_t tier = 0; tier < storage_tiers ; tier++) {
1518
+ for(size_t tier = 0; tier < nd_profile.storage_tiers; tier++) {
1519
buffer_json_add_array_item_object(wb);
1520
buffer_json_member_add_uint64(wb, "tier", tier);
1521
buffer_json_member_add_uint64(wb, "queries", qt->db.tiers[tier].queries);
src/web/api/formatters/rrd2json.c
+1
-1
@@ -63,7 +63,7 @@ int rrdset2value_api_v1(
63
*db_points_read += r->stats.db_points_read;
64
65
if(db_points_per_tier) {
66
- for(size_t t = 0; t < storage_tiers ;t++)
66
+ for(size_t t = 0; t < nd_profile.storage_tiers;t++)
67
db_points_per_tier[t] += r->internal.qt->db.tiers[t].points;
68
}
69
src/web/api/formatters/value/value.c
+1
-1
@@ -133,7 +133,7 @@ QUERY_VALUE rrdmetric2value(RRDHOST *host,
133
storage_point_merge_to(qv.sp, r->internal.qt->query.array[d].query_points);
134
}
135
136
- for(size_t t = 0; t < storage_tiers ;t++)
136
+ for(size_t t = 0; t < nd_profile.storage_tiers;t++)
137
qv.storage_points_per_tier[t] = r->internal.qt->db.tiers[t].points;
138
139
long i = (!(options & RRDR_OPTION_REVERSED))?(long)rrdr_rows(r) - 1:0;
src/web/api/queries/backfill.c
+55
-22
@@ -3,11 +3,11 @@
3
#include "backfill.h"
4
5
struct backfill_request {
6
- size_t rrdhost_receiver_state_id;
6
+ OBJECT_STATE_ID host_state_id;
7
RRDSET_ACQUIRED *rsa;
8
- uint32_t works;
9
- uint32_t successful;
10
- uint32_t failed;
8
+ int32_t works;
9
+ int32_t successful;
10
+ int32_t failed;
11
backfill_callback_t cb;
12
struct backfill_request_data data;
13
};
@@ -28,6 +28,9 @@ static struct {
28
size_t queue_size;
29
BACKFILL_JudyLSet queue;
30
31
+ size_t charts_added;
32
+ size_t callbacks_executed;
33
+
34
ARAL *ar_br;
35
ARAL *ar_bdm;
36
@@ -46,9 +49,9 @@ bool backfill_request_add(RRDSET *st, backfill_callback_t cb, struct backfill_re
49
struct backfill_dim_work *array[dimensions];
50
51
if(backfill_globals.running) {
49
- struct backfill_request *br = aral_mallocz(backfill_globals.ar_br);
52
+ struct backfill_request *br = aral_callocz(backfill_globals.ar_br);
53
br->data = *data;
51
- br->rrdhost_receiver_state_id = rrdhost_state_id(st->rrdhost);
54
+ br->host_state_id = object_state_id(&st->rrdhost->state_id);
55
br->rsa = rrdset_find_and_acquire(st->rrdhost, string2str(st->id));
56
if(br->rsa) {
57
br->cb = cb;
@@ -59,7 +62,7 @@ bool backfill_request_add(RRDSET *st, backfill_callback_t cb, struct backfill_re
62
break;
63
64
if (!rrddim_option_check(rd, RRDDIM_OPTION_BACKFILLED_HIGH_TIERS)) {
62
- struct backfill_dim_work *bdm = aral_mallocz(backfill_globals.ar_bdm);
65
+ struct backfill_dim_work *bdm = aral_callocz(backfill_globals.ar_bdm);
66
bdm->rda = (RRDDIM_ACQUIRED *)dictionary_acquired_item_dup(st->rrddim_root_index, rd_dfe.item);
67
bdm->br = br;
68
br->works++;
@@ -69,9 +72,13 @@ bool backfill_request_add(RRDSET *st, backfill_callback_t cb, struct backfill_re
72
rrddim_foreach_done(rd);
73
}
74
75
+ internal_fatal((size_t)br->works != added, "works and added are not the same");
76
+
77
if(added) {
78
spinlock_lock(&backfill_globals.spinlock);
79
80
+ __atomic_add_fetch(&backfill_globals.charts_added, 1, __ATOMIC_RELAXED);
81
+
82
for(size_t i = 0; i < added ;i++) {
83
backfill_globals.queue_size++;
84
BACKFILL_SET(&backfill_globals.queue, backfill_globals.id++, array[i]);
@@ -95,20 +102,20 @@ bool backfill_request_add(RRDSET *st, backfill_callback_t cb, struct backfill_re
102
bool backfill_execute(struct backfill_dim_work *bdm) {
103
RRDSET *st = rrdset_acquired_to_rrdset(bdm->br->rsa);
104
98
- if(!rrdhost_state_acquire(st->rrdhost, bdm->br->rrdhost_receiver_state_id))
105
+ if(!object_state_acquire(&st->rrdhost->state_id, bdm->br->host_state_id))
106
return false;
107
108
size_t success = 0;
109
RRDDIM *rd = rrddim_acquired_to_rrddim(bdm->rda);
110
104
- for (size_t tier = 1; tier < storage_tiers; tier++)
111
+ for (size_t tier = 1; tier < nd_profile.storage_tiers; tier++)
112
if (backfill_tier_from_smaller_tiers(rd, tier, now_realtime_sec()))
113
success++;
114
115
if (success > 0)
116
rrddim_option_set(rd, RRDDIM_OPTION_BACKFILLED_HIGH_TIERS);
117
111
- rrdhost_state_release(st->rrdhost);
118
+ object_state_release(&st->rrdhost->state_id);
119
return success > 0;
120
}
121
@@ -120,12 +127,20 @@ static void backfill_dim_work_free(bool successful, struct backfill_dim_work *bd
127
else
128
__atomic_add_fetch(&br->failed, 1, __ATOMIC_RELAXED);
129
123
- uint32_t works = __atomic_sub_fetch(&br->works, 1, __ATOMIC_RELAXED);
124
- if(!works) {
125
- if(br->cb)
126
- br->cb(__atomic_load_n(&br->successful, __ATOMIC_RELAXED),
127
- __atomic_load_n(&br->failed, __ATOMIC_RELAXED),
128
- &br->data);
130
+ int32_t works = __atomic_sub_fetch(&br->works, 1, __ATOMIC_RELAXED);
131
+ internal_fatal(works < 0, "negative backfill jobs");
132
+
133
+ if(works == 0) {
134
+ // we are the last dimension of the chart
135
+
136
+ if(br->cb) {
137
+ __atomic_add_fetch(&backfill_globals.callbacks_executed, 1, __ATOMIC_RELAXED);
138
+
139
+ br->cb(
140
+ __atomic_load_n(&br->successful, __ATOMIC_RELAXED),
141
+ __atomic_load_n(&br->failed, __ATOMIC_RELAXED),
142
+ &br->data);
143
+ }
144
145
rrdset_acquired_release(br->rsa);
146
aral_freez(backfill_globals.ar_br, br);
@@ -135,7 +150,13 @@ static void backfill_dim_work_free(bool successful, struct backfill_dim_work *bd
150
aral_freez(backfill_globals.ar_bdm, bdm);
151
}
152
138
-void *backfill_worker_thread(void *ptr __maybe_unused) {
153
+#define LOG_WARNING_EVERY 10
154
+
155
+void *backfill_worker_thread(void *ptr) {
156
+ bool main_thread = (ptr == (void *)0x01);
157
+ size_t warning = LOG_WARNING_EVERY;
158
+ bool timeout = false;
159
+
160
worker_register("BACKFILL");
161
162
worker_register_job_name(0, "get");
@@ -156,16 +177,31 @@ void *backfill_worker_thread(void *ptr __maybe_unused) {
177
spinlock_unlock(&backfill_globals.spinlock);
178
179
if(bdm) {
180
+ warning = LOG_WARNING_EVERY;
181
worker_is_busy(1);
182
bool success = backfill_execute(bdm);
183
backfill_dim_work_free(success, bdm);
184
continue;
185
}
186
+ else if(main_thread && timeout) {
187
+ size_t added = __atomic_load_n(&backfill_globals.charts_added, __ATOMIC_RELAXED);
188
+ size_t executed = __atomic_load_n(&backfill_globals.callbacks_executed, __ATOMIC_RELAXED);
189
+
190
+ if(executed != added && --warning == 0) {
191
+ warning = LOG_WARNING_EVERY;
192
+
193
+ nd_log(NDLS_DAEMON, NDLP_WARNING,
194
+ "BACKFILL: the queue is empty, but the commands executed %zu is not equal to the commands added %zu",
195
+ executed, added);
196
+ }
197
+ }
198
199
worker_set_metric(2, (NETDATA_DOUBLE)queue_size);
200
201
worker_is_idle();
168
- job_id = completion_wait_for_a_job_with_timeout(&backfill_globals.completion, job_id, 1000);
202
+ size_t new_job_id = completion_wait_for_a_job_with_timeout(&backfill_globals.completion, job_id, 1000);
203
+ timeout = new_job_id == job_id;
204
+ job_id = new_job_id;
205
}
206
207
worker_unregister();
@@ -199,7 +235,7 @@ void *backfill_thread(void *ptr) {
235
th[t] = nd_thread_create(tag, NETDATA_THREAD_OPTION_JOINABLE, backfill_worker_thread, NULL);
236
}
237
202
- backfill_worker_thread(NULL);
238
+ backfill_worker_thread((void *)0x01);
239
static_thread->enabled = NETDATA_MAIN_THREAD_EXITING;
240
241
for(size_t t = 0; t < threads - 1 ;t++) {
@@ -227,6 +263,3 @@ void *backfill_thread(void *ptr) {
263
return NULL;
264
}
265
230
-bool backfill_threads_detect_from_stream_conf(void) {
231
- return stream_conf_is_parent(false);
232
-}
src/web/api/queries/backfill.h
+1
-3
@@ -7,7 +7,7 @@
7
8
struct parser;
9
struct backfill_request_data {
10
- size_t rrdhost_receiver_state_id;
10
+ OBJECT_STATE_ID host_state_id;
11
struct parser *parser;
12
RRDHOST *host;
13
RRDSET *st;
@@ -21,6 +21,4 @@ typedef bool (*backfill_callback_t)(size_t successful_dims, size_t failed_dims,
21
void *backfill_thread(void *ptr);
22
bool backfill_request_add(RRDSET *st, backfill_callback_t cb, struct backfill_request_data *data);
23
24
-bool backfill_threads_detect_from_stream_conf(void);
25
-
24
#endif //NETDATA_BACKFILL_H
src/web/api/queries/query.c
+17
-18
@@ -957,7 +957,7 @@ static bool query_metric_is_valid_tier(QUERY_METRIC *qm, size_t tier) {
957
}
958
959
static size_t query_metric_first_working_tier(QUERY_METRIC *qm) {
960
- for(size_t tier = 0; tier < storage_tiers ; tier++) {
960
+ for(size_t tier = 0; tier < nd_profile.storage_tiers; tier++) {
961
962
// find the db time-range for this tier for all metrics
963
STORAGE_METRIC_HANDLE *smh = qm->tiers[tier].smh;
@@ -1002,7 +1002,7 @@ static long query_plan_points_coverage_weight(time_t db_first_time_s, time_t db_
1002
}
1003
1004
static size_t query_metric_best_tier_for_timeframe(QUERY_METRIC *qm, time_t after_wanted, time_t before_wanted, size_t points_wanted) {
1005
- if(unlikely(storage_tiers < 2))
1005
+ if(unlikely(nd_profile.storage_tiers < 2))
1006
return 0;
1007
1008
if(unlikely(after_wanted == before_wanted || points_wanted <= 0))
@@ -1015,7 +1015,7 @@ static size_t query_metric_best_tier_for_timeframe(QUERY_METRIC *qm, time_t afte
1015
time_t min_first_time_s = 0;
1016
time_t max_last_time_s = 0;
1017
1018
- for(size_t tier = 0; tier < storage_tiers ; tier++) {
1018
+ for(size_t tier = 0; tier < nd_profile.storage_tiers; tier++) {
1019
time_t first_time_s = qm->tiers[tier].db_first_time_s;
1020
time_t last_time_s = qm->tiers[tier].db_last_time_s;
1021
@@ -1026,7 +1026,7 @@ static size_t query_metric_best_tier_for_timeframe(QUERY_METRIC *qm, time_t afte
1026
max_last_time_s = last_time_s;
1027
}
1028
1029
- for(size_t tier = 0; tier < storage_tiers ; tier++) {
1029
+ for(size_t tier = 0; tier < nd_profile.storage_tiers; tier++) {
1030
1031
// find the db time-range for this tier for all metrics
1032
STORAGE_METRIC_HANDLE *smh = qm->tiers[tier].smh;
@@ -1053,7 +1053,7 @@ static size_t query_metric_best_tier_for_timeframe(QUERY_METRIC *qm, time_t afte
1053
}
1054
1055
size_t best_tier = 0;
1056
- for(size_t tier = 1; tier < storage_tiers ; tier++) {
1056
+ for(size_t tier = 1; tier < nd_profile.storage_tiers; tier++) {
1057
if(qm->tiers[tier].weight >= qm->tiers[best_tier].weight)
1058
best_tier = tier;
1059
}
@@ -1062,7 +1062,7 @@ static size_t query_metric_best_tier_for_timeframe(QUERY_METRIC *qm, time_t afte
1062
}
1063
1064
static size_t rrddim_find_best_tier_for_timeframe(QUERY_TARGET *qt, time_t after_wanted, time_t before_wanted, size_t points_wanted) {
1065
- if(unlikely(storage_tiers < 2))
1065
+ if(unlikely(nd_profile.storage_tiers < 2))
1066
return 0;
1067
1068
if(unlikely(after_wanted == before_wanted || points_wanted <= 0)) {
@@ -1070,9 +1070,9 @@ static size_t rrddim_find_best_tier_for_timeframe(QUERY_TARGET *qt, time_t after
1070
return 0;
1071
}
1072
1073
- long weight[storage_tiers];
1073
+ long weight[nd_profile.storage_tiers];
1074
1075
- for(size_t tier = 0; tier < storage_tiers ; tier++) {
1075
+ for(size_t tier = 0; tier < nd_profile.storage_tiers; tier++) {
1076
1077
time_t common_first_time_s = 0;
1078
time_t common_last_time_s = 0;
@@ -1109,7 +1109,7 @@ static size_t rrddim_find_best_tier_for_timeframe(QUERY_TARGET *qt, time_t after
1109
}
1110
1111
size_t best_tier = 0;
1112
- for(size_t tier = 1; tier < storage_tiers ; tier++) {
1112
+ for(size_t tier = 1; tier < nd_profile.storage_tiers; tier++) {
1113
if(weight[tier] >= weight[best_tier])
1114
best_tier = tier;
1115
}
@@ -1122,13 +1122,13 @@ static size_t rrddim_find_best_tier_for_timeframe(QUERY_TARGET *qt, time_t after
1122
1123
static time_t rrdset_find_natural_update_every_for_timeframe(QUERY_TARGET *qt, time_t after_wanted, time_t before_wanted, size_t points_wanted, RRDR_OPTIONS options, size_t tier) {
1124
size_t best_tier;
1125
- if((options & RRDR_OPTION_SELECTED_TIER) && tier < storage_tiers)
1125
+ if((options & RRDR_OPTION_SELECTED_TIER) && tier < nd_profile.storage_tiers)
1126
best_tier = tier;
1127
else
1128
best_tier = rrddim_find_best_tier_for_timeframe(qt, after_wanted, before_wanted, points_wanted);
1129
1130
// find the db minimum update every for this tier for all metrics
1131
- time_t common_update_every_s = default_rrd_update_every;
1131
+ time_t common_update_every_s = nd_profile.update_every;
1132
for(size_t i = 0, used = qt->query.used; i < used ; i++) {
1133
QUERY_METRIC *qm = query_metric(qt, i);
1134
@@ -1365,8 +1365,7 @@ static bool query_plan(QUERY_ENGINE_OPS *ops, time_t after_wanted, time_t before
1365
bool switch_tiers = true;
1366
1367
if((ops->r->internal.qt->window.options & RRDR_OPTION_SELECTED_TIER)
1368
- && ops->r->internal.qt->window.tier < storage_tiers
1369
- && query_metric_is_valid_tier(qm, ops->r->internal.qt->window.tier)) {
1368
+ && ops->r->internal.qt->window.tier < nd_profile.storage_tiers && query_metric_is_valid_tier(qm, ops->r->internal.qt->window.tier)) {
1369
selected_tier = ops->r->internal.qt->window.tier;
1370
switch_tiers = false;
1371
}
@@ -1396,7 +1395,7 @@ static bool query_plan(QUERY_ENGINE_OPS *ops, time_t after_wanted, time_t before
1395
// check if our selected tier can start the query
1396
if (selected_tier_first_time_s > after_wanted) {
1397
// we need some help from other tiers
1399
- for (size_t tr = (int)selected_tier + 1; tr < storage_tiers && qm->plan.used < QUERY_PLANS_MAX ; tr++) {
1398
+ for (size_t tr = (int)selected_tier + 1; tr < nd_profile.storage_tiers && qm->plan.used < QUERY_PLANS_MAX ; tr++) {
1399
if(!query_metric_is_valid_tier(qm, tr))
1400
continue;
1401
@@ -1955,7 +1954,7 @@ static void rrd2rrdr_query_execute(RRDR *r, size_t dim_id_in_rrdr, QUERY_ENGINE_
1954
1955
r->stats.result_points_generated += points_added;
1956
r->stats.db_points_read += ops->db_total_points_read;
1958
- for(size_t tr = 0; tr < storage_tiers ; tr++)
1957
+ for(size_t tr = 0; tr < nd_profile.storage_tiers; tr++)
1958
qt->db.tiers[tr].points += ops->db_points_read_per_tier[tr];
1959
}
1960
@@ -1965,7 +1964,7 @@ static void rrd2rrdr_query_execute(RRDR *r, size_t dim_id_in_rrdr, QUERY_ENGINE_
1964
void store_metric_at_tier(RRDDIM *rd, size_t tier, struct rrddim_tier *t, STORAGE_POINT sp, usec_t now_ut);
1965
1966
bool backfill_tier_from_smaller_tiers(RRDDIM *rd, size_t tier, time_t now_s) {
1968
- if(unlikely(tier >= storage_tiers)) return false;
1967
+ if(unlikely(tier >= nd_profile.storage_tiers)) return false;
1968
#ifdef ENABLE_DBENGINE
1969
if(default_backfill == RRD_BACKFILL_NONE) return false;
1970
#else
@@ -2232,7 +2231,7 @@ bool query_target_calculate_window(QUERY_TARGET *qt) {
2231
rrdr_relative_window_to_absolute_query(&after_wanted, &before_wanted, NULL, unittest_running);
2232
query_debug_log(":relative2absolute after %ld, before %ld", after_wanted, before_wanted);
2233
2235
- if (natural_points && (options & RRDR_OPTION_SELECTED_TIER) && tier > 0 && storage_tiers > 1) {
2234
+ if (natural_points && (options & RRDR_OPTION_SELECTED_TIER) && tier > 0 && nd_profile.storage_tiers > 1) {
2235
update_every = rrdset_find_natural_update_every_for_timeframe(
2236
qt, after_wanted, before_wanted, points_wanted, options, tier);
2237
@@ -3498,7 +3497,7 @@ RRDR *rrd2rrdr(ONEWAYALLOC *owa, QUERY_TARGET *qt) {
3497
if(qt->query.used)
3498
ops = onewayalloc_callocz(owa, qt->query.used, sizeof(QUERY_ENGINE_OPS *));
3499
3501
- size_t capacity = libuv_worker_threads * 10;
3500
+ size_t capacity = MAX(netdata_conf_cpus() / 2, 4);
3501
size_t max_queries_to_prepare = (qt->query.used > (capacity - 1)) ? (capacity - 1) : qt->query.used;
3502
size_t queries_prepared = 0;
3503
while(queries_prepared < max_queries_to_prepare) {
src/web/api/queries/weights.c
+5
-5
@@ -147,7 +147,7 @@ static void results_header_to_json(DICTIONARY *results __maybe_unused, BUFFER *w
147
148
buffer_json_member_add_array(wb, "db_points_per_tier");
149
{
150
- for (size_t tier = 0; tier < storage_tiers; tier++)
150
+ for (size_t tier = 0; tier < nd_profile.storage_tiers; tier++)
151
buffer_json_add_array_item_uint64(wb, stats->db_points_per_tier[tier]);
152
}
153
buffer_json_array_close(wb);
@@ -459,7 +459,7 @@ static void results_header_to_json_v2(DICTIONARY *results __maybe_unused, BUFFER
459
460
buffer_json_member_add_array(wb, "db_points_per_tier");
461
{
462
- for (size_t tier = 0; tier < storage_tiers; tier++)
462
+ for (size_t tier = 0; tier < nd_profile.storage_tiers; tier++)
463
buffer_json_add_array_item_uint64(wb, stats->db_points_per_tier[tier]);
464
}
465
buffer_json_array_close(wb);
@@ -1295,7 +1295,7 @@ NETDATA_DOUBLE *rrd2rrdr_ks2(
1295
stats->db_queries++;
1296
stats->result_points += r->stats.result_points_generated;
1297
stats->db_points += r->stats.db_points_read;
1298
- for(size_t tr = 0; tr < storage_tiers ; tr++)
1298
+ for(size_t tr = 0; tr < nd_profile.storage_tiers; tr++)
1299
stats->db_points_per_tier[tr] += r->internal.qt->db.tiers[tr].points;
1300
1301
if(r->d != 1 || r->internal.qt->query.used != 1) {
@@ -1402,7 +1402,7 @@ static void merge_query_value_to_stats(QUERY_VALUE *qv, WEIGHTS_STATS *stats, si
1402
stats->db_queries += queries;
1403
stats->result_points += qv->result_points;
1404
stats->db_points += qv->points_read;
1405
- for(size_t tier = 0; tier < storage_tiers ; tier++)
1405
+ for(size_t tier = 0; tier < nd_profile.storage_tiers; tier++)
1406
stats->db_points_per_tier[tier] += qv->storage_points_per_tier[tier];
1407
}
1408
@@ -1522,7 +1522,7 @@ static void rrdset_weights_multi_dimensional_value(struct query_weights_data *qw
1522
.tier = qwd->qwr->tier,
1523
.timeout_ms = qwd->qwr->timeout_ms,
1524
.query_source = QUERY_SOURCE_API_WEIGHTS,
1525
- .priority = STORAGE_PRIORITY_NORMAL,
1525
+ .priority = STORAGE_PRIORITY_SYNCHRONOUS_FIRST,
1526
};
1527
1528
ONEWAYALLOC *owa = onewayalloc_create(16 * 1024);
src/web/api/v1/api_v1_badge/web_buffer_svg.c
+1
-1
@@ -1121,7 +1121,7 @@ int api_v1_badge(RRDHOST *host, struct web_client *w, char *url) {
1121
NULL, &latest_timestamp,
1122
NULL, NULL, NULL,
1123
&value_is_null, NULL, 0, 0,
1124
- QUERY_SOURCE_API_BADGE, STORAGE_PRIORITY_NORMAL);
1124
+ QUERY_SOURCE_API_BADGE, STORAGE_PRIORITY_SYNCHRONOUS_FIRST);
1125
1126
// if the value cannot be calculated, show empty badge
1127
if (ret != HTTP_RESP_OK) {
src/web/api/v1/api_v1_data.c
+1
-1
@@ -111,7 +111,7 @@ int api_v1_data(RRDHOST *host, struct web_client *w, char *url) {
111
}
112
else if(!strcmp(name, "tier")) {
113
tier = str2ul(value);
114
- if(tier < storage_tiers)
114
+ if(tier < nd_profile.storage_tiers)
115
options |= RRDR_OPTION_SELECTED_TIER;
116
else
117
tier = 0;
src/web/api/v1/api_v1_dbengine.c
+1
-1
@@ -85,7 +85,7 @@ int api_v1_dbengine_stats(RRDHOST *host __maybe_unused, struct web_client *w, ch
85
wb->content_type = CT_APPLICATION_JSON;
86
buffer_no_cacheable(wb);
87
buffer_strcat(wb, "{");
88
- for(size_t tier = 0; tier < storage_tiers ;tier++) {
88
+ for(size_t tier = 0; tier < nd_profile.storage_tiers;tier++) {
89
buffer_sprintf(wb, "%s\n\t\"tier%zu\": {", tier?",":"", tier);
90
web_client_api_v1_dbengine_stats_for_tier(wb, tier);
91
buffer_strcat(wb, "\n\t}");
src/web/api/v2/api_v2_data.c
+1
-1
@@ -180,7 +180,7 @@ int api_v2_data(RRDHOST *host __maybe_unused, struct web_client *w, char *url) {
180
181
if(tier_str && *tier_str) {
182
tier = str2ul(tier_str);
183
- if(tier < storage_tiers)
183
+ if(tier < nd_profile.storage_tiers)
184
options |= RRDR_OPTION_SELECTED_TIER;
185
else
186
tier = 0;
src/web/api/v2/api_v2_weights.c
+1
-1
@@ -85,7 +85,7 @@ int web_client_api_request_weights(RRDHOST *host, struct web_client *w, char *ur
85
86
else if(!strcmp(name, "tier")) {
87
tier = str2ul(value);
88
- if(tier < storage_tiers)
88
+ if(tier < nd_profile.storage_tiers)
89
options |= RRDR_OPTION_SELECTED_TIER;
90
else
91
tier = 0;
src/web/api/web_api.c
+1
-1
@@ -125,7 +125,7 @@ bool web_client_interrupt_callback(void *data) {
125
if(w->interrupt.callback)
126
ret = w->interrupt.callback(w, w->interrupt.callback_data);
127
else
128
- ret = is_socket_closed(w->ofd);
128
+ ret = is_socket_closed(w->fd);
129
130
return ret;
131
}
src/web/server/static/static-threaded.c
+35
-180
@@ -29,7 +29,7 @@ static struct web_client *web_client_create_on_fd(POLLINFO *pi) {
29
struct web_client *w;
30
31
w = web_client_get_from_cache();
32
- w->ifd = w->ofd = pi->fd;
32
+ w->fd = pi->fd;
33
34
strncpyz(w->client_ip, pi->client_ip, sizeof(w->client_ip) - 1);
35
strncpyz(w->client_port, pi->client_port, sizeof(w->client_port) - 1);
@@ -41,19 +41,18 @@ static struct web_client *web_client_create_on_fd(POLLINFO *pi) {
41
42
int flag = 1;
43
if(unlikely(
44
- web_client_check_conn_tcp(w) && setsockopt(w->ifd, IPPROTO_TCP, TCP_NODELAY, (char *) &flag, sizeof(int)) != 0))
45
- netdata_log_debug(D_WEB_CLIENT, "%llu: failed to enable TCP_NODELAY on socket fd %d.", w->id, w->ifd);
44
+ web_client_check_conn_tcp(w) && setsockopt(w->fd, IPPROTO_TCP, TCP_NODELAY, (char *) &flag, sizeof(int)) != 0))
45
+ netdata_log_debug(D_WEB_CLIENT, "%llu: failed to enable TCP_NODELAY on socket fd %d.", w->id, w->fd);
46
47
flag = 1;
48
- if(unlikely(setsockopt(w->ifd, SOL_SOCKET, SO_KEEPALIVE, (char *) &flag, sizeof(int)) != 0))
49
- netdata_log_debug(D_WEB_CLIENT, "%llu: failed to enable SO_KEEPALIVE on socket fd %d.", w->id, w->ifd);
48
+ if(unlikely(setsockopt(w->fd, SOL_SOCKET, SO_KEEPALIVE, (char *) &flag, sizeof(int)) != 0))
49
+ netdata_log_debug(D_WEB_CLIENT, "%llu: failed to enable SO_KEEPALIVE on socket fd %d.", w->id, w->fd);
50
51
web_client_update_acl_matches(w);
52
web_client_enable_wait_receive(w);
53
54
web_server_log_connection(w, "CONNECTED");
55
56
- w->pollinfo_slot = pi->slot;
56
return(w);
57
}
58
@@ -73,9 +72,6 @@ struct web_server_static_threaded_worker {
72
volatile size_t receptions;
73
volatile size_t sends;
74
volatile size_t max_concurrent;
76
-
77
- volatile size_t files_read;
78
- volatile size_t file_reads;
75
};
76
77
static long long static_threaded_workers_count = 1;
@@ -92,107 +88,10 @@ static inline int web_server_check_client_status(struct web_client *w) {
88
return 0;
89
}
90
95
-// ----------------------------------------------------------------------------
96
-// web server files
97
-
98
-static void *web_server_file_add_callback(POLLINFO *pi, short int *events, void *data) {
99
- struct web_client *w = (struct web_client *)data;
100
-
101
- worker_is_busy(WORKER_JOB_ADD_FILE);
102
-
103
- worker_private->files_read++;
104
-
105
- netdata_log_debug(D_WEB_CLIENT, "%llu: ADDED FILE READ ON FD %d", w->id, pi->fd);
106
- *events = POLLIN;
107
- pi->data = w;
108
-
109
- worker_is_idle();
110
- return w;
111
-}
112
-
113
-static void web_server_file_del_callback(POLLINFO *pi) {
114
- struct web_client *w = (struct web_client *)pi->data;
115
- netdata_log_debug(D_WEB_CLIENT, "%llu: RELEASE FILE READ ON FD %d", w->id, pi->fd);
116
-
117
- worker_is_busy(WORKER_JOB_DEL_FILE);
118
-
119
- w->pollinfo_filecopy_slot = 0;
120
-
121
- if(unlikely(!w->pollinfo_slot)) {
122
- netdata_log_debug(D_WEB_CLIENT, "%llu: CROSS WEB CLIENT CLEANUP (iFD %d, oFD %d)", w->id, pi->fd, w->ofd);
123
- web_server_log_connection(w, "DISCONNECTED");
124
- web_client_request_done(w);
125
- web_client_release_to_cache(w);
126
- pulse_web_client_disconnected();
127
- }
128
-
129
- worker_is_idle();
130
-}
131
-
132
-static int web_server_file_read_callback(POLLINFO *pi, short int *events) {
133
- int retval = -1;
134
- struct web_client *w = (struct web_client *)pi->data;
135
-
136
- worker_is_busy(WORKER_JOB_READ_FILE);
137
-
138
- // if there is no POLLINFO linked to this, it means the client disconnected
139
- // stop the file reading too
140
- if(unlikely(!w->pollinfo_slot)) {
141
- netdata_log_debug(D_WEB_CLIENT, "%llu: PREVENTED ATTEMPT TO READ FILE ON FD %d, ON CLOSED WEB CLIENT", w->id, pi->fd);
142
- retval = -1;
143
- goto cleanup;
144
- }
145
-
146
- if(unlikely(w->mode != HTTP_REQUEST_MODE_FILECOPY || w->ifd == w->ofd)) {
147
- netdata_log_debug(D_WEB_CLIENT, "%llu: PREVENTED ATTEMPT TO READ FILE ON FD %d, ON NON-FILECOPY WEB CLIENT", w->id, pi->fd);
148
- retval = -1;
149
- goto cleanup;
150
- }
151
-
152
- netdata_log_debug(D_WEB_CLIENT, "%llu: READING FILE ON FD %d", w->id, pi->fd);
153
-
154
- worker_private->file_reads++;
155
- ssize_t ret = unlikely(web_client_read_file(w));
156
-
157
- if(likely(web_client_has_wait_send(w))) {
158
- POLLJOB *p = pi->p; // our POLLJOB
159
- POLLINFO *wpi = pollinfo_from_slot(p, w->pollinfo_slot); // POLLINFO of the client socket
160
-
161
- netdata_log_debug(D_WEB_CLIENT, "%llu: SIGNALING W TO SEND (iFD %d, oFD %d)", w->id, pi->fd, wpi->fd);
162
- p->fds[wpi->slot].events |= POLLOUT;
163
- }
164
-
165
- if(unlikely(ret <= 0 || w->ifd == w->ofd)) {
166
- netdata_log_debug(D_WEB_CLIENT, "%llu: DONE READING FILE ON FD %d", w->id, pi->fd);
167
- retval = -1;
168
- goto cleanup;
169
- }
170
-
171
- *events = POLLIN;
172
- retval = 0;
173
-
174
-cleanup:
175
- worker_is_idle();
176
- return retval;
177
-}
178
-
179
-static int web_server_file_write_callback(POLLINFO *pi, short int *events) {
180
- (void)pi;
181
- (void)events;
182
-
183
- worker_is_busy(WORKER_JOB_WRITE_FILE);
184
- netdata_log_error("Writing to web files is not supported!");
185
- worker_is_idle();
186
-
187
- return -1;
188
-}
189
-
91
// ----------------------------------------------------------------------------
92
// web server clients
93
193
-static void *web_server_add_callback(POLLINFO *pi, short int *events, void *data) {
194
- (void)data; // Suppress warning on unused argument
195
-
94
+static void *web_server_add_callback(POLLINFO *pi, nd_poll_event_t *events, void *data __maybe_unused) {
95
worker_is_busy(WORKER_JOB_ADD_CONNECTION);
96
worker_private->connected++;
97
@@ -200,7 +99,7 @@ static void *web_server_add_callback(POLLINFO *pi, short int *events, void *data
99
if(unlikely(concurrent > worker_private->max_concurrent))
100
worker_private->max_concurrent = concurrent;
101
203
- *events = POLLIN;
102
+ *events = ND_POLL_READ;
103
104
netdata_log_debug(D_WEB_CLIENT_ACCESS, "LISTENER on %d: new connection.", pi->fd);
105
struct web_client *w = web_client_create_on_fd(pi);
@@ -212,17 +111,17 @@ static void *web_server_add_callback(POLLINFO *pi, short int *events, void *data
111
}
112
113
if ((web_client_check_conn_tcp(w)) && (netdata_ssl_web_server_ctx)) {
215
- sock_delnonblock(w->ifd);
114
+ sock_setnonblock(w->fd, false);
115
116
//Read the first 7 bytes from the message, but the message
117
//is not removed from the queue, because we are using MSG_PEEK
118
char test[8];
220
- if ( recv(w->ifd,test, 7, MSG_PEEK) == 7 ) {
119
+ if ( recv(w->fd,test, 7, MSG_PEEK) == 7 ) {
120
test[7] = '\0';
121
}
122
else {
123
// we couldn't read 7 bytes
225
- sock_setnonblock(w->ifd);
124
+ sock_setnonblock(w->fd, true);
125
goto cleanup;
126
}
127
@@ -232,11 +131,11 @@ static void *web_server_add_callback(POLLINFO *pi, short int *events, void *data
131
}
132
else {
133
// SSL
235
- if(!netdata_ssl_open(&w->ssl, netdata_ssl_web_server_ctx, w->ifd) || !netdata_ssl_accept(&w->ssl))
134
+ if(!netdata_ssl_open(&w->ssl, netdata_ssl_web_server_ctx, w->fd) || !netdata_ssl_accept(&w->ssl))
135
WEB_CLIENT_IS_DEAD(w);
136
}
137
239
- sock_setnonblock(w->ifd);
138
+ sock_setnonblock(w->fd, true);
139
}
140
141
netdata_log_debug(D_WEB_CLIENT, "%llu: ADDED CLIENT FD %d", w->id, pi->fd);
@@ -254,28 +153,19 @@ static void web_server_del_callback(POLLINFO *pi) {
153
154
struct web_client *w = (struct web_client *)pi->data;
155
257
- w->pollinfo_slot = 0;
258
- if(unlikely(w->pollinfo_filecopy_slot)) {
259
- POLLINFO *fpi = pollinfo_from_slot(pi->p, w->pollinfo_filecopy_slot); // POLLINFO of the client socket
260
- (void)fpi;
156
+ if(web_client_flag_check(w, WEB_CLIENT_FLAG_DONT_CLOSE_SOCKET))
157
+ pi->flags |= POLLINFO_FLAG_DONT_CLOSE;
158
262
- netdata_log_debug(D_WEB_CLIENT, "%llu: THE CLIENT WILL BE FRED BY READING FILE JOB ON FD %d", w->id, fpi->fd);
263
- }
264
- else {
265
- if(web_client_flag_check(w, WEB_CLIENT_FLAG_DONT_CLOSE_SOCKET))
266
- pi->flags |= POLLINFO_FLAG_DONT_CLOSE;
267
-
268
- netdata_log_debug(D_WEB_CLIENT, "%llu: CLOSING CLIENT FD %d", w->id, pi->fd);
269
- web_server_log_connection(w, "DISCONNECTED");
270
- web_client_request_done(w);
271
- web_client_release_to_cache(w);
272
- pulse_web_client_disconnected();
273
- }
159
+ netdata_log_debug(D_WEB_CLIENT, "%llu: CLOSING CLIENT FD %d", w->id, pi->fd);
160
+ web_server_log_connection(w, "DISCONNECTED");
161
+ web_client_request_done(w);
162
+ web_client_release_to_cache(w);
163
+ pulse_web_client_disconnected();
164
165
worker_is_idle();
166
}
167
278
-static int web_server_rcv_callback(POLLINFO *pi, short int *events) {
168
+static int web_server_rcv_callback(POLLINFO *pi, nd_poll_event_t *events) {
169
int ret = -1;
170
worker_is_busy(WORKER_JOB_RCV_DATA);
171
@@ -296,57 +186,21 @@ static int web_server_rcv_callback(POLLINFO *pi, short int *events) {
186
if (unlikely(w->mode == HTTP_REQUEST_MODE_STREAM)) {
187
web_client_send(w);
188
}
189
+ else if(unlikely(w->fd == fd && web_client_has_wait_receive(w)))
190
+ *events |= ND_POLL_READ;
191
300
- else if(unlikely(w->mode == HTTP_REQUEST_MODE_FILECOPY)) {
301
- if(w->pollinfo_filecopy_slot == 0) {
302
- netdata_log_debug(D_WEB_CLIENT, "%llu: FILECOPY DETECTED ON FD %d", w->id, pi->fd);
303
-
304
- if (unlikely(w->ifd != -1 && w->ifd != w->ofd && w->ifd != fd)) {
305
- // add a new socket to poll_events, with the same
306
- netdata_log_debug(D_WEB_CLIENT, "%llu: CREATING FILECOPY SLOT ON FD %d", w->id, pi->fd);
307
-
308
- POLLINFO *fpi = poll_add_fd(
309
- pi->p
310
- , w->ifd
311
- , pi->port_acl
312
- , 0
313
- , POLLINFO_FLAG_CLIENT_SOCKET
314
- , "FILENAME"
315
- , ""
316
- , ""
317
- , web_server_file_add_callback
318
- , web_server_file_del_callback
319
- , web_server_file_read_callback
320
- , web_server_file_write_callback
321
- , (void *) w
322
- );
323
-
324
- if(fpi)
325
- w->pollinfo_filecopy_slot = fpi->slot;
326
- else {
327
- netdata_log_error("Failed to add filecopy fd. Closing client.");
328
- ret = -1;
329
- goto cleanup;
330
- }
331
- }
332
- }
333
- }
334
- else {
335
- if(unlikely(w->ifd == fd && web_client_has_wait_receive(w)))
336
- *events |= POLLIN;
337
- }
192
+ if(unlikely(w->fd == fd && web_client_has_wait_send(w)))
193
+ *events |= ND_POLL_WRITE;
194
339
- if(unlikely(w->ofd == fd && web_client_has_wait_send(w)))
340
- *events |= POLLOUT;
195
} else if(unlikely(bytes < 0)) {
196
ret = -1;
197
goto cleanup;
198
} else if (unlikely(bytes == 0)) {
345
- if(unlikely(w->ifd == fd && web_client_has_ssl_wait_receive(w)))
346
- *events |= POLLIN;
199
+ if(unlikely(w->fd == fd && web_client_has_ssl_wait_receive(w)))
200
+ *events |= ND_POLL_READ;
201
348
- if(unlikely(w->ofd == fd && web_client_has_ssl_wait_send(w)))
349
- *events |= POLLOUT;
202
+ if(unlikely(w->fd == fd && web_client_has_ssl_wait_send(w)))
203
+ *events |= ND_POLL_WRITE;
204
}
205
206
ret = web_server_check_client_status(w);
@@ -356,7 +210,7 @@ cleanup:
210
return ret;
211
}
212
359
-static int web_server_snd_callback(POLLINFO *pi, short int *events) {
213
+static int web_server_snd_callback(POLLINFO *pi, nd_poll_event_t *events) {
214
int retval = -1;
215
worker_is_busy(WORKER_JOB_SND_DATA);
216
@@ -374,11 +228,11 @@ static int web_server_snd_callback(POLLINFO *pi, short int *events) {
228
goto cleanup;
229
}
230
377
- if(unlikely(w->ifd == fd && web_client_has_wait_receive(w)))
378
- *events |= POLLIN;
231
+ if(unlikely(w->fd == fd && web_client_has_wait_receive(w)))
232
+ *events |= ND_POLL_READ;
233
380
- if(unlikely(w->ofd == fd && web_client_has_wait_send(w)))
381
- *events |= POLLOUT;
234
+ if(unlikely(w->fd == fd && web_client_has_wait_send(w)))
235
+ *events |= ND_POLL_WRITE;
236
237
retval = web_server_check_client_status(w);
238
@@ -437,7 +291,8 @@ void *socket_listen_main_static_threaded_worker(void *ptr) {
291
, NULL
292
, web_client_first_request_timeout
293
, web_client_timeout
440
- , default_rrd_update_every * 1000 // timer_milliseconds
294
+ ,
295
+ nd_profile.update_every * 1000 // timer_milliseconds
296
, ptr // timer_data
297
, worker_private->max_sockets
298
);
src/web/server/web_client.c
+44
-170
@@ -82,22 +82,6 @@ static inline int bad_request_multiple_dashboard_versions(struct web_client *w)
82
return HTTP_RESP_BAD_REQUEST;
83
}
84
85
-static inline int web_client_cork_socket(struct web_client *w __maybe_unused) {
86
-#ifdef TCP_CORK
87
- if(likely(web_client_check_conn_tcp(w) && !w->tcp_cork && w->ofd != -1)) {
88
- w->tcp_cork = true;
89
- if(unlikely(setsockopt(w->ofd, IPPROTO_TCP, TCP_CORK, (char *) &w->tcp_cork, sizeof(int)) != 0)) {
90
- netdata_log_error("%llu: failed to enable TCP_CORK on socket.", w->id);
91
-
92
- w->tcp_cork = false;
93
- return -1;
94
- }
95
- }
96
-#endif /* TCP_CORK */
97
-
98
- return 0;
99
-}
100
-
85
static inline void web_client_enable_wait_from_ssl(struct web_client *w) {
86
if (w->ssl.ssl_errno == SSL_ERROR_WANT_READ)
87
web_client_enable_ssl_wait_receive(w);
@@ -109,22 +93,6 @@ static inline void web_client_enable_wait_from_ssl(struct web_client *w) {
93
}
94
}
95
112
-static inline int web_client_uncork_socket(struct web_client *w __maybe_unused) {
113
-#ifdef TCP_CORK
114
- if(likely(w->tcp_cork && w->ofd != -1)) {
115
- w->tcp_cork = false;
116
- if(unlikely(setsockopt(w->ofd, IPPROTO_TCP, TCP_CORK, (char *) &w->tcp_cork, sizeof(int)) != 0)) {
117
- netdata_log_error("%llu: failed to disable TCP_CORK on socket.", w->id);
118
- w->tcp_cork = true;
119
- return -1;
120
- }
121
- }
122
-#endif /* TCP_CORK */
123
-
124
- w->tcp_cork = false;
125
- return 0;
126
-}
127
-
96
static inline char *strip_control_characters(char *url) {
97
if(!url) return "";
98
@@ -221,7 +189,7 @@ void web_client_log_completed_request(struct web_client *w, bool update_web_stat
189
struct timeval tv;
190
now_monotonic_high_precision_timeval(&tv);
191
224
- size_t size = (w->mode == HTTP_REQUEST_MODE_FILECOPY) ? w->response.rlen : w->response.data->len;
192
+ size_t size = w->response.data->len;
193
size_t sent = w->response.zoutput ? (size_t)w->response.zstream.total_out : size;
194
195
if(update_web_stats)
@@ -276,26 +244,11 @@ void web_client_log_completed_request(struct web_client *w, bool update_web_stat
244
}
245
246
void web_client_request_done(struct web_client *w) {
279
- web_client_uncork_socket(w);
247
+ sock_setcork(w->fd, false);
248
249
netdata_log_debug(D_WEB_CLIENT, "%llu: Resetting client.", w->id);
250
251
web_client_log_completed_request(w, true);
284
-
285
- if(unlikely(w->mode == HTTP_REQUEST_MODE_FILECOPY)) {
286
- if(w->ifd != w->ofd) {
287
- netdata_log_debug(D_WEB_CLIENT, "%llu: Closing filecopy input file descriptor %d.", w->id, w->ifd);
288
-
289
- if(web_server_mode != WEB_SERVER_MODE_STATIC_THREADED) {
290
- if (w->ifd != -1){
291
- close(w->ifd);
292
- }
293
- }
294
-
295
- w->ifd = w->ofd;
296
- }
297
- }
298
-
252
web_client_reset_allocations(w, false);
253
254
w->mode = HTTP_REQUEST_MODE_GET;
@@ -311,7 +264,6 @@ void web_client_request_done(struct web_client *w) {
264
web_client_disable_wait_send(w);
265
266
w->response.has_cookies = false;
314
- w->response.rlen = 0;
267
w->response.sent = 0;
268
w->response.code = 0;
269
w->response.zoutput = false;
@@ -467,7 +419,7 @@ static bool find_filename_to_serve(const char *filename, char *dst, size_t dst_l
419
return true;
420
}
421
470
-static int mysendfile(struct web_client *w, char *filename) {
422
+static int web_server_static_file(struct web_client *w, char *filename) {
423
netdata_log_debug(D_WEB_CLIENT, "%llu: Looking for file '%s/%s'", w->id, netdata_configured_web_dir, filename);
424
425
if(!http_can_access_dashboard(w))
@@ -511,10 +463,24 @@ static int mysendfile(struct web_client *w, char *filename) {
463
if(is_dir && !web_client_flag_check(w, WEB_CLIENT_FLAG_PATH_HAS_TRAILING_SLASH))
464
return append_slash_to_url_and_redirect(w);
465
466
+ buffer_flush(w->response.data);
467
+ buffer_need_bytes(w->response.data, (size_t)statbuf.st_size);
468
+ w->response.data->len = (size_t)statbuf.st_size;
469
+
470
// open the file
515
- w->ifd = open(web_filename, O_NONBLOCK, O_RDONLY | O_CLOEXEC);
516
- if(w->ifd == -1) {
517
- w->ifd = w->ofd;
471
+ int fd = open(web_filename, O_RDONLY | O_CLOEXEC);
472
+
473
+ // read the file
474
+ if(fd != -1 && read(fd, w->response.data->buffer, statbuf.st_size) != statbuf.st_size) {
475
+ // cannot read the whole file
476
+ nd_log(NDLS_DAEMON, NDLP_ERR, "Web server failed to read file '%s'", web_filename);
477
+ close(fd);
478
+ fd = -1;
479
+ }
480
+
481
+ // check for failures
482
+ if(fd == -1) {
483
+ buffer_flush(w->response.data);
484
485
if(errno == EBUSY || errno == EAGAIN) {
486
netdata_log_error("%llu: File '%s' is busy, sending 307 Moved Temporarily to force retry.", w->id, web_filename);
@@ -532,24 +498,23 @@ static int mysendfile(struct web_client *w, char *filename) {
498
return HTTP_RESP_NOT_FOUND;
499
}
500
}
535
-
536
- sock_setnonblock(w->ifd);
501
+ else
502
+ close(fd);
503
504
w->response.data->content_type = contenttype_for_filename(web_filename);
539
- netdata_log_debug(D_WEB_CLIENT_ACCESS, "%llu: Sending file '%s' (%"PRId64" bytes, ifd %d, ofd %d).", w->id, web_filename, (int64_t)statbuf.st_size, w->ifd, w->ofd);
505
+ netdata_log_debug(D_WEB_CLIENT_ACCESS, "%llu: Sending file '%s' (%"PRId64" bytes, fd %d).", w->id, web_filename, (int64_t)statbuf.st_size, w->fd);
506
+
507
+ w->mode = HTTP_REQUEST_MODE_GET;
508
+ web_client_enable_wait_send(w);
509
+ web_client_disable_wait_receive(w);
510
541
- w->mode = HTTP_REQUEST_MODE_FILECOPY;
542
- web_client_enable_wait_receive(w);
543
- web_client_disable_wait_send(w);
544
- buffer_flush(w->response.data);
545
- buffer_need_bytes(w->response.data, (size_t)statbuf.st_size);
546
- w->response.rlen = (size_t)statbuf.st_size;
511
#ifdef __APPLE__
512
w->response.data->date = statbuf.st_mtimespec.tv_sec;
513
#else
514
w->response.data->date = statbuf.st_mtim.tv_sec;
515
#endif
516
w->response.data->expires = now_realtime_sec() + 86400;
517
+
518
buffer_cacheable(w->response.data);
519
520
return HTTP_RESP_OK;
@@ -814,9 +779,9 @@ static inline ssize_t web_client_send_data(struct web_client *w,const void *buf,
779
bytes = netdata_ssl_write(&w->ssl, buf, len);
780
web_client_enable_wait_from_ssl(w);
781
} else
817
- bytes = send(w->ofd, buf, len, flags);
782
+ bytes = send(w->fd, buf, len, flags);
783
} else if (web_client_check_conn_tcp(w) || web_client_check_conn_unix(w))
819
- bytes = send(w->ofd, buf, len, flags);
784
+ bytes = send(w->fd, buf, len, flags);
785
else
786
bytes = -999;
787
@@ -923,9 +888,9 @@ void web_client_build_http_header(struct web_client *w) {
888
if(likely(w->flags & WEB_CLIENT_CHUNKED_TRANSFER))
889
buffer_strcat(w->response.header_output, "Transfer-Encoding: chunked\r\n");
890
else {
926
- if(likely((w->response.data->len || w->response.rlen))) {
891
+ if(likely(w->response.data->len)) {
892
// we know the content length, put it
928
- buffer_sprintf(w->response.header_output, "Content-Length: %zu\r\n", w->response.data->len? w->response.data->len: w->response.rlen);
893
+ buffer_sprintf(w->response.header_output, "Content-Length: %zu\r\n", (size_t)w->response.data->len);
894
}
895
else {
896
// we don't know the content length, disable keep-alive
@@ -952,7 +917,7 @@ static inline void web_client_send_http_header(struct web_client *w) {
917
, buffer_tostring(w->response.header_output)
918
);
919
955
- web_client_cork_socket(w);
920
+ sock_setcork(w->fd, true);
921
922
size_t count = 0;
923
ssize_t bytes;
@@ -963,7 +928,7 @@ static inline void web_client_send_http_header(struct web_client *w) {
928
web_client_enable_wait_from_ssl(w);
929
}
930
else {
966
- while((bytes = send(w->ofd, buffer_tostring(w->response.header_output), buffer_strlen(w->response.header_output), 0)) == -1) {
931
+ while((bytes = send(w->fd, buffer_tostring(w->response.header_output), buffer_strlen(w->response.header_output), 0)) == -1) {
932
count++;
933
934
if(count > 100 || (errno != EAGAIN && errno != EWOULDBLOCK)) {
@@ -974,7 +939,7 @@ static inline void web_client_send_http_header(struct web_client *w) {
939
}
940
}
941
else if(web_client_check_conn_tcp(w) || web_client_check_conn_unix(w)) {
977
- while((bytes = send(w->ofd, buffer_tostring(w->response.header_output), buffer_strlen(w->response.header_output), 0)) == -1) {
942
+ while((bytes = send(w->fd, buffer_tostring(w->response.header_output), buffer_strlen(w->response.header_output), 0)) == -1) {
943
count++;
944
945
if(count > 100 || (errno != EAGAIN && errno != EWOULDBLOCK)) {
@@ -1285,7 +1250,7 @@ static inline int web_client_process_url(RRDHOST *host, struct web_client *w, ch
1250
}
1251
1252
buffer_flush(w->response.data);
1288
- return mysendfile(w, filename);
1253
+ return web_server_static_file(w, filename);
1254
}
1255
1256
static bool web_server_log_transport(BUFFER *wb, void *ptr) {
@@ -1365,7 +1330,6 @@ void web_client_process_request_from_web_server(struct web_client *w) {
1330
w->response.code = HTTP_RESP_OK;
1331
break;
1332
1368
- case HTTP_REQUEST_MODE_FILECOPY:
1333
case HTTP_REQUEST_MODE_POST:
1334
case HTTP_REQUEST_MODE_GET:
1335
case HTTP_REQUEST_MODE_PUT:
@@ -1508,28 +1472,6 @@ void web_client_process_request_from_web_server(struct web_client *w) {
1472
w->id, (size_t)w->response.data->len);
1473
break;
1474
1511
- case HTTP_REQUEST_MODE_FILECOPY:
1512
- if(w->response.rlen) {
1513
- netdata_log_debug(D_WEB_CLIENT, "%llu: Done preparing the response. Will be sending data file of %zu bytes to client.", w->id, w->response.rlen);
1514
- web_client_enable_wait_receive(w);
1515
-
1516
- /*
1517
- // utilize the kernel sendfile() for copying the file to the socket.
1518
- // this block of code can be commented, without anything missing.
1519
- // when it is commented, the program will copy the data using async I/O.
1520
- {
1521
- long len = sendfile(w->ofd, w->ifd, NULL, w->response.data->rbytes);
1522
- if(len != w->response.data->rbytes)
1523
- netdata_log_error("%llu: sendfile() should copy %ld bytes, but copied %ld. Falling back to manual copy.", w->id, w->response.data->rbytes, len);
1524
- else
1525
- web_client_request_done(w);
1526
- }
1527
- */
1528
- }
1529
- else
1530
- netdata_log_debug(D_WEB_CLIENT, "%llu: Done preparing the response. Will be sending an unknown amount of bytes to client.", w->id);
1531
- break;
1532
-
1475
default:
1476
fatal("%llu: Unknown client mode %u.", w->id, w->mode);
1477
break;
@@ -1627,13 +1569,6 @@ ssize_t web_client_send_deflate(struct web_client *w)
1569
if(t < 0) return t;
1570
}
1571
1630
- if(w->mode == HTTP_REQUEST_MODE_FILECOPY && web_client_has_wait_receive(w) && w->response.rlen && w->response.rlen > w->response.data->len) {
1631
- // we have to wait, more data will come
1632
- netdata_log_debug(D_WEB_CLIENT, "%llu: Waiting for more data to become available.", w->id);
1633
- web_client_disable_wait_send(w);
1634
- return t;
1635
- }
1636
-
1572
if(unlikely(!web_client_has_keepalive(w))) {
1573
netdata_log_debug(D_WEB_CLIENT, "%llu: Closing (keep-alive is not enabled). %zu bytes sent.", w->id, w->response.sent);
1574
WEB_CLIENT_IS_DEAD(w);
@@ -1669,8 +1604,10 @@ ssize_t web_client_send_deflate(struct web_client *w)
1604
1605
// ask for FINISH if we have all the input
1606
int flush = Z_SYNC_FLUSH;
1672
- if((w->mode == HTTP_REQUEST_MODE_GET || w->mode == HTTP_REQUEST_MODE_POST || w->mode == HTTP_REQUEST_MODE_PUT || w->mode == HTTP_REQUEST_MODE_DELETE)
1673
- || (w->mode == HTTP_REQUEST_MODE_FILECOPY && !web_client_has_wait_receive(w) && w->response.data->len == w->response.rlen)) {
1607
+ if((w->mode == HTTP_REQUEST_MODE_GET ||
1608
+ w->mode == HTTP_REQUEST_MODE_POST ||
1609
+ w->mode == HTTP_REQUEST_MODE_PUT ||
1610
+ w->mode == HTTP_REQUEST_MODE_DELETE)) {
1611
flush = Z_FINISH;
1612
netdata_log_debug(D_DEFLATE, "%llu: Requesting Z_FINISH, if possible.", w->id);
1613
}
@@ -1735,13 +1672,6 @@ ssize_t web_client_send(struct web_client *w) {
1672
// A. we have done everything
1673
// B. we temporarily have nothing to send, waiting for the buffer to be filled by ifd
1674
1738
- if(w->mode == HTTP_REQUEST_MODE_FILECOPY && web_client_has_wait_receive(w) && w->response.rlen && w->response.rlen > w->response.data->len) {
1739
- // we have to wait, more data will come
1740
- netdata_log_debug(D_WEB_CLIENT, "%llu: Waiting for more data to become available.", w->id);
1741
- web_client_disable_wait_send(w);
1742
- return 0;
1743
- }
1744
-
1675
if(unlikely(!web_client_has_keepalive(w))) {
1676
netdata_log_debug(D_WEB_CLIENT, "%llu: Closing (keep-alive is not enabled). %zu bytes sent.", w->id, w->response.sent);
1677
WEB_CLIENT_IS_DEAD(w);
@@ -1770,63 +1700,7 @@ ssize_t web_client_send(struct web_client *w) {
1700
return(bytes);
1701
}
1702
1773
-ssize_t web_client_read_file(struct web_client *w)
1774
-{
1775
- if(unlikely(w->response.rlen > w->response.data->size))
1776
- buffer_need_bytes(w->response.data, w->response.rlen - w->response.data->size);
1777
-
1778
- if(unlikely(w->response.rlen <= w->response.data->len))
1779
- return 0;
1780
-
1781
- ssize_t left = (ssize_t)(w->response.rlen - w->response.data->len);
1782
- ssize_t bytes = read(w->ifd, &w->response.data->buffer[w->response.data->len], (size_t)left);
1783
- if(likely(bytes > 0)) {
1784
- size_t old = w->response.data->len;
1785
- (void)old;
1786
-
1787
- w->response.data->len += bytes;
1788
- w->response.data->buffer[w->response.data->len] = '\0';
1789
-
1790
- netdata_log_debug(D_WEB_CLIENT, "%llu: Read %zd bytes.", w->id, bytes);
1791
- netdata_log_debug(D_WEB_DATA, "%llu: Read data: '%s'.", w->id, &w->response.data->buffer[old]);
1792
-
1793
- web_client_enable_wait_send(w);
1794
-
1795
- if(w->response.rlen && w->response.data->len >= w->response.rlen)
1796
- web_client_disable_wait_receive(w);
1797
- }
1798
- else if(likely(bytes == 0)) {
1799
- netdata_log_debug(D_WEB_CLIENT, "%llu: Out of input file data.", w->id);
1800
-
1801
- // if we cannot read, it means we have an error on input.
1802
- // if however, we are copying a file from ifd to ofd, we should not return an error.
1803
- // in this case, the error should be generated when the file has been sent to the client.
1804
-
1805
- // we are copying data from ifd to ofd
1806
- // let it finish copying...
1807
- web_client_disable_wait_receive(w);
1808
-
1809
- netdata_log_debug(D_WEB_CLIENT, "%llu: Read the whole file.", w->id);
1810
-
1811
- if(web_server_mode != WEB_SERVER_MODE_STATIC_THREADED) {
1812
- if (w->ifd != w->ofd) close(w->ifd);
1813
- }
1814
-
1815
- w->ifd = w->ofd;
1816
- }
1817
- else {
1818
- netdata_log_debug(D_WEB_CLIENT, "%llu: read data failed.", w->id);
1819
- WEB_CLIENT_IS_DEAD(w);
1820
- }
1821
-
1822
- return(bytes);
1823
-}
1824
-
1825
-ssize_t web_client_receive(struct web_client *w)
1826
-{
1827
- if(unlikely(w->mode == HTTP_REQUEST_MODE_FILECOPY))
1828
- return web_client_read_file(w);
1829
-
1703
+ssize_t web_client_receive(struct web_client *w) {
1704
ssize_t bytes;
1705
1706
// do we have any space for more data?
@@ -1842,11 +1716,11 @@ ssize_t web_client_receive(struct web_client *w)
1716
web_client_enable_wait_from_ssl(w);
1717
}
1718
else {
1845
- bytes = recv(w->ifd, &w->response.data->buffer[w->response.data->len], (size_t) (left - 1), MSG_DONTWAIT);
1719
+ bytes = recv(w->fd, &w->response.data->buffer[w->response.data->len], (size_t) (left - 1), MSG_DONTWAIT);
1720
}
1721
}
1722
else if(web_client_check_conn_tcp(w) || web_client_check_conn_unix(w)) {
1849
- bytes = recv(w->ifd, &w->response.data->buffer[w->response.data->len], (size_t) (left - 1), MSG_DONTWAIT);
1723
+ bytes = recv(w->fd, &w->response.data->buffer[w->response.data->len], (size_t) (left - 1), MSG_DONTWAIT);
1724
}
1725
else // other connection methods
1726
bytes = -1;
@@ -1940,7 +1814,7 @@ void web_client_reuse_from_cache(struct web_client *w) {
1814
// zero everything
1815
memset(w, 0, sizeof(struct web_client));
1816
1943
- w->ifd = w->ofd = -1;
1817
+ w->fd = -1;
1818
w->statistics.memory_accounting = statistics_memory_accounting;
1819
w->use_count = use_count;
1820
src/web/server/web_client.h
+3
-15
@@ -140,15 +140,10 @@ struct response {
140
BUFFER *header; // our response header
141
BUFFER *header_output; // internal use
142
BUFFER *data; // our response data buffer
143
-
143
+ size_t sent; // current data length sent to output
144
short int code; // the HTTP response code
145
bool has_cookies;
146
-
147
- size_t rlen; // if non-zero, the excepted size of ifd (input of firecopy)
148
- size_t sent; // current data length sent to output
149
-
150
- bool zoutput; // if set to 1, web_client_send() will send compressed data
151
-
146
+ bool zoutput; // if set to 1, web_client_send() will send compressed data
147
bool zinitialized;
148
z_stream zstream; // zlib stream for sending compressed output to client
149
size_t zsent; // the compressed bytes we have sent to the client
@@ -174,9 +169,7 @@ struct web_client {
169
size_t header_parse_tries;
170
size_t header_parse_last_size;
171
177
- bool tcp_cork;
178
- int ifd;
179
- int ofd;
172
+ int fd;
173
174
char client_ip[INET6_ADDRSTRLEN]; // Defined buffer sizes include null-terminators
175
char client_port[NI_MAXSERV];
@@ -196,10 +189,6 @@ struct web_client {
189
190
BUFFER *payload; // when this request is a POST, this has the payload
191
199
- // STATIC-THREADED WEB SERVER MEMBERS
200
- size_t pollinfo_slot; // POLLINFO slot of the web client
201
- size_t pollinfo_filecopy_slot; // POLLINFO slot of the file read
202
-
192
NETDATA_SSL ssl;
193
194
struct {
@@ -241,7 +230,6 @@ int web_client_service_unavailable(struct web_client *w);
230
231
ssize_t web_client_send(struct web_client *w);
232
ssize_t web_client_receive(struct web_client *w);
244
-ssize_t web_client_read_file(struct web_client *w);
233
234
void web_client_process_request_from_web_server(struct web_client *w);
235
void web_client_request_done(struct web_client *w);
src/web/server/web_client_cache.c
-1
@@ -125,7 +125,6 @@ struct web_client *web_client_get_from_cache(void) {
125
}
126
127
void web_client_release_to_cache(struct web_client *w) {
128
-
128
netdata_ssl_close(&w->ssl);
129
130
// unlink it from the used
src/web/server/web_server.c
+6
-6
@@ -95,32 +95,32 @@ void web_client_update_acl_matches(struct web_client *w) {
95
96
if(!(w->port_acl & HTTP_ACL_TRANSPORTS_WITHOUT_CLIENT_IP_VALIDATION)) {
97
if (!web_allow_dashboard_from ||
98
- connection_allowed(w->ifd, w->client_ip, w->client_host, sizeof(w->client_host),
98
+ connection_allowed(w->fd, w->client_ip, w->client_host, sizeof(w->client_host),
99
web_allow_dashboard_from, "dashboard", web_allow_dashboard_dns))
100
w->acl |= HTTP_ACL_DASHBOARD;
101
102
if (!web_allow_registry_from ||
103
- connection_allowed(w->ifd, w->client_ip, w->client_host, sizeof(w->client_host),
103
+ connection_allowed(w->fd, w->client_ip, w->client_host, sizeof(w->client_host),
104
web_allow_registry_from, "registry", web_allow_registry_dns))
105
w->acl |= HTTP_ACL_REGISTRY;
106
107
if (!web_allow_badges_from ||
108
- connection_allowed(w->ifd, w->client_ip, w->client_host, sizeof(w->client_host),
108
+ connection_allowed(w->fd, w->client_ip, w->client_host, sizeof(w->client_host),
109
web_allow_badges_from, "badges", web_allow_badges_dns))
110
w->acl |= HTTP_ACL_BADGES;
111
112
if (!web_allow_mgmt_from ||
113
- connection_allowed(w->ifd, w->client_ip, w->client_host, sizeof(w->client_host),
113
+ connection_allowed(w->fd, w->client_ip, w->client_host, sizeof(w->client_host),
114
web_allow_mgmt_from, "management", web_allow_mgmt_dns))
115
w->acl |= HTTP_ACL_MANAGEMENT;
116
117
if (!web_allow_streaming_from ||
118
- connection_allowed(w->ifd, w->client_ip, w->client_host, sizeof(w->client_host),
118
+ connection_allowed(w->fd, w->client_ip, w->client_host, sizeof(w->client_host),
119
web_allow_streaming_from, "streaming", web_allow_streaming_dns))
120
w->acl |= HTTP_ACL_STREAMING;
121
122
if (!web_allow_netdataconf_from ||
123
- connection_allowed(w->ifd, w->client_ip, w->client_host, sizeof(w->client_host),
123
+ connection_allowed(w->fd, w->client_ip, w->client_host, sizeof(w->client_host),
124
web_allow_netdataconf_from, "netdata.conf", web_allow_netdataconf_dns))
125
w->acl |= HTTP_ACL_NETDATACONF;
126
}
system/systemd/netdata.service.in
+1
-1
@@ -1,6 +1,6 @@
1
# SPDX-License-Identifier: GPL-3.0-or-later
2
[Unit]
3
-Description=Real time performance monitoring
3
+Description=infrastructure monitoring and troubleshooting, transformed
4
5
# append here other services you want netdata to wait for them to start
6
After=network.target network-online.target nss-lookup.target
system/systemd/netdata.service.v235.in
+1
-1
@@ -1,6 +1,6 @@
1
# SPDX-License-Identifier: GPL-3.0-or-later
2
[Unit]
3
-Description=Real time performance monitoring
3
+Description=infrastructure monitoring and troubleshooting, transformed
4
5
# append here other services you want netdata to wait for them to start
6
After=network.target network-online.target nss-lookup.target