@cryptotaxi247 / netdata / commits / b018c0a13

Address eBPF and Cgroup exit and messages (#22574)

thiagoftsm committed May 28, 2026 at 11:28 UTC b018c0a13ee7adef95b3685835214c73cf95f963

20 files changed +149 -18

src/collectors/cgroups.plugin/cgroup-discovery.c

+63 -4

@@ -24,6 +24,10 @@ struct cgroup *discovered_cgroup_root = NULL;

24 char cgroup_chart_id_prefix[] = "cgroup_";

25 char services_chart_id_prefix[] = "systemd_";

26 const char *cgroups_rename_script = NULL;

27 +static DICTIONARY *discovery_walkdir_open_errors = NULL;

28 +static size_t discovery_walkdir_open_errors_count = 0;

29 +

30 +#define DISCOVERY_WALKDIR_OPEN_ERRORS_MAX 256

32 // (legacy SHM globals removed — replaced by netipc in cgroup-netipc.c)

@@ -349,6 +353,54 @@ static inline struct cgroup *discovery_cgroup_find(const char *id) {

353 return cg;

354 }

355

356 +struct discovery_walkdir_open_error_prune {

357 + size_t remaining;

358 +};

359 +

360 +static int discovery_walkdir_open_error_prune_cb(const DICTIONARY_ITEM *item, void *value __maybe_unused, void *data) {

361 + struct discovery_walkdir_open_error_prune *prune = data;

362 +

363 + if (!prune->remaining)

364 + return -1;

365 +

366 + dictionary_del(discovery_walkdir_open_errors, dictionary_acquired_item_name(item));

367 + discovery_walkdir_open_errors_count--;

368 + prune->remaining--;

369 +

370 + return prune->remaining ? 0 : -1;

371 +}

372 +

373 +static inline bool discovery_walkdir_open_error_first(const char *dirpath) {

374 + if (unlikely(!discovery_walkdir_open_errors))

375 + discovery_walkdir_open_errors = dictionary_create(DICT_OPTION_SINGLE_THREADED);

376 +

377 + if (dictionary_get(discovery_walkdir_open_errors, dirpath))

378 + return false;

379 +

380 + if (discovery_walkdir_open_errors_count >= DISCOVERY_WALKDIR_OPEN_ERRORS_MAX) {

381 + struct discovery_walkdir_open_error_prune prune = {

382 + .remaining = discovery_walkdir_open_errors_count - DISCOVERY_WALKDIR_OPEN_ERRORS_MAX + 1,

383 + };

384 +

385 + dictionary_walkthrough_write(discovery_walkdir_open_errors, discovery_walkdir_open_error_prune_cb, &prune);

386 + }

387 +

388 + uint8_t logged = 1;

389 + dictionary_set(discovery_walkdir_open_errors, dirpath, &logged, sizeof(logged));

390 + discovery_walkdir_open_errors_count++;

391 +

392 + return true;

393 +}

394 +

395 +static inline void discovery_walkdir_open_errors_cleanup(void) {

396 + if (!discovery_walkdir_open_errors)

397 + return;

398 +

399 + dictionary_destroy(discovery_walkdir_open_errors);

400 + discovery_walkdir_open_errors = NULL;

401 + discovery_walkdir_open_errors_count = 0;

402 +}

403 +

404 static int calc_cgroup_depth(const char *id) {

405 int depth = 0;

406 const char *s;

@@ -407,10 +459,16 @@ static inline int discovery_find_walkdir(const char *base, const char *dirpath)

459

460 DIR *dir = opendir(dirpath);

461 if(!dir) {

410 - if(errno == EACCES && strcmp(dirpath, base) != 0)

411 - nd_log_collector(NDLP_DEBUG, "CGROUP: cannot open directory '%s': %s", dirpath, strerror(errno));

412 - else

413 - collector_error("CGROUP: cannot open directory '%s': %s", dirpath, strerror(errno));

462 + int err = errno;

463 + bool first_log_for_path = discovery_walkdir_open_error_first(dirpath);

464 +

465 + if(err == EACCES && strcmp(dirpath, base) != 0) {

466 + if (first_log_for_path)

467 + nd_log_collector(NDLP_DEBUG, "CGROUP: cannot open directory '%s': %s", dirpath, strerror(err));

468 + }

469 + else if (first_log_for_path) {

470 + collector_error("CGROUP: cannot open directory '%s': %s", dirpath, strerror(err));

471 + }

472 return ret;

473 }

474 ret = 1;

@@ -1187,6 +1245,7 @@ void cgroup_discovery_worker(void *ptr)

1245

1246 // Stop the netipc server first so its worker threads cannot iterate cgroup_root while we free it.

1247 cgroup_netipc_cleanup();

1248 + discovery_walkdir_open_errors_cleanup();

1249

1250 // free all cgroups

1251 netdata_mutex_lock(&cgroup_root_mutex);

src/collectors/ebpf.plugin/ebpf.c

+39 -7

@@ -34,6 +34,7 @@ uint32_t integration_with_collectors = NETDATA_EBPF_INTEGRATION_DISABLED;

34 ND_THREAD *socket_ipc = NULL;

35 static size_t global_iterations_counter = 1;

36 bool publish_internal_metrics = true;

37 +bool ebpf_program_loaded_any = false;

39 netdata_mutex_t lock;

40 netdata_mutex_t ebpf_exit_cleanup;

@@ -1014,7 +1015,7 @@ static inline void ebpf_check_before2go()

1015 /**

1016 * Close the collector gracefully

1017 */

1017 -static void ebpf_exit()

1018 +static void ebpf_cleanup(void)

1019 {

1020 #ifdef LIBBPF_MAJOR_VERSION

1021 netdata_mutex_lock(&ebpf_exit_cleanup);

@@ -1042,8 +1043,12 @@ static void ebpf_exit()

1043 }

1044 ebpf_cgroup_cache_cleanup();

1045 netdata_integration_cleanup_shm();

1046 +}

1047

1046 - exit(0);

1048 +static void ebpf_exit(int exit_code)

1049 +{

1050 + ebpf_cleanup();

1051 + exit(exit_code);

1052 }

1053

1054 /**

@@ -1161,7 +1166,7 @@ void ebpf_stop_threads(int sig)

1166 #endif

1167 netdata_mutex_unlock(&mutex_cgroup_shm);

1168

1164 - // Join the cgroup integration thread before ebpf_exit() tears down the netipc cache it reads.

1169 + // Join the cgroup integration thread before cleanup tears down the netipc cache it reads.

1170 if (cgroup_integration_thread.thread) {

1171 nd_thread_join(cgroup_integration_thread.thread);

1172 cgroup_integration_thread.thread = NULL;

@@ -1185,7 +1190,7 @@ void ebpf_stop_threads(int sig)

1190 netdata_log_info(

1191 "EBPF SHUTDOWN: total stop duration %llums.", (unsigned long long)(total_duration_ut / USEC_PER_MS));

1192

1188 - ebpf_exit();

1193 + ebpf_cleanup();

1194 }

1195

1196 /**

@@ -1545,7 +1550,7 @@ static void ebpf_parse_args(int argc, char **argv)

1550 netdata_log_error(

1551 "Cannot read process groups '%s/apps_groups.conf'. There are no internal defaults. Failing.",

1552 ebpf_stock_config_dir);

1548 - ebpf_exit();

1553 + ebpf_exit(1);

1554 }

1555 } else

1556 netdata_log_info("Loaded config file '%s/apps_groups.conf'", ebpf_user_config_dir);

@@ -2295,6 +2300,26 @@ static void ebpf_signal_stop_handler(int sig)

2300 ebpf_stop_signal = (sig > 0) ? sig : 1;

2301 }

2302

2303 +static bool ebpf_all_enabled_threads_stopped(void)

2304 +{

2305 + bool any_enabled = false;

2306 +

2307 + for (size_t i = 0; ebpf_threads[i].name != NULL; i++) {

2308 + if (!ebpf_threads[i].enabled)

2309 + continue;

2310 +

2311 + any_enabled = true;

2312 +

2313 + if (!ebpf_threads[i].thread)

2314 + continue;

2315 +

2316 + if (ebpf_module_enabled_get(&ebpf_modules[i]) < NETDATA_THREAD_EBPF_STOPPING)

2317 + return false;

2318 + }

2319 +

2320 + return any_enabled;

2321 +}

2322 +

2323 /**

2324 * Entry point

2325 *

@@ -2344,7 +2369,7 @@ int main(int argc, char **argv)

2369

2370 netdata_configured_host_prefix = getenv("NETDATA_HOST_PREFIX");

2371 if (verify_netdata_host_prefix(true) == -1)

2347 - ebpf_exit();

2372 + ebpf_exit(1);

2373

2374 ebpf_allocate_common_vectors();

2375

@@ -2402,6 +2427,7 @@ int main(int argc, char **argv)

2427 heartbeat_init(&hb, USEC_PER_SEC);

2428 int update_apps_every = (int)EBPF_CFG_UPDATE_APPS_EVERY_DEFAULT;

2429 int update_apps_list = update_apps_every - 1;

2430 + int exit_code = 0;

2431 //Plugin will be killed when it receives a signal

2432 for (; !ebpf_plugin_stop(); global_iterations_counter++) {

2433 (void)heartbeat_next(&hb);

@@ -2414,6 +2440,12 @@ int main(int argc, char **argv)

2440 if (ebpf_plugin_stop())

2441 break;

2442

2443 + if (!ebpf_program_loaded() && ebpf_all_enabled_threads_stopped()) {

2444 + netdata_log_error("EBPF: failed to load any eBPF program, shutting down.");

2445 + exit_code = 1;

2446 + break;

2447 + }

2448 +

2449 if (global_iterations_counter % EBPF_DEFAULT_UPDATE_EVERY == 0) {

2450 netdata_mutex_lock(&lock);

2451 ebpf_create_statistic_charts(EBPF_DEFAULT_UPDATE_EVERY);

@@ -2438,5 +2470,5 @@ int main(int argc, char **argv)

2470

2471 ebpf_stop_threads((int)ebpf_stop_signal);

2472

2441 - return 0;

2473 + return exit_code;

2474 }

src/collectors/ebpf.plugin/ebpf.h

+11

@@ -121,6 +121,7 @@ typedef struct netdata_ebpf_judy_pid_stats {

121 } netdata_ebpf_judy_pid_stats_t;

122

123 extern ebpf_module_t ebpf_modules[];

124 +extern bool ebpf_program_loaded_any;

125

126 typedef struct ebpf_tracepoint {

127 bool enabled;

@@ -367,6 +368,16 @@ static inline bool ebpf_plugin_stop(void)

368 nd_thread_signaled_to_cancel();

369 }

370

371 +static inline void ebpf_mark_program_loaded(void)

372 +{

373 + __atomic_store_n(&ebpf_program_loaded_any, true, __ATOMIC_RELEASE);

374 +}

375 +

376 +static inline bool ebpf_program_loaded(void)

377 +{

378 + return __atomic_load_n(&ebpf_program_loaded_any, __ATOMIC_ACQUIRE);

379 +}

380 +

381 // `enabled` is sampled from stats/shutdown paths without a single shared mutex.

382 // Keep those state transitions defined without changing the plugin's lock layout.

383 static inline enum ebpf_threads_status ebpf_module_enabled_get(ebpf_module_t *em)

src/collectors/ebpf.plugin/ebpf_cachestat.c

@@ -1850,6 +1850,7 @@ void ebpf_cachestat_thread(void *ptr)

1850 if (ebpf_cachestat_load_bpf(em)) {

1851 goto endcachestat;

1852 }

1853 + ebpf_mark_program_loaded();

1854

1855 ebpf_cachestat_allocate_global_vectors();

1856

src/collectors/ebpf.plugin/ebpf_dcstat.c

@@ -1590,6 +1590,7 @@ void ebpf_dcstat_thread(void *ptr)

1590 if (ebpf_dcstat_load_bpf(em)) {

1591 goto enddcstat;

1592 }

1593 + ebpf_mark_program_loaded();

1594

1595 ebpf_dcstat_allocate_global_vectors();

1596

src/collectors/ebpf.plugin/ebpf_disk.c

@@ -925,6 +925,7 @@ void ebpf_disk_thread(void *ptr)

925 if (ebpf_disk_load_bpf(em)) {

926 goto enddisk;

927 }

928 + ebpf_mark_program_loaded();

929

930 int algorithms[NETDATA_EBPF_HIST_MAX_BINS];

931 ebpf_fill_algorithms(algorithms, NETDATA_EBPF_HIST_MAX_BINS, NETDATA_EBPF_INCREMENTAL_IDX);

src/collectors/ebpf.plugin/ebpf_fd.c

@@ -1689,6 +1689,7 @@ void ebpf_fd_thread(void *ptr)

1689 if (ebpf_fd_load_bpf(em)) {

1690 goto endfd;

1691 }

1692 + ebpf_mark_program_loaded();

1693

1694 ebpf_fd_allocate_global_vectors();

1695

src/collectors/ebpf.plugin/ebpf_filesystem.c

+16 -7

@@ -618,8 +618,15 @@ int ebpf_filesystem_initialize_ebpf_data(ebpf_module_t *em)

618 int i;

619 const char *saved_name = em->info.thread_name;

620 uint64_t kernels = em->kernels;

621 + bool loaded_any = false;

622 for (i = 0; localfs[i].filesystem; i++) {

623 ebpf_filesystem_partitions_t *efp = &localfs[i];

624 + if (efp->flags & NETDATA_FILESYSTEM_FLAG_HAS_PARTITION)

625 + loaded_any = true;

626 +

627 + if (efp->load_failed)

628 + continue;

629 +

630 if (!efp->probe_links && efp->flags & NETDATA_FILESYSTEM_LOAD_EBPF_PROGRAM) {

631 em->info.thread_name = efp->filesystem;

632 em->kernels = efp->kernels;

@@ -630,34 +637,36 @@ int ebpf_filesystem_initialize_ebpf_data(ebpf_module_t *em)

637 if (em->load & EBPF_LOAD_LEGACY) {

638 efp->probe_links = ebpf_load_program(ebpf_plugin_dir, em, running_on_kernel, isrh, &efp->objects);

639 if (!efp->probe_links) {

640 + efp->load_failed = true;

641 em->info.thread_name = saved_name;

642 em->kernels = kernels;

643 em->maps = NULL;

636 - netdata_mutex_unlock(&lock);

637 - return -1;

644 + continue;

645 }

646 }

647 #ifdef LIBBPF_MAJOR_VERSION

648 else {

649 efp->fs_obj = filesystem_bpf__open();

650 if (!efp->fs_obj) {

651 + efp->load_failed = true;

652 em->info.thread_name = saved_name;

653 em->kernels = kernels;

654 em->maps = NULL;

647 - netdata_mutex_unlock(&lock);

648 - return -1;

655 + continue;

656 } else if (ebpf_fs_load_and_attach(em->maps, efp->fs_obj, efp->functions, NULL)) {

657 + efp->load_failed = true;

658 filesystem_bpf__destroy(efp->fs_obj);

659 efp->fs_obj = NULL;

660 em->info.thread_name = saved_name;

661 em->kernels = kernels;

662 em->maps = NULL;

655 - netdata_mutex_unlock(&lock);

656 - return -1;

663 + continue;

664 }

665 }

666 #endif

667 efp->flags |= NETDATA_FILESYSTEM_FLAG_HAS_PARTITION;

668 + loaded_any = true;

669 + ebpf_mark_program_loaded();

670 ebpf_update_kernel_memory(&plugin_statistics, efp->fs_maps, EBPF_ACTION_STAT_ADD);

671

672 // Needed for filesystems like btrfs

@@ -682,7 +691,7 @@ int ebpf_filesystem_initialize_ebpf_data(ebpf_module_t *em)

691 filesystem_hash_values = callocz(ebpf_nprocs, sizeof(netdata_idx_t));

692 }

693

685 - return 0;

694 + return loaded_any ? 0 : -1;

695 }

696

697 /**

src/collectors/ebpf.plugin/ebpf_hardirq.c

@@ -696,6 +696,7 @@ void ebpf_hardirq_thread(void *ptr)

696 if (ebpf_hardirq_load_bpf(em)) {

697 goto endhardirq;

698 }

699 + ebpf_mark_program_loaded();

700

701 ebpf_hardirq_allocate_global_vectors();

702

src/collectors/ebpf.plugin/ebpf_mdflush.c

@@ -418,6 +418,7 @@ void ebpf_mdflush_thread(void *ptr)

418 netdata_log_error("Cannot load eBPF software.");

419 goto endmdflush;

420 }

421 + ebpf_mark_program_loaded();

422

423 mdflush_safe_clean = true;

424 mdflush_collector(em);

src/collectors/ebpf.plugin/ebpf_mount.c

@@ -527,6 +527,7 @@ void ebpf_mount_thread(void *ptr)

527 if (ebpf_mount_load_bpf(em)) {

528 goto endmount;

529 }

530 + ebpf_mark_program_loaded();

531

532 int algorithms[NETDATA_EBPF_MOUNT_SYSCALL] = {NETDATA_EBPF_INCREMENTAL_IDX, NETDATA_EBPF_INCREMENTAL_IDX};

533

src/collectors/ebpf.plugin/ebpf_oomkill.c

@@ -590,6 +590,7 @@ void ebpf_oomkill_thread(void *ptr)

590 if (!em->probe_links) {

591 goto endoomkill;

592 }

593 + ebpf_mark_program_loaded();

594

595 netdata_mutex_lock(&lock);

596 ebpf_update_stats(&plugin_statistics, em);

src/collectors/ebpf.plugin/ebpf_process.c

@@ -1825,6 +1825,8 @@ void ebpf_process_thread(void *ptr)

1825 if (ebpf_process_load_bpf(em)) {

1826 em->global_charts = em->apps_charts = em->cgroup_charts = NETDATA_THREAD_EBPF_STOPPING;

1827 ebpf_module_enabled_set(em, NETDATA_THREAD_EBPF_STOPPING);

1828 + } else {

1829 + ebpf_mark_program_loaded();

1830 }

1831

1832 int algorithms[NETDATA_KEY_PUBLISH_PROCESS_END] = {

src/collectors/ebpf.plugin/ebpf_shm.c

@@ -1447,6 +1447,7 @@ void ebpf_shm_thread(void *ptr)

1447 if (ebpf_shm_load_bpf(em)) {

1448 goto endshm;

1449 }

1450 + ebpf_mark_program_loaded();

1451

1452 ebpf_shm_allocate_global_vectors();

1453

src/collectors/ebpf.plugin/ebpf_socket.c

@@ -3126,6 +3126,7 @@ void ebpf_socket_thread(void *ptr)

3126 if (ebpf_socket_load_bpf(em)) {

3127 goto endsocket;

3128 }

3129 + ebpf_mark_program_loaded();

3130

3131 int algorithms[NETDATA_MAX_SOCKET_VECTOR] = {

3132 NETDATA_EBPF_ABSOLUTE_IDX,

src/collectors/ebpf.plugin/ebpf_softirq.c

@@ -263,6 +263,7 @@ void ebpf_softirq_thread(void *ptr)

263 if (!em->probe_links) {

264 goto endsoftirq;

265 }

266 + ebpf_mark_program_loaded();

267

268 softirq_safe_clean = true;

269 softirq_collector(em);

src/collectors/ebpf.plugin/ebpf_swap.c

@@ -1332,6 +1332,7 @@ void ebpf_swap_thread(void *ptr)

1332 if (ebpf_swap_load_bpf(em)) {

1333 goto endswap;

1334 }

1335 + ebpf_mark_program_loaded();

1336

1337 ebpf_swap_allocate_global_vectors();

1338

src/collectors/ebpf.plugin/ebpf_sync.c

@@ -360,6 +360,8 @@ static int ebpf_sync_initialize_syscall(ebpf_module_t *em)

360 if (em->load & EBPF_LOAD_LEGACY) {

361 if (ebpf_sync_load_legacy(w, em))

362 errors++;

363 + else

364 + ebpf_mark_program_loaded();

365

366 em->info.thread_name = saved_name;

367 }

@@ -378,6 +380,8 @@ static int ebpf_sync_initialize_syscall(ebpf_module_t *em)

380 w->sync_obj = NULL;

381 w->enabled = false;

382 errors++;

383 + } else {

384 + ebpf_mark_program_loaded();

385 }

386 }

387 } else {

src/collectors/ebpf.plugin/ebpf_vfs.c

@@ -2985,6 +2985,7 @@ void ebpf_vfs_thread(void *ptr)

2985 if (ebpf_vfs_load_bpf(em)) {

2986 goto endvfs;

2987 }

2988 + ebpf_mark_program_loaded();

2989

2990 int algorithms[NETDATA_KEY_PUBLISH_VFS_END] = {

2991 NETDATA_EBPF_INCREMENTAL_IDX,

src/collectors/ebpf.plugin/libbpf_api/ebpf.h

@@ -418,6 +418,7 @@ typedef struct ebpf_filesystem_partitions {

418 char *family_name;

419 struct bpf_object *objects;

420 struct bpf_link **probe_links;

421 + bool load_failed;

422

423 netdata_ebpf_histogram_t hread;

424 netdata_ebpf_histogram_t hwrite;