backend now sends all metrics for all hosts
Costa Tsaousis (ktsaou) committed
Feb 18, 2017 at 17:23 UTC
11f48b1835cdb3f8a9f4fd495b9b30cb6b3807a7
3 files changed
+73
-74
src/backends.c
+71
-72
@@ -152,19 +152,20 @@ void *backends_main(void *ptr) {
152
int default_port = 0;
153
int sock = -1;
154
uint32_t options;
155
- int enabled = config_get_boolean("backend", "enabled", 0);
156
- const char *source = config_get("backend", "data source", "average");
157
- const char *type = config_get("backend", "type", "graphite");
155
+ int enabled = config_get_boolean("backend", "enabled", 0);
156
+ const char *source = config_get("backend", "data source", "average");
157
+ const char *type = config_get("backend", "type", "graphite");
158
const char *destination = config_get("backend", "destination", "localhost");
159
- const char *prefix = config_get("backend", "prefix", "netdata");
160
- const char *hostname = config_get("backend", "hostname", localhost->hostname);
161
- int frequency = (int)config_get_number("backend", "update every", 10);
162
- int buffer_on_failures = (int)config_get_number("backend", "buffer on failures", 10);
163
- long timeoutms = config_get_number("backend", "timeout ms", frequency * 2 * 1000);
159
+ const char *prefix = config_get("backend", "prefix", "netdata");
160
+ const char *hostname = config_get("backend", "hostname", localhost->hostname);
161
+ int frequency = (int)config_get_number("backend", "update every", 10);
162
+ int buffer_on_failures = (int)config_get_number("backend", "buffer on failures", 10);
163
+ long timeoutms = config_get_number("backend", "timeout ms", frequency * 2 * 1000);
164
165
// ------------------------------------------------------------------------
166
// validate configuration options
167
// and prepare for sending data to our backend
168
+
169
if(!enabled || frequency < 1)
170
goto cleanup;
171
@@ -182,6 +183,17 @@ void *backends_main(void *ptr) {
183
goto cleanup;
184
}
185
186
+ if(timeoutms < 1) {
187
+ error("BACKED invalid timeout %ld ms given. Assuming %d ms.", timeoutms, frequency * 2 * 1000);
188
+ timeoutms = frequency * 2 * 1000;
189
+ }
190
+ timeout.tv_sec = (timeoutms * 1000) / 1000000;
191
+ timeout.tv_usec = (timeoutms * 1000) % 1000000;
192
+
193
+
194
+ // ------------------------------------------------------------------------
195
+ // select the backend type
196
+
197
if(!strcmp(type, "graphite") || !strcmp(type, "graphite:plaintext")) {
198
default_port = 2003;
199
if(options == BACKEND_SOURCE_DATA_AS_COLLECTED)
@@ -210,15 +222,9 @@ void *backends_main(void *ptr) {
222
goto cleanup;
223
}
224
213
- if(timeoutms < 1) {
214
- error("BACKED invalid timeout %ld ms given. Assuming %d ms.", timeoutms, frequency * 2 * 1000);
215
- timeoutms = frequency * 2 * 1000;
216
- }
217
- timeout.tv_sec = (timeoutms * 1000) / 1000000;
218
- timeout.tv_usec = (timeoutms * 1000) % 1000000;
225
226
// ------------------------------------------------------------------------
221
- // prepare the charts for monitoring the backend
227
+ // prepare the charts for monitoring the backend operation
228
229
struct rusage thread;
230
@@ -237,37 +243,23 @@ void *backends_main(void *ptr) {
243
chart_backend_reconnects = 0,
244
chart_backend_latency = 0;
245
240
- RRDSET *chart_metrics = rrdset_find_localhost("netdata.backend_metrics");
241
- if(!chart_metrics) {
242
- chart_metrics = rrdset_create_localhost("netdata", "backend_metrics", NULL, "backend", NULL
243
- , "Netdata Buffered Metrics", "metrics", 130600, frequency
244
- , RRDSET_TYPE_LINE);
245
- rrddim_add(chart_metrics, "buffered", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
246
- rrddim_add(chart_metrics, "lost", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
247
- rrddim_add(chart_metrics, "sent", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
248
- }
246
+ RRDSET *chart_metrics = rrdset_create_localhost("netdata", "backend_metrics", NULL, "backend", NULL, "Netdata Buffered Metrics", "metrics", 130600, frequency, RRDSET_TYPE_LINE);
247
+ rrddim_add(chart_metrics, "buffered", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
248
+ rrddim_add(chart_metrics, "lost", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
249
+ rrddim_add(chart_metrics, "sent", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
250
250
- RRDSET *chart_bytes = rrdset_find_localhost("netdata.backend_bytes");
251
- if(!chart_bytes) {
252
- chart_bytes = rrdset_create_localhost("netdata", "backend_bytes", NULL, "backend", NULL
253
- , "Netdata Backend Data Size", "KB", 130610, frequency, RRDSET_TYPE_AREA);
254
- rrddim_add(chart_bytes, "buffered", NULL, 1, 1024, RRD_ALGORITHM_ABSOLUTE);
255
- rrddim_add(chart_bytes, "lost", NULL, 1, 1024, RRD_ALGORITHM_ABSOLUTE);
256
- rrddim_add(chart_bytes, "sent", NULL, 1, 1024, RRD_ALGORITHM_ABSOLUTE);
257
- rrddim_add(chart_bytes, "received", NULL, 1, 1024, RRD_ALGORITHM_ABSOLUTE);
258
- }
251
+ RRDSET *chart_bytes = rrdset_create_localhost("netdata", "backend_bytes", NULL, "backend", NULL, "Netdata Backend Data Size", "KB", 130610, frequency, RRDSET_TYPE_AREA);
252
+ rrddim_add(chart_bytes, "buffered", NULL, 1, 1024, RRD_ALGORITHM_ABSOLUTE);
253
+ rrddim_add(chart_bytes, "lost", NULL, 1, 1024, RRD_ALGORITHM_ABSOLUTE);
254
+ rrddim_add(chart_bytes, "sent", NULL, 1, 1024, RRD_ALGORITHM_ABSOLUTE);
255
+ rrddim_add(chart_bytes, "received", NULL, 1, 1024, RRD_ALGORITHM_ABSOLUTE);
256
260
- RRDSET *chart_ops = rrdset_find_localhost("netdata.backend_ops");
261
- if(!chart_ops) {
262
- chart_ops = rrdset_create_localhost("netdata", "backend_ops", NULL, "backend", NULL
263
- , "Netdata Backend Operations", "operations", 130630, frequency
264
- , RRDSET_TYPE_LINE);
265
- rrddim_add(chart_ops, "write", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
266
- rrddim_add(chart_ops, "discard", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
267
- rrddim_add(chart_ops, "reconnect", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
268
- rrddim_add(chart_ops, "failure", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
269
- rrddim_add(chart_ops, "read", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
270
- }
257
+ RRDSET *chart_ops = rrdset_create_localhost("netdata", "backend_ops", NULL, "backend", NULL, "Netdata Backend Operations", "operations", 130630, frequency, RRDSET_TYPE_LINE);
258
+ rrddim_add(chart_ops, "write", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
259
+ rrddim_add(chart_ops, "discard", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
260
+ rrddim_add(chart_ops, "reconnect", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
261
+ rrddim_add(chart_ops, "failure", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
262
+ rrddim_add(chart_ops, "read", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
263
264
/*
265
* this is misleading - we can only measure the time we need to send data
@@ -276,21 +268,14 @@ void *backends_main(void *ptr) {
268
*
269
* issue #1432 and https://www.softlab.ntua.gr/facilities/documentation/unix/unix-socket-faq/unix-socket-faq-2.html
270
*
279
- RRDSET *chart_latency = rrdset_find_localhost("netdata.backend_latency");
280
- if(!chart_latency) {
281
- chart_latency = rrdset_create_localhost("netdata", "backend_latency", NULL, "backend", NULL, "Netdata Backend Latency", "ms", 130620, frequency, RRDSET_TYPE_AREA);
282
- rrddim_add(chart_latency, "latency", NULL, 1, 1000, RRD_ALGORITHM_ABSOLUTE);
283
- }
271
+ RRDSET *chart_latency = rrdset_create_localhost("netdata", "backend_latency", NULL, "backend", NULL, "Netdata Backend Latency", "ms", 130620, frequency, RRDSET_TYPE_AREA);
272
+ rrddim_add(chart_latency, "latency", NULL, 1, 1000, RRD_ALGORITHM_ABSOLUTE);
273
*/
274
286
- RRDSET *chart_rusage = rrdset_find_localhost("netdata.backend_thread_cpu");
287
- if(!chart_rusage) {
288
- chart_rusage = rrdset_create_localhost("netdata", "backend_thread_cpu", NULL, "backend", NULL
289
- , "NetData Backend Thread CPU usage", "milliseconds/s", 130630, frequency
290
- , RRDSET_TYPE_STACKED);
291
- rrddim_add(chart_rusage, "user", NULL, 1, 1000, RRD_ALGORITHM_INCREMENTAL);
292
- rrddim_add(chart_rusage, "system", NULL, 1, 1000, RRD_ALGORITHM_INCREMENTAL);
293
- }
275
+ RRDSET *chart_rusage = rrdset_create_localhost("netdata", "backend_thread_cpu", NULL, "backend", NULL, "NetData Backend Thread CPU usage", "milliseconds/s", 130630, frequency, RRDSET_TYPE_STACKED);
276
+ rrddim_add(chart_rusage, "user", NULL, 1, 1000, RRD_ALGORITHM_INCREMENTAL);
277
+ rrddim_add(chart_rusage, "system", NULL, 1, 1000, RRD_ALGORITHM_INCREMENTAL);
278
+
279
280
// ------------------------------------------------------------------------
281
// prepare the backend main loop
@@ -304,32 +289,46 @@ void *backends_main(void *ptr) {
289
heartbeat_init(&hb);
290
291
for(;;) {
292
+
293
// ------------------------------------------------------------------------
294
// Wait for the next iteration point.
295
heartbeat_next(&hb, step_ut);
296
time_t before = now_realtime_sec();
297
298
+
299
// ------------------------------------------------------------------------
300
// add to the buffer the data we need to send to the backend
314
- RRDSET *st;
301
+
302
int pthreadoldcancelstate;
303
304
if(unlikely(pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &pthreadoldcancelstate) != 0))
305
error("Cannot set pthread cancel state to DISABLE.");
306
320
- rrdhost_rdlock(localhost);
321
- for(st = localhost->rrdset_root; st ;st = st->next) {
322
- pthread_rwlock_rdlock(&st->rwlock);
307
+ RRDHOST *host;
308
+ for(host = localhost; host ; host = host->next) {
309
+ // for each host
310
+
311
+ rrdhost_rdlock(host);
312
+
313
+ RRDSET *st;
314
+ for(st = host->rrdset_root; st; st = st->next) {
315
+ // for each chart
316
+
317
+ pthread_rwlock_rdlock(&st->rwlock);
318
+
319
+ RRDDIM *rd;
320
+ for(rd = st->dimensions; rd; rd = rd->next) {
321
+ // for each dimension
322
+
323
+ if(rd->last_collected_time.tv_sec >= after)
324
+ chart_buffered_metrics += backend_request_formatter(b, prefix, host, (host == localhost)?hostname:host->hostname, st, rd, after, before, options);
325
+ }
326
324
- RRDDIM *rd;
325
- for(rd = st->dimensions; rd ;rd = rd->next) {
326
- if(rd->last_collected_time.tv_sec >= after)
327
- chart_buffered_metrics += backend_request_formatter(b, prefix, localhost, hostname, st, rd, after, before, options);
327
+ pthread_rwlock_unlock(&st->rwlock);
328
}
329
330
- pthread_rwlock_unlock(&st->rwlock);
330
+ rrdhost_unlock(host);
331
}
332
- rrdhost_unlock(localhost);
332
333
if(unlikely(pthread_setcancelstate(pthreadoldcancelstate, NULL) != 0))
334
error("Cannot set pthread cancel state to RESTORE (%d).", pthreadoldcancelstate);
@@ -488,7 +487,7 @@ void *backends_main(void *ptr) {
487
// ------------------------------------------------------------------------
488
// update the monitoring charts
489
491
- if(chart_ops->counter_done) rrdset_next(chart_ops);
490
+ if(likely(chart_ops->counter_done)) rrdset_next(chart_ops);
491
rrddim_set(chart_ops, "read", chart_receptions);
492
rrddim_set(chart_ops, "write", chart_transmission_successes);
493
rrddim_set(chart_ops, "discard", chart_data_lost_events);
@@ -496,13 +495,13 @@ void *backends_main(void *ptr) {
495
rrddim_set(chart_ops, "reconnect", chart_backend_reconnects);
496
rrdset_done(chart_ops);
497
499
- if(chart_metrics->counter_done) rrdset_next(chart_metrics);
498
+ if(likely(chart_metrics->counter_done)) rrdset_next(chart_metrics);
499
rrddim_set(chart_metrics, "buffered", chart_buffered_metrics);
500
rrddim_set(chart_metrics, "lost", chart_lost_metrics);
501
rrddim_set(chart_metrics, "sent", chart_sent_metrics);
502
rrdset_done(chart_metrics);
503
505
- if(chart_bytes->counter_done) rrdset_next(chart_bytes);
504
+ if(likely(chart_bytes->counter_done)) rrdset_next(chart_bytes);
505
rrddim_set(chart_bytes, "buffered", chart_buffered_bytes);
506
rrddim_set(chart_bytes, "lost", chart_lost_bytes);
507
rrddim_set(chart_bytes, "sent", chart_sent_bytes);
@@ -510,13 +509,13 @@ void *backends_main(void *ptr) {
509
rrdset_done(chart_bytes);
510
511
/*
513
- if(chart_latency->counter_done) rrdset_next(chart_latency);
512
+ if(likely(chart_latency->counter_done)) rrdset_next(chart_latency);
513
rrddim_set(chart_latency, "latency", chart_backend_latency);
514
rrdset_done(chart_latency);
515
*/
516
517
getrusage(RUSAGE_THREAD, &thread);
519
- if(chart_rusage->counter_done) rrdset_next(chart_rusage);
518
+ if(likely(chart_rusage->counter_done)) rrdset_next(chart_rusage);
519
rrddim_set(chart_rusage, "user", thread.ru_utime.tv_sec * 1000000ULL + thread.ru_utime.tv_usec);
520
rrddim_set(chart_rusage, "system", thread.ru_stime.tv_sec * 1000000ULL + thread.ru_stime.tv_usec);
521
rrdset_done(chart_rusage);
src/rrd.c
+1
-1
@@ -33,7 +33,7 @@ inline const char *rrd_memory_mode_name(RRD_MEMORY_MODE id) {
33
}
34
}
35
36
-int rrd_memory_mode_id(const char *name) {
36
+RRD_MEMORY_MODE rrd_memory_mode_id(const char *name) {
37
if(unlikely(!strcmp(name, RRD_MEMORY_MODE_RAM_NAME)))
38
return RRD_MEMORY_MODE_RAM;
39
else if(unlikely(!strcmp(name, RRD_MEMORY_MODE_MAP_NAME)))
src/rrd.h
+1
-1
@@ -50,7 +50,7 @@ typedef enum rrd_memory_mode {
50
RRD_MEMORY_MODE rrd_memory_mode;
51
52
extern const char *rrd_memory_mode_name(RRD_MEMORY_MODE id);
53
-extern int rrd_memory_mode_id(const char *name);
53
+extern RRD_MEMORY_MODE rrd_memory_mode_id(const char *name);
54
55
56
// ----------------------------------------------------------------------------