3 #define BACKEND_SOURCE_DATA_AS_COLLECTED 0x00000001
4 #define BACKEND_SOURCE_DATA_AVERAGE 0x00000002
5 #define BACKEND_SOURCE_DATA_SUM 0x00000004
7 static inline calculated_number backend_calculate_value_from_stored_data(RRDSET *st, RRDDIM *rd, time_t after, time_t before, uint32_t options) {
8 time_t first_t = rrdset_first_entry_t(st);
9 time_t last_t = rrdset_last_entry_t(st);
11 if(unlikely(before < first_t || after > last_t))
12 // the chart has not been updated in the wanted timeframe
15 // align the time-frame
16 // for 'after' also skip the first value by adding st->update_every
17 after = after - after % st->update_every + st->update_every;
18 before = before - before % st->update_every;
20 if(unlikely(after < first_t))
23 if(unlikely(after > before))
24 // this can happen when the st->update_every > before - after
27 if(unlikely(before > last_t))
31 calculated_number sum = 0;
33 long start_at_slot = rrdset_time2slot(st, before),
34 stop_at_slot = rrdset_time2slot(st, after),
37 for(slot = start_at_slot; !stop_now ; slot--) {
38 if(unlikely(slot < 0)) slot = st->entries - 1;
39 if(unlikely(slot == stop_at_slot)) stop_now = 1;
41 storage_number n = rd->values[slot];
42 if(unlikely(!does_storage_number_exist(n))) continue;
44 calculated_number value = unpack_storage_number(n);
49 if(unlikely(!counter))
52 if(unlikely(options & BACKEND_SOURCE_DATA_SUM))
55 return sum / (calculated_number)counter;
58 static inline int format_dimension_collected_graphite_plaintext(BUFFER *b, const char *prefix, RRDHOST *host, const char *hostname, RRDSET *st, RRDDIM *rd, time_t after, time_t before, uint32_t options) {
63 buffer_sprintf(b, "%s.%s.%s.%s " COLLECTED_NUMBER_FORMAT " %u\n", prefix, hostname, st->id, rd->id, rd->last_collected_value, (uint32_t)rd->last_collected_time.tv_sec);
67 static inline int format_dimension_stored_graphite_plaintext(BUFFER *b, const char *prefix, RRDHOST *host, const char *hostname, RRDSET *st, RRDDIM *rd, time_t after, time_t before, uint32_t options) {
69 calculated_number value = backend_calculate_value_from_stored_data(st, rd, after, before, options);
71 buffer_sprintf(b, "%s.%s.%s.%s " CALCULATED_NUMBER_FORMAT " %u\n", prefix, hostname, st->id, rd->id, value, (uint32_t) before);
77 static inline int format_dimension_collected_opentsdb_telnet(BUFFER *b, const char *prefix, RRDHOST *host, const char *hostname, RRDSET *st, RRDDIM *rd, time_t after, time_t before, uint32_t options) {
82 buffer_sprintf(b, "put %s.%s.%s %u " COLLECTED_NUMBER_FORMAT " host=%s\n", prefix, st->id, rd->id, (uint32_t)rd->last_collected_time.tv_sec, rd->last_collected_value, hostname);
86 static inline int format_dimension_stored_opentsdb_telnet(BUFFER *b, const char *prefix, RRDHOST *host, const char *hostname, RRDSET *st, RRDDIM *rd, time_t after, time_t before, uint32_t options) {
88 calculated_number value = backend_calculate_value_from_stored_data(st, rd, after, before, options);
90 buffer_sprintf(b, "put %s.%s.%s %u " CALCULATED_NUMBER_FORMAT " host=%s\n", prefix, st->id, rd->id, (uint32_t) before, value, hostname);
96 static inline int format_dimension_collected_json_plaintext(BUFFER *b, const char *prefix, RRDHOST *host, const char *hostname, RRDSET *st, RRDDIM *rd, time_t after, time_t before, uint32_t options) {
101 buffer_sprintf(b, "{"
103 "\"hostname\":\"%s\","
105 "\"chart_id\":\"%s\","
106 "\"chart_name\":\"%s\","
108 "\"context\": \"%s\","
114 "\"value\":" COLLECTED_NUMBER_FORMAT ","
116 "\"timestamp\": %u}\n",
129 rd->last_collected_value,
131 (uint32_t)rd->last_collected_time.tv_sec
137 static inline int format_dimension_stored_json_plaintext(BUFFER *b, const char *prefix, RRDHOST *host, const char *hostname, RRDSET *st, RRDDIM *rd, time_t after, time_t before, uint32_t options) {
139 calculated_number value = backend_calculate_value_from_stored_data(st, rd, after, before, options);
141 buffer_sprintf(b, "{"
143 "\"hostname\":\"%s\","
145 "\"chart_id\":\"%s\","
146 "\"chart_name\":\"%s\","
148 "\"context\": \"%s\","
154 "\"value\":" CALCULATED_NUMBER_FORMAT ","
156 "\"timestamp\": %u}\n",
179 static inline int process_graphite_response(BUFFER *b) {
181 const char *s = buffer_tostring(b);
182 char *d = sample, *e = &sample[sizeof(sample) - 1];
184 for(; *s && d < e ;s++) {
186 if(unlikely(!isprint(c))) c = ' ';
191 info("Received %zu bytes from graphite backend. Ignoring them. Sample: '%s'", buffer_strlen(b), sample);
196 static inline int process_json_response(BUFFER *b) {
198 const char *s = buffer_tostring(b);
199 char *d = sample, *e = &sample[sizeof(sample) - 1];
201 for(; *s && d < e ;s++) {
203 if(unlikely(!isprint(c))) c = ' ';
208 info("Received %zu bytes from json backend. Ignoring them. Sample: '%s'", buffer_strlen(b), sample);
213 static inline int process_opentsdb_response(BUFFER *b) {
215 const char *s = buffer_tostring(b);
216 char *d = sample, *e = &sample[sizeof(sample) - 1];
218 for(; *s && d < e ;s++) {
220 if(unlikely(!isprint(c))) c = ' ';
225 info("Received %zu bytes from opentsdb backend. Ignoring them. Sample: '%s'", buffer_strlen(b), sample);
230 void *backends_main(void *ptr) {
231 int default_port = 0;
233 struct netdata_static_thread *static_thread = (struct netdata_static_thread *)ptr;
235 BUFFER *b = buffer_create(1), *response = buffer_create(1);
236 int (*backend_request_formatter)(BUFFER *, const char *, RRDHOST *, const char *, RRDSET *, RRDDIM *, time_t, time_t, uint32_t) = NULL;
237 int (*backend_response_checker)(BUFFER *) = NULL;
239 info("BACKEND thread created with task id %d", gettid());
241 if(pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, NULL) != 0)
242 error("Cannot set pthread cancel type to DEFERRED.");
244 if(pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL) != 0)
245 error("Cannot set pthread cancel state to ENABLE.");
247 // ------------------------------------------------------------------------
248 // collect configuration options
250 struct timeval timeout = {
255 int enabled = config_get_boolean(CONFIG_SECTION_BACKEND, "enabled", 0);
256 const char *source = config_get(CONFIG_SECTION_BACKEND, "data source", "average");
257 const char *type = config_get(CONFIG_SECTION_BACKEND, "type", "graphite");
258 const char *destination = config_get(CONFIG_SECTION_BACKEND, "destination", "localhost");
259 const char *prefix = config_get(CONFIG_SECTION_BACKEND, "prefix", "netdata");
260 const char *hostname = config_get(CONFIG_SECTION_BACKEND, "hostname", localhost->hostname);
261 int frequency = (int)config_get_number(CONFIG_SECTION_BACKEND, "update every", 10);
262 int buffer_on_failures = (int)config_get_number(CONFIG_SECTION_BACKEND, "buffer on failures", 10);
263 long timeoutms = config_get_number(CONFIG_SECTION_BACKEND, "timeout ms", frequency * 2 * 1000);
265 // ------------------------------------------------------------------------
266 // validate configuration options
267 // and prepare for sending data to our backend
269 if(!enabled || frequency < 1)
272 if(!strcmp(source, "as collected")) {
273 options = BACKEND_SOURCE_DATA_AS_COLLECTED;
275 else if(!strcmp(source, "average")) {
276 options = BACKEND_SOURCE_DATA_AVERAGE;
278 else if(!strcmp(source, "sum") || !strcmp(source, "volume")) {
279 options = BACKEND_SOURCE_DATA_SUM;
282 error("Invalid data source method '%s' for backend given. Disabling backed.", source);
287 error("BACKED invalid timeout %ld ms given. Assuming %d ms.", timeoutms, frequency * 2 * 1000);
288 timeoutms = frequency * 2 * 1000;
290 timeout.tv_sec = (timeoutms * 1000) / 1000000;
291 timeout.tv_usec = (timeoutms * 1000) % 1000000;
294 // ------------------------------------------------------------------------
295 // select the backend type
297 if(!strcmp(type, "graphite") || !strcmp(type, "graphite:plaintext")) {
299 if(options == BACKEND_SOURCE_DATA_AS_COLLECTED)
300 backend_request_formatter = format_dimension_collected_graphite_plaintext;
302 backend_request_formatter = format_dimension_stored_graphite_plaintext;
304 backend_response_checker = process_graphite_response;
306 else if(!strcmp(type, "opentsdb") || !strcmp(type, "opentsdb:telnet")) {
308 if(options == BACKEND_SOURCE_DATA_AS_COLLECTED)
309 backend_request_formatter = format_dimension_collected_opentsdb_telnet;
311 backend_request_formatter = format_dimension_stored_opentsdb_telnet;
313 backend_response_checker = process_opentsdb_response;
315 else if (!strcmp(type, "json") || !strcmp(type, "json:plaintext"))
319 if (options == BACKEND_SOURCE_DATA_AS_COLLECTED)
321 backend_request_formatter = format_dimension_collected_json_plaintext;
325 backend_request_formatter = format_dimension_stored_json_plaintext;
328 backend_response_checker = process_json_response;
331 error("Unknown backend type '%s'", type);
335 if(backend_request_formatter == NULL || backend_response_checker == NULL) {
336 error("backend is misconfigured - disabling it.");
341 // ------------------------------------------------------------------------
342 // prepare the charts for monitoring the backend operation
344 struct rusage thread;
347 chart_buffered_metrics = 0,
348 chart_lost_metrics = 0,
349 chart_sent_metrics = 0,
350 chart_buffered_bytes = 0,
351 chart_received_bytes = 0,
352 chart_sent_bytes = 0,
353 chart_receptions = 0,
354 chart_transmission_successes = 0,
355 chart_transmission_failures = 0,
356 chart_data_lost_events = 0,
357 chart_lost_bytes = 0,
358 chart_backend_reconnects = 0,
359 chart_backend_latency = 0;
361 RRDSET *chart_metrics = rrdset_create_localhost("netdata", "backend_metrics", NULL, "backend", NULL, "Netdata Buffered Metrics", "metrics", 130600, frequency, RRDSET_TYPE_LINE);
362 rrddim_add(chart_metrics, "buffered", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
363 rrddim_add(chart_metrics, "lost", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
364 rrddim_add(chart_metrics, "sent", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
366 RRDSET *chart_bytes = rrdset_create_localhost("netdata", "backend_bytes", NULL, "backend", NULL, "Netdata Backend Data Size", "KB", 130610, frequency, RRDSET_TYPE_AREA);
367 rrddim_add(chart_bytes, "buffered", NULL, 1, 1024, RRD_ALGORITHM_ABSOLUTE);
368 rrddim_add(chart_bytes, "lost", NULL, 1, 1024, RRD_ALGORITHM_ABSOLUTE);
369 rrddim_add(chart_bytes, "sent", NULL, 1, 1024, RRD_ALGORITHM_ABSOLUTE);
370 rrddim_add(chart_bytes, "received", NULL, 1, 1024, RRD_ALGORITHM_ABSOLUTE);
372 RRDSET *chart_ops = rrdset_create_localhost("netdata", "backend_ops", NULL, "backend", NULL, "Netdata Backend Operations", "operations", 130630, frequency, RRDSET_TYPE_LINE);
373 rrddim_add(chart_ops, "write", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
374 rrddim_add(chart_ops, "discard", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
375 rrddim_add(chart_ops, "reconnect", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
376 rrddim_add(chart_ops, "failure", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
377 rrddim_add(chart_ops, "read", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE);
380 * this is misleading - we can only measure the time we need to send data
381 * this time is not related to the time required for the data to travel to
382 * the backend database and the time that server needed to process them
384 * issue #1432 and https://www.softlab.ntua.gr/facilities/documentation/unix/unix-socket-faq/unix-socket-faq-2.html
386 RRDSET *chart_latency = rrdset_create_localhost("netdata", "backend_latency", NULL, "backend", NULL, "Netdata Backend Latency", "ms", 130620, frequency, RRDSET_TYPE_AREA);
387 rrddim_add(chart_latency, "latency", NULL, 1, 1000, RRD_ALGORITHM_ABSOLUTE);
390 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);
391 rrddim_add(chart_rusage, "user", NULL, 1, 1000, RRD_ALGORITHM_INCREMENTAL);
392 rrddim_add(chart_rusage, "system", NULL, 1, 1000, RRD_ALGORITHM_INCREMENTAL);
395 // ------------------------------------------------------------------------
396 // prepare the backend main loop
398 info("BACKEND configured ('%s' on '%s' sending '%s' data, every %d seconds, as host '%s', with prefix '%s')", type, destination, source, frequency, hostname, prefix);
400 usec_t step_ut = frequency * USEC_PER_SEC;
401 time_t after = now_realtime_sec();
408 // ------------------------------------------------------------------------
409 // Wait for the next iteration point.
410 heartbeat_next(&hb, step_ut);
411 time_t before = now_realtime_sec();
414 // ------------------------------------------------------------------------
415 // add to the buffer the data we need to send to the backend
417 int pthreadoldcancelstate;
419 if(unlikely(pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &pthreadoldcancelstate) != 0))
420 error("Cannot set pthread cancel state to DISABLE.");
424 rrdhost_foreach_read(host) {
425 if(host->rrd_memory_mode == RRD_MEMORY_MODE_NONE)
428 rrdhost_rdlock(host);
431 rrdset_foreach_read(st, host) {
435 rrddim_foreach_read(rd, st) {
436 if(rd->last_collected_time.tv_sec >= after)
437 chart_buffered_metrics += backend_request_formatter(b, prefix, host, (host == localhost)?hostname:host->hostname, st, rd, after, before, options);
441 rrdhost_unlock(host);
445 if(unlikely(pthread_setcancelstate(pthreadoldcancelstate, NULL) != 0))
446 error("Cannot set pthread cancel state to RESTORE (%d).", pthreadoldcancelstate);
448 // ------------------------------------------------------------------------
450 chart_buffered_bytes = (collected_number)buffer_strlen(b);
452 // reset the monitoring chart counters
453 chart_received_bytes =
457 chart_transmission_successes =
458 chart_transmission_failures =
459 chart_data_lost_events =
461 chart_backend_reconnects =
462 chart_backend_latency = 0;
464 if(unlikely(netdata_exit)) break;
466 //fprintf(stderr, "\nBACKEND BEGIN:\n%s\nBACKEND END\n", buffer_tostring(b)); // FIXME
467 //fprintf(stderr, "after = %lu, before = %lu\n", after, before);
469 // prepare for the next iteration
470 // to add incrementally data to buffer
473 // ------------------------------------------------------------------------
474 // if we are connected, receive a response, without blocking
476 if(likely(sock != -1)) {
479 // loop through to collect all data
480 while(sock != -1 && errno != EWOULDBLOCK) {
481 buffer_need_bytes(response, 4096);
483 ssize_t r = recv(sock, &response->buffer[response->len], response->size - response->len, MSG_DONTWAIT);
485 // we received some data
487 chart_received_bytes += r;
491 error("Backend '%s' closed the socket", destination);
496 // failed to receive data
497 if(errno != EAGAIN && errno != EWOULDBLOCK) {
498 error("Cannot receive data from backend '%s'.", destination);
503 // if we received data, process them
504 if(buffer_strlen(response))
505 backend_response_checker(response);
508 // ------------------------------------------------------------------------
509 // if we are not connected, connect to a backend server
511 if(unlikely(sock == -1)) {
512 usec_t start_ut = now_monotonic_usec();
513 size_t reconnects = 0;
515 sock = connect_to_one_of(destination, default_port, &timeout, &reconnects, NULL, 0);
517 chart_backend_reconnects += reconnects;
518 chart_backend_latency += now_monotonic_usec() - start_ut;
521 if(unlikely(netdata_exit)) break;
523 // ------------------------------------------------------------------------
524 // if we are connected, send our buffer to the backend server
526 if(likely(sock != -1)) {
527 size_t len = buffer_strlen(b);
528 usec_t start_ut = now_monotonic_usec();
531 flags += MSG_NOSIGNAL;
534 ssize_t written = send(sock, buffer_tostring(b), len, flags);
535 chart_backend_latency += now_monotonic_usec() - start_ut;
536 if(written != -1 && (size_t)written == len) {
537 // we sent the data successfully
538 chart_transmission_successes++;
539 chart_sent_bytes += written;
540 chart_sent_metrics = chart_buffered_metrics;
542 // reset the failures count
549 // oops! we couldn't send (all or some of the) data
550 error("Failed to write data to database backend '%s'. Willing to write %zu bytes, wrote %zd bytes. Will re-connect.", destination, len, written);
551 chart_transmission_failures++;
554 chart_sent_bytes += written;
556 // increment the counter we check for data loss
559 // close the socket - we will re-open it next time
565 error("Failed to update database backend '%s'", destination);
566 chart_transmission_failures++;
568 // increment the counter we check for data loss
572 if(failures > buffer_on_failures) {
573 // too bad! we are going to lose data
574 chart_lost_bytes += buffer_strlen(b);
575 error("Reached %d backend failures. Flushing buffers to protect this host - this results in data loss on back-end server '%s'", failures, destination);
578 chart_data_lost_events++;
579 chart_lost_metrics = chart_buffered_metrics;
582 if(unlikely(netdata_exit)) break;
584 // ------------------------------------------------------------------------
585 // update the monitoring charts
587 if(likely(chart_ops->counter_done)) rrdset_next(chart_ops);
588 rrddim_set(chart_ops, "read", chart_receptions);
589 rrddim_set(chart_ops, "write", chart_transmission_successes);
590 rrddim_set(chart_ops, "discard", chart_data_lost_events);
591 rrddim_set(chart_ops, "failure", chart_transmission_failures);
592 rrddim_set(chart_ops, "reconnect", chart_backend_reconnects);
593 rrdset_done(chart_ops);
595 if(likely(chart_metrics->counter_done)) rrdset_next(chart_metrics);
596 rrddim_set(chart_metrics, "buffered", chart_buffered_metrics);
597 rrddim_set(chart_metrics, "lost", chart_lost_metrics);
598 rrddim_set(chart_metrics, "sent", chart_sent_metrics);
599 rrdset_done(chart_metrics);
601 if(likely(chart_bytes->counter_done)) rrdset_next(chart_bytes);
602 rrddim_set(chart_bytes, "buffered", chart_buffered_bytes);
603 rrddim_set(chart_bytes, "lost", chart_lost_bytes);
604 rrddim_set(chart_bytes, "sent", chart_sent_bytes);
605 rrddim_set(chart_bytes, "received", chart_received_bytes);
606 rrdset_done(chart_bytes);
609 if(likely(chart_latency->counter_done)) rrdset_next(chart_latency);
610 rrddim_set(chart_latency, "latency", chart_backend_latency);
611 rrdset_done(chart_latency);
614 getrusage(RUSAGE_THREAD, &thread);
615 if(likely(chart_rusage->counter_done)) rrdset_next(chart_rusage);
616 rrddim_set(chart_rusage, "user", thread.ru_utime.tv_sec * 1000000ULL + thread.ru_utime.tv_usec);
617 rrddim_set(chart_rusage, "system", thread.ru_stime.tv_sec * 1000000ULL + thread.ru_stime.tv_usec);
618 rrdset_done(chart_rusage);
620 if(likely(buffer_strlen(b) == 0))
621 chart_buffered_metrics = 0;
623 if(unlikely(netdata_exit)) break;
631 buffer_free(response);
633 info("BACKEND thread exiting");
635 static_thread->enabled = 0;