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 - after < st->update_every && after != after - after % st->update_every))
12 // when st->update_every is bigger than the frequency we send data to backend
13 // skip the iterations that are not aligned to the database
16 // align the time-frame
17 // for 'after' also skip the first value by adding st->update_every
18 after = after - after % st->update_every + st->update_every;
19 before = before - before % st->update_every;
21 if(unlikely(after < first_t))
24 if(unlikely(after > before))
25 // this can happen when the st->update_every > before - after
28 if(unlikely(before > last_t))
32 calculated_number sum = 0;
34 long start_at_slot = rrdset_time2slot(st, before),
35 stop_at_slot = rrdset_time2slot(st, after),
38 for(slot = start_at_slot; !stop_now ; slot--) {
39 if(unlikely(slot < 0)) slot = st->entries - 1;
40 if(unlikely(slot == stop_at_slot)) stop_now = 1;
42 storage_number n = rd->values[slot];
43 if(unlikely(!does_storage_number_exist(n))) continue;
45 calculated_number value = unpack_storage_number(n);
50 if(unlikely(!counter))
53 if(unlikely(options & BACKEND_SOURCE_DATA_SUM))
56 return sum / (calculated_number)counter;
59 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) {
64 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);
68 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) {
70 calculated_number value = backend_calculate_value_from_stored_data(st, rd, after, before, options);
72 buffer_sprintf(b, "%s.%s.%s.%s " CALCULATED_NUMBER_FORMAT " %u\n", prefix, hostname, st->id, rd->id, value, (uint32_t) before);
78 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) {
83 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);
87 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) {
89 calculated_number value = backend_calculate_value_from_stored_data(st, rd, after, before, options);
91 buffer_sprintf(b, "put %s.%s.%s %u " CALCULATED_NUMBER_FORMAT " host=%s\n", prefix, st->id, rd->id, (uint32_t) before, value, hostname);
97 static inline int process_graphite_response(BUFFER *b) {
99 const char *s = buffer_tostring(b);
100 char *d = sample, *e = &sample[sizeof(sample) - 1];
102 for(; *s && d < e ;s++) {
104 if(unlikely(!isprint(c))) c = ' ';
109 info("Received %zu bytes from graphite backend. Ignoring them. Sample: '%s'", buffer_strlen(b), sample);
114 static inline int process_opentsdb_response(BUFFER *b) {
116 const char *s = buffer_tostring(b);
117 char *d = sample, *e = &sample[sizeof(sample) - 1];
119 for(; *s && d < e ;s++) {
121 if(unlikely(!isprint(c))) c = ' ';
126 info("Received %zu bytes from opentsdb backend. Ignoring them. Sample: '%s'", buffer_strlen(b), sample);
131 void *backends_main(void *ptr) {
132 struct netdata_static_thread *static_thread = (struct netdata_static_thread *)ptr;
134 BUFFER *b = buffer_create(1), *response = buffer_create(1);
135 int (*backend_request_formatter)(BUFFER *b, const char *prefix, RRDHOST *host, const char *hostname, RRDSET *st, RRDDIM *rd, time_t after, time_t before, uint32_t options) = NULL;
136 int (*backend_response_checker)(BUFFER *b) = NULL;
138 info("BACKEND thread created with task id %d", gettid());
140 if(pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, NULL) != 0)
141 error("Cannot set pthread cancel type to DEFERRED.");
143 if(pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL) != 0)
144 error("Cannot set pthread cancel state to ENABLE.");
146 // ------------------------------------------------------------------------
147 // collect configuration options
149 struct timeval timeout = {
153 int default_port = 0;
156 int enabled = config_get_boolean("backend", "enabled", 0);
157 const char *source = config_get("backend", "data source", "average");
158 const char *type = config_get("backend", "type", "graphite");
159 const char *destination = config_get("backend", "destination", "localhost");
160 const char *prefix = config_get("backend", "prefix", "netdata");
161 const char *hostname = config_get("backend", "hostname", localhost.hostname);
162 int frequency = (int)config_get_number("backend", "update every", 10);
163 int buffer_on_failures = (int)config_get_number("backend", "buffer on failures", 10);
164 long timeoutms = config_get_number("backend", "timeout ms", frequency * 2 * 1000);
166 // ------------------------------------------------------------------------
167 // validate configuration options
168 // and prepare for sending data to our backend
169 if(!enabled || frequency < 1)
172 if(!strcmp(source, "as collected")) {
173 options = BACKEND_SOURCE_DATA_AS_COLLECTED;
175 else if(!strcmp(source, "average")) {
176 options = BACKEND_SOURCE_DATA_AVERAGE;
178 else if(!strcmp(source, "sum") || !strcmp(source, "volume")) {
179 options = BACKEND_SOURCE_DATA_SUM;
182 error("Invalid data source method '%s' for backend given. Disabling backed.", source);
186 if(!strcmp(type, "graphite") || !strcmp(type, "graphite:plaintext")) {
188 if(options == BACKEND_SOURCE_DATA_AS_COLLECTED)
189 backend_request_formatter = format_dimension_collected_graphite_plaintext;
191 backend_request_formatter = format_dimension_stored_graphite_plaintext;
193 backend_response_checker = process_graphite_response;
195 else if(!strcmp(type, "opentsdb") || !strcmp(type, "opentsdb:telnet")) {
197 if(options == BACKEND_SOURCE_DATA_AS_COLLECTED)
198 backend_request_formatter = format_dimension_collected_opentsdb_telnet;
200 backend_request_formatter = format_dimension_stored_opentsdb_telnet;
202 backend_response_checker = process_opentsdb_response;
205 error("Unknown backend type '%s'", type);
209 if(backend_request_formatter == NULL || backend_response_checker == NULL) {
210 error("backend is misconfigured - disabling it.");
215 error("BACKED invalid timeout %ld ms given. Assuming %d ms.", timeoutms, frequency * 2 * 1000);
216 timeoutms = frequency * 2 * 1000;
218 timeout.tv_sec = (timeoutms * 1000) / 1000000;
219 timeout.tv_usec = (timeoutms * 1000) % 1000000;
221 // ------------------------------------------------------------------------
222 // prepare the charts for monitoring the backend
224 struct rusage thread;
227 chart_buffered_metrics = 0,
228 chart_lost_metrics = 0,
229 chart_sent_metrics = 0,
230 chart_buffered_bytes = 0,
231 chart_received_bytes = 0,
232 chart_sent_bytes = 0,
233 chart_receptions = 0,
234 chart_transmission_successes = 0,
235 chart_transmission_failures = 0,
236 chart_data_lost_events = 0,
237 chart_lost_bytes = 0,
238 chart_backend_reconnects = 0,
239 chart_backend_latency = 0;
241 RRDSET *chart_metrics = rrdset_find("netdata.backend_metrics");
243 chart_metrics = rrdset_create("netdata", "backend_metrics", NULL, "backend", NULL, "Netdata Buffered Metrics", "metrics", 130600, frequency, RRDSET_TYPE_LINE);
244 rrddim_add(chart_metrics, "buffered", NULL, 1, 1, RRDDIM_ABSOLUTE);
245 rrddim_add(chart_metrics, "lost", NULL, 1, 1, RRDDIM_ABSOLUTE);
246 rrddim_add(chart_metrics, "sent", NULL, 1, 1, RRDDIM_ABSOLUTE);
249 RRDSET *chart_bytes = rrdset_find("netdata.backend_bytes");
251 chart_bytes = rrdset_create("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, RRDDIM_ABSOLUTE);
253 rrddim_add(chart_bytes, "lost", NULL, 1, 1024, RRDDIM_ABSOLUTE);
254 rrddim_add(chart_bytes, "sent", NULL, 1, 1024, RRDDIM_ABSOLUTE);
255 rrddim_add(chart_bytes, "received", NULL, 1, 1024, RRDDIM_ABSOLUTE);
258 RRDSET *chart_ops = rrdset_find("netdata.backend_ops");
260 chart_ops = rrdset_create("netdata", "backend_ops", NULL, "backend", NULL, "Netdata Backend Operations", "operations", 130630, frequency, RRDSET_TYPE_LINE);
261 rrddim_add(chart_ops, "write", NULL, 1, 1, RRDDIM_ABSOLUTE);
262 rrddim_add(chart_ops, "discard", NULL, 1, 1, RRDDIM_ABSOLUTE);
263 rrddim_add(chart_ops, "reconnect", NULL, 1, 1, RRDDIM_ABSOLUTE);
264 rrddim_add(chart_ops, "failure", NULL, 1, 1, RRDDIM_ABSOLUTE);
265 rrddim_add(chart_ops, "read", NULL, 1, 1, RRDDIM_ABSOLUTE);
269 * this is misleading - we can only measure the time we need to send data
270 * this time is not related to the time required for the data to travel to
271 * the backend database and the time that server needed to process them
273 * issue #1432 and https://www.softlab.ntua.gr/facilities/documentation/unix/unix-socket-faq/unix-socket-faq-2.html
275 RRDSET *chart_latency = rrdset_find("netdata.backend_latency");
277 chart_latency = rrdset_create("netdata", "backend_latency", NULL, "backend", NULL, "Netdata Backend Latency", "ms", 130620, frequency, RRDSET_TYPE_AREA);
278 rrddim_add(chart_latency, "latency", NULL, 1, 1000, RRDDIM_ABSOLUTE);
282 RRDSET *chart_rusage = rrdset_find("netdata.backend_thread_cpu");
284 chart_rusage = rrdset_create("netdata", "backend_thread_cpu", NULL, "backend", NULL, "NetData Backend Thread CPU usage", "milliseconds/s", 130630, frequency, RRDSET_TYPE_STACKED);
285 rrddim_add(chart_rusage, "user", NULL, 1, 1000, RRDDIM_INCREMENTAL);
286 rrddim_add(chart_rusage, "system", NULL, 1, 1000, RRDDIM_INCREMENTAL);
289 // ------------------------------------------------------------------------
290 // prepare the backend main loop
292 info("BACKEND configured ('%s' on '%s' sending '%s' data, every %d seconds, as host '%s', with prefix '%s')", type, destination, source, frequency, hostname, prefix);
294 usec_t step_ut = frequency * USEC_PER_SEC;
295 usec_t random_ut = now_realtime_usec() % (step_ut / 2);
296 time_t before = (time_t)((now_realtime_usec() - step_ut) / USEC_PER_SEC);
297 time_t after = before;
301 // ------------------------------------------------------------------------
302 // wait for the next iteration point
304 usec_t now_ut = now_realtime_usec();
305 usec_t next_ut = now_ut - (now_ut % step_ut) + step_ut;
306 before = (time_t)(next_ut / USEC_PER_SEC);
308 // add a little delay (1/4 of the step) plus some randomness
309 next_ut += (step_ut / 4) + random_ut;
311 while(now_ut < next_ut) {
312 sleep_usec(next_ut - now_ut);
313 now_ut = now_realtime_usec();
316 // ------------------------------------------------------------------------
317 // add to the buffer the data we need to send to the backend
320 int pthreadoldcancelstate;
322 if(unlikely(pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &pthreadoldcancelstate) != 0))
323 error("Cannot set pthread cancel state to DISABLE.");
325 rrdhost_rdlock(&localhost);
326 for(st = localhost.rrdset_root; st ;st = st->next) {
327 pthread_rwlock_rdlock(&st->rwlock);
330 for(rd = st->dimensions; rd ;rd = rd->next) {
331 if(rd->last_collected_time.tv_sec >= after)
332 chart_buffered_metrics += backend_request_formatter(b, prefix, &localhost, hostname, st, rd, after, before, options);
335 pthread_rwlock_unlock(&st->rwlock);
337 rrdhost_unlock(&localhost);
339 if(unlikely(pthread_setcancelstate(pthreadoldcancelstate, NULL) != 0))
340 error("Cannot set pthread cancel state to RESTORE (%d).", pthreadoldcancelstate);
342 // ------------------------------------------------------------------------
344 chart_buffered_bytes = (collected_number)buffer_strlen(b);
346 // reset the monitoring chart counters
347 chart_received_bytes =
351 chart_transmission_successes =
352 chart_transmission_failures =
353 chart_data_lost_events =
355 chart_backend_reconnects =
356 chart_backend_latency = 0;
358 if(unlikely(netdata_exit)) break;
360 //fprintf(stderr, "\nBACKEND BEGIN:\n%s\nBACKEND END\n", buffer_tostring(b)); // FIXME
361 //fprintf(stderr, "after = %lu, before = %lu\n", after, before);
363 // ------------------------------------------------------------------------
364 // if we are connected, receive a response, without blocking
366 if(likely(sock != -1)) {
369 // loop through to collect all data
370 while(sock != -1 && errno != EWOULDBLOCK) {
371 buffer_need_bytes(response, 4096);
373 ssize_t r = recv(sock, &response->buffer[response->len], response->size - response->len, MSG_DONTWAIT);
375 // we received some data
377 chart_received_bytes += r;
381 error("Backend '%s' closed the socket", destination);
386 // failed to receive data
387 if(errno != EAGAIN && errno != EWOULDBLOCK) {
388 error("Cannot receive data from backend '%s'.", destination);
393 // if we received data, process them
394 if(buffer_strlen(response))
395 backend_response_checker(response);
398 // ------------------------------------------------------------------------
399 // if we are not connected, connect to a backend server
401 if(unlikely(sock == -1)) {
402 usec_t start_ut = now_realtime_usec();
403 const char *s = destination;
407 // skip separators, moving both s(tart) and e(nd)
408 while(isspace(*e) || *e == ',') s = ++e;
410 // move e(nd) to the first separator
411 while(*e && !isspace(*e) && *e != ',') e++;
413 // is there anything?
414 if(!*s || s == e) break;
417 strncpyz(buf, s, e - s);
418 chart_backend_reconnects++;
419 sock = connect_to(buf, default_port, &timeout);
420 if(sock != -1) break;
423 chart_backend_latency += now_realtime_usec() - start_ut;
426 if(unlikely(netdata_exit)) break;
428 // ------------------------------------------------------------------------
429 // if we are connected, send our buffer to the backend server
431 if(likely(sock != -1)) {
432 size_t len = buffer_strlen(b);
433 usec_t start_ut = now_realtime_usec();
436 flags += MSG_NOSIGNAL;
439 ssize_t written = send(sock, buffer_tostring(b), len, flags);
440 chart_backend_latency += now_realtime_usec() - start_ut;
441 if(written != -1 && (size_t)written == len) {
442 // we sent the data successfully
443 chart_transmission_successes++;
444 chart_sent_bytes += written;
445 chart_sent_metrics = chart_buffered_metrics;
447 // reset the failures count
454 // oops! we couldn't send (all or some of the) data
455 error("Failed to write data to database backend '%s'. Willing to write %zu bytes, wrote %zd bytes. Will re-connect.", destination, len, written);
456 chart_transmission_failures++;
459 chart_sent_bytes += written;
461 // increment the counter we check for data loss
464 // close the socket - we will re-open it next time
469 // either the buffer is empty
470 // or is holding the data we couldn't send
471 // so, make sure the next iteration will continue
472 // from where we are now
476 error("Failed to update database backend '%s'", destination);
477 chart_transmission_failures++;
479 // increment the counter we check for data loss
483 if(failures > buffer_on_failures) {
484 // too bad! we are going to lose data
485 chart_lost_bytes += buffer_strlen(b);
486 error("Reached %d backend failures. Flushing buffers to protect this host - this results in data loss on back-end server '%s'", failures, destination);
489 chart_data_lost_events++;
490 chart_lost_metrics = chart_buffered_metrics;
493 if(unlikely(netdata_exit)) break;
495 // ------------------------------------------------------------------------
496 // update the monitoring charts
498 if(chart_ops->counter_done) rrdset_next(chart_ops);
499 rrddim_set(chart_ops, "read", chart_receptions);
500 rrddim_set(chart_ops, "write", chart_transmission_successes);
501 rrddim_set(chart_ops, "discard", chart_data_lost_events);
502 rrddim_set(chart_ops, "failure", chart_transmission_failures);
503 rrddim_set(chart_ops, "reconnect", chart_backend_reconnects);
504 rrdset_done(chart_ops);
506 if(chart_metrics->counter_done) rrdset_next(chart_metrics);
507 rrddim_set(chart_metrics, "buffered", chart_buffered_metrics);
508 rrddim_set(chart_metrics, "lost", chart_lost_metrics);
509 rrddim_set(chart_metrics, "sent", chart_sent_metrics);
510 rrdset_done(chart_metrics);
512 if(chart_bytes->counter_done) rrdset_next(chart_bytes);
513 rrddim_set(chart_bytes, "buffered", chart_buffered_bytes);
514 rrddim_set(chart_bytes, "lost", chart_lost_bytes);
515 rrddim_set(chart_bytes, "sent", chart_sent_bytes);
516 rrddim_set(chart_bytes, "received", chart_received_bytes);
517 rrdset_done(chart_bytes);
520 if(chart_latency->counter_done) rrdset_next(chart_latency);
521 rrddim_set(chart_latency, "latency", chart_backend_latency);
522 rrdset_done(chart_latency);
525 getrusage(RUSAGE_THREAD, &thread);
526 if(chart_rusage->counter_done) rrdset_next(chart_rusage);
527 rrddim_set(chart_rusage, "user", thread.ru_utime.tv_sec * 1000000ULL + thread.ru_utime.tv_usec);
528 rrddim_set(chart_rusage, "system", thread.ru_stime.tv_sec * 1000000ULL + thread.ru_stime.tv_usec);
529 rrdset_done(chart_rusage);
531 if(likely(buffer_strlen(b) == 0))
532 chart_buffered_metrics = 0;
534 if(unlikely(netdata_exit)) break;
542 buffer_free(response);
544 info("BACKEND thread exiting");
546 static_thread->enabled = 0;