src/backends.c

   1 #include "common.h"
   2
   3 #define BACKEND_SOURCE_DATA_AS_COLLECTED 0x00000001
   4 #define BACKEND_SOURCE_DATA_AVERAGE      0x00000002
   5 #define BACKEND_SOURCE_DATA_SUM          0x00000004
   6
   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);
  10
  11     if(unlikely(before < first_t || after > last_t))
  12         // the chart has not been updated in the wanted timeframe
  13         return NAN;
  14
  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;
  19
  20     if(unlikely(after < first_t))
  21         after = first_t;
  22
  23     if(unlikely(after > before))
  24         // this can happen when the st->update_every > before - after
  25         before = after;
  26
  27     if(unlikely(before > last_t))
  28         before = last_t;
  29
  30     size_t counter = 0;
  31     calculated_number sum = 0;
  32
  33     long    start_at_slot = rrdset_time2slot(st, before),
  34             stop_at_slot  = rrdset_time2slot(st, after),
  35             slot, stop_now = 0;
  36
  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;
  40
  41         storage_number n = rd->values[slot];
  42         if(unlikely(!does_storage_number_exist(n))) continue;
  43
  44         calculated_number value = unpack_storage_number(n);
  45         sum += value;
  46         counter++;
  47     }
  48
  49     if(unlikely(!counter))
  50         return NAN;
  51
  52     if(unlikely(options & BACKEND_SOURCE_DATA_SUM))
  53         return sum;
  54
  55     return sum / (calculated_number)counter;
  56 }
  57
  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) {
  59     (void)host;
  60     (void)after;
  61     (void)before;
  62     (void)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);
  64     return 1;
  65 }
  66
  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) {
  68     (void)host;
  69     calculated_number value = backend_calculate_value_from_stored_data(st, rd, after, before, options);
  70     if(!isnan(value)) {
  71         buffer_sprintf(b, "%s.%s.%s.%s " CALCULATED_NUMBER_FORMAT " %u\n", prefix, hostname, st->id, rd->id, value, (uint32_t) before);
  72         return 1;
  73     }
  74     return 0;
  75 }
  76
  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) {
  78     (void)host;
  79     (void)after;
  80     (void)before;
  81     (void)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);
  83     return 1;
  84 }
  85
  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) {
  87     (void)host;
  88     calculated_number value = backend_calculate_value_from_stored_data(st, rd, after, before, options);
  89     if(!isnan(value)) {
  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);
  91         return 1;
  92     }
  93     return 0;
  94 }
  95
  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) {
  97     (void)host;
  98     (void)after;
  99     (void)before;
 100     (void)options;
 101     buffer_sprintf(b, "{"
 102         "\"prefix\":\"%s\","
 103         "\"hostname\":\"%s\","
 104
 105         "\"chart_id\":\"%s\","
 106         "\"chart_name\":\"%s\","
 107         "\"family\":\"%s\","
 108         "\"context\": \"%s\","
 109         "\"type\":\"%s\","
 110         "\"units\": \"%s\","
 111
 112         "\"id\":\"%s\","
 113         "\"name\":\"%s\","
 114         "\"value\":" COLLECTED_NUMBER_FORMAT ","
 115
 116         "\"timestamp\": %u}\n",
 117             prefix,
 118             hostname,
 119
 120             st->id,
 121             st->name,
 122             st->family,
 123             st->context,
 124             st->type,
 125             st->units,
 126
 127             rd->id,
 128             rd->name,
 129             rd->last_collected_value,
 130
 131             (uint32_t)rd->last_collected_time.tv_sec
 132     );
 133
 134     return 1;
 135 }
 136
 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) {
 138     (void)host;
 139     calculated_number value = backend_calculate_value_from_stored_data(st, rd, after, before, options);
 140     if(!isnan(value)) {
 141         buffer_sprintf(b, "{"
 142             "\"prefix\":\"%s\","
 143             "\"hostname\":\"%s\","
 144
 145             "\"chart_id\":\"%s\","
 146             "\"chart_name\":\"%s\","
 147             "\"family\":\"%s\","
 148             "\"context\": \"%s\","
 149             "\"type\":\"%s\","
 150             "\"units\": \"%s\","
 151
 152             "\"id\":\"%s\","
 153             "\"name\":\"%s\","
 154             "\"value\":" CALCULATED_NUMBER_FORMAT ","
 155
 156             "\"timestamp\": %u}\n",
 157                 prefix,
 158                 hostname,
 159
 160                 st->id,
 161                 st->name,
 162                 st->family,
 163                 st->context,
 164                 st->type,
 165                 st->units,
 166
 167                 rd->id,
 168                 rd->name,
 169                 value,
 170
 171                 (uint32_t)before
 172         );
 173
 174         return 1;
 175     }
 176     return 0;
 177 }
 178
 179 static inline int process_graphite_response(BUFFER *b) {
 180     char sample[1024];
 181     const char *s = buffer_tostring(b);
 182     char *d = sample, *e = &sample[sizeof(sample) - 1];
 183
 184     for(; *s && d < e ;s++) {
 185         char c = *s;
 186         if(unlikely(!isprint(c))) c = ' ';
 187         *d++ = c;
 188     }
 189     *d = '\0';
 190
 191     info("Received %zu bytes from graphite backend. Ignoring them. Sample: '%s'", buffer_strlen(b), sample);
 192     buffer_flush(b);
 193     return 0;
 194 }
 195
 196 static inline int process_json_response(BUFFER *b) {
 197     char sample[1024];
 198     const char *s = buffer_tostring(b);
 199     char *d = sample, *e = &sample[sizeof(sample) - 1];
 200
 201     for(; *s && d < e ;s++) {
 202         char c = *s;
 203         if(unlikely(!isprint(c))) c = ' ';
 204         *d++ = c;
 205     }
 206     *d = '\0';
 207
 208     info("Received %zu bytes from json backend. Ignoring them. Sample: '%s'", buffer_strlen(b), sample);
 209     buffer_flush(b);
 210     return 0;
 211 }
 212
 213 static inline int process_opentsdb_response(BUFFER *b) {
 214     char sample[1024];
 215     const char *s = buffer_tostring(b);
 216     char *d = sample, *e = &sample[sizeof(sample) - 1];
 217
 218     for(; *s && d < e ;s++) {
 219         char c = *s;
 220         if(unlikely(!isprint(c))) c = ' ';
 221         *d++ = c;
 222     }
 223     *d = '\0';
 224
 225     info("Received %zu bytes from opentsdb backend. Ignoring them. Sample: '%s'", buffer_strlen(b), sample);
 226     buffer_flush(b);
 227     return 0;
 228 }
 229
 230 void *backends_main(void *ptr) {
 231     struct netdata_static_thread *static_thread = (struct netdata_static_thread *)ptr;
 232
 233     BUFFER *b = buffer_create(1), *response = buffer_create(1);
 234     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;
 235     int (*backend_response_checker)(BUFFER *b) = NULL;
 236
 237     info("BACKEND thread created with task id %d", gettid());
 238
 239     if(pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, NULL) != 0)
 240         error("Cannot set pthread cancel type to DEFERRED.");
 241
 242     if(pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL) != 0)
 243         error("Cannot set pthread cancel state to ENABLE.");
 244
 245     // ------------------------------------------------------------------------
 246     // collect configuration options
 247
 248     struct timeval timeout = {
 249             .tv_sec = 0,
 250             .tv_usec = 0
 251     };
 252     int default_port = 0;
 253     int sock = -1;
 254     uint32_t options;
 255     int enabled = config_get_boolean("backend", "enabled", 0);
 256     const char *source = config_get("backend", "data source", "average");
 257     const char *type = config_get("backend", "type", "graphite");
 258     const char *destination = config_get("backend", "destination", "localhost");
 259     const char *prefix = config_get("backend", "prefix", "netdata");
 260     const char *hostname = config_get("backend", "hostname", localhost.hostname);
 261     int frequency = (int)config_get_number("backend", "update every", 10);
 262     int buffer_on_failures = (int)config_get_number("backend", "buffer on failures", 10);
 263     long timeoutms = config_get_number("backend", "timeout ms", frequency * 2 * 1000);
 264
 265     // ------------------------------------------------------------------------
 266     // validate configuration options
 267     // and prepare for sending data to our backend
 268     if(!enabled || frequency < 1)
 269         goto cleanup;
 270
 271     if(!strcmp(source, "as collected")) {
 272         options = BACKEND_SOURCE_DATA_AS_COLLECTED;
 273     }
 274     else if(!strcmp(source, "average")) {
 275         options = BACKEND_SOURCE_DATA_AVERAGE;
 276     }
 277     else if(!strcmp(source, "sum") || !strcmp(source, "volume")) {
 278         options = BACKEND_SOURCE_DATA_SUM;
 279     }
 280     else {
 281         error("Invalid data source method '%s' for backend given. Disabling backed.", source);
 282         goto cleanup;
 283     }
 284
 285     if(!strcmp(type, "graphite") || !strcmp(type, "graphite:plaintext")) {
 286         default_port = 2003;
 287         if(options == BACKEND_SOURCE_DATA_AS_COLLECTED)
 288             backend_request_formatter = format_dimension_collected_graphite_plaintext;
 289         else
 290             backend_request_formatter = format_dimension_stored_graphite_plaintext;
 291
 292         backend_response_checker = process_graphite_response;
 293     }
 294     else if(!strcmp(type, "opentsdb") || !strcmp(type, "opentsdb:telnet")) {
 295         default_port = 4242;
 296         if(options == BACKEND_SOURCE_DATA_AS_COLLECTED)
 297             backend_request_formatter = format_dimension_collected_opentsdb_telnet;
 298         else
 299             backend_request_formatter = format_dimension_stored_opentsdb_telnet;
 300
 301         backend_response_checker = process_opentsdb_response;
 302     }
 303     else if (!strcmp(type, "json") || !strcmp(type, "json:plaintext"))
 304     {
 305         default_port = 5448;
 306
 307         if (options == BACKEND_SOURCE_DATA_AS_COLLECTED)
 308         {
 309             backend_request_formatter = format_dimension_collected_json_plaintext;
 310         }
 311         else
 312         {
 313             backend_request_formatter = format_dimension_stored_json_plaintext;
 314         }
 315
 316         backend_response_checker = process_json_response;
 317     }
 318     else {
 319         error("Unknown backend type '%s'", type);
 320         goto cleanup;
 321     }
 322
 323     if(backend_request_formatter == NULL || backend_response_checker == NULL) {
 324         error("backend is misconfigured - disabling it.");
 325         goto cleanup;
 326     }
 327
 328     if(timeoutms < 1) {
 329         error("BACKED invalid timeout %ld ms given. Assuming %d ms.", timeoutms, frequency * 2 * 1000);
 330         timeoutms = frequency * 2 * 1000;
 331     }
 332     timeout.tv_sec  = (timeoutms * 1000) / 1000000;
 333     timeout.tv_usec = (timeoutms * 1000) % 1000000;
 334
 335     // ------------------------------------------------------------------------
 336     // prepare the charts for monitoring the backend
 337
 338     struct rusage thread;
 339
 340     collected_number
 341             chart_buffered_metrics = 0,
 342             chart_lost_metrics = 0,
 343             chart_sent_metrics = 0,
 344             chart_buffered_bytes = 0,
 345             chart_received_bytes = 0,
 346             chart_sent_bytes = 0,
 347             chart_receptions = 0,
 348             chart_transmission_successes = 0,
 349             chart_transmission_failures = 0,
 350             chart_data_lost_events = 0,
 351             chart_lost_bytes = 0,
 352             chart_backend_reconnects = 0,
 353             chart_backend_latency = 0;
 354
 355     RRDSET *chart_metrics = rrdset_find("netdata.backend_metrics");
 356     if(!chart_metrics) {
 357         chart_metrics = rrdset_create("netdata", "backend_metrics", NULL, "backend", NULL, "Netdata Buffered Metrics", "metrics", 130600, frequency, RRDSET_TYPE_LINE);
 358         rrddim_add(chart_metrics, "buffered", NULL,  1, 1, RRDDIM_ABSOLUTE);
 359         rrddim_add(chart_metrics, "lost",     NULL,  1, 1, RRDDIM_ABSOLUTE);
 360         rrddim_add(chart_metrics, "sent",     NULL,  1, 1, RRDDIM_ABSOLUTE);
 361     }
 362
 363     RRDSET *chart_bytes = rrdset_find("netdata.backend_bytes");
 364     if(!chart_bytes) {
 365         chart_bytes = rrdset_create("netdata", "backend_bytes", NULL, "backend", NULL, "Netdata Backend Data Size", "KB", 130610, frequency, RRDSET_TYPE_AREA);
 366         rrddim_add(chart_bytes, "buffered", NULL, 1, 1024, RRDDIM_ABSOLUTE);
 367         rrddim_add(chart_bytes, "lost",     NULL, 1, 1024, RRDDIM_ABSOLUTE);
 368         rrddim_add(chart_bytes, "sent",     NULL, 1, 1024, RRDDIM_ABSOLUTE);
 369         rrddim_add(chart_bytes, "received", NULL, 1, 1024, RRDDIM_ABSOLUTE);
 370     }
 371
 372     RRDSET *chart_ops = rrdset_find("netdata.backend_ops");
 373     if(!chart_ops) {
 374         chart_ops = rrdset_create("netdata", "backend_ops", NULL, "backend", NULL, "Netdata Backend Operations", "operations", 130630, frequency, RRDSET_TYPE_LINE);
 375         rrddim_add(chart_ops, "write",     NULL, 1, 1, RRDDIM_ABSOLUTE);
 376         rrddim_add(chart_ops, "discard",   NULL, 1, 1, RRDDIM_ABSOLUTE);
 377         rrddim_add(chart_ops, "reconnect", NULL, 1, 1, RRDDIM_ABSOLUTE);
 378         rrddim_add(chart_ops, "failure",   NULL, 1, 1, RRDDIM_ABSOLUTE);
 379         rrddim_add(chart_ops, "read",      NULL, 1, 1, RRDDIM_ABSOLUTE);
 380     }
 381
 382     /*
 383      * this is misleading - we can only measure the time we need to send data
 384      * this time is not related to the time required for the data to travel to
 385      * the backend database and the time that server needed to process them
 386      *
 387      * issue #1432 and https://www.softlab.ntua.gr/facilities/documentation/unix/unix-socket-faq/unix-socket-faq-2.html
 388      *
 389     RRDSET *chart_latency = rrdset_find("netdata.backend_latency");
 390     if(!chart_latency) {
 391         chart_latency = rrdset_create("netdata", "backend_latency", NULL, "backend", NULL, "Netdata Backend Latency", "ms", 130620, frequency, RRDSET_TYPE_AREA);
 392         rrddim_add(chart_latency, "latency",   NULL,  1, 1000, RRDDIM_ABSOLUTE);
 393     }
 394     */
 395
 396     RRDSET *chart_rusage = rrdset_find("netdata.backend_thread_cpu");
 397     if(!chart_rusage) {
 398         chart_rusage = rrdset_create("netdata", "backend_thread_cpu", NULL, "backend", NULL, "NetData Backend Thread CPU usage", "milliseconds/s", 130630, frequency, RRDSET_TYPE_STACKED);
 399         rrddim_add(chart_rusage, "user",   NULL, 1, 1000, RRDDIM_INCREMENTAL);
 400         rrddim_add(chart_rusage, "system", NULL, 1, 1000, RRDDIM_INCREMENTAL);
 401     }
 402
 403     // ------------------------------------------------------------------------
 404     // prepare the backend main loop
 405
 406     info("BACKEND configured ('%s' on '%s' sending '%s' data, every %d seconds, as host '%s', with prefix '%s')", type, destination, source, frequency, hostname, prefix);
 407
 408     usec_t step_ut = frequency * USEC_PER_SEC;
 409     time_t after = now_realtime_sec();
 410     int failures = 0;
 411     heartbeat_t hb;
 412     heartbeat_init(&hb);
 413
 414     for(;;) {
 415         // ------------------------------------------------------------------------
 416         // Wait for the next iteration point.
 417         heartbeat_next(&hb, step_ut);
 418         time_t before = now_realtime_sec();
 419
 420         // ------------------------------------------------------------------------
 421         // add to the buffer the data we need to send to the backend
 422         RRDSET *st;
 423         int pthreadoldcancelstate;
 424
 425         if(unlikely(pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &pthreadoldcancelstate) != 0))
 426             error("Cannot set pthread cancel state to DISABLE.");
 427
 428         rrdhost_rdlock(&localhost);
 429         for(st = localhost.rrdset_root; st ;st = st->next) {
 430             pthread_rwlock_rdlock(&st->rwlock);
 431
 432             RRDDIM *rd;
 433             for(rd = st->dimensions; rd ;rd = rd->next) {
 434                 if(rd->last_collected_time.tv_sec >= after)
 435                     chart_buffered_metrics += backend_request_formatter(b, prefix, &localhost, hostname, st, rd, after, before, options);
 436             }
 437
 438             pthread_rwlock_unlock(&st->rwlock);
 439         }
 440         rrdhost_unlock(&localhost);
 441
 442         if(unlikely(pthread_setcancelstate(pthreadoldcancelstate, NULL) != 0))
 443             error("Cannot set pthread cancel state to RESTORE (%d).", pthreadoldcancelstate);
 444
 445         // ------------------------------------------------------------------------
 446
 447         chart_buffered_bytes = (collected_number)buffer_strlen(b);
 448
 449         // reset the monitoring chart counters
 450         chart_received_bytes =
 451         chart_sent_bytes =
 452         chart_sent_metrics =
 453         chart_lost_metrics =
 454         chart_transmission_successes =
 455         chart_transmission_failures =
 456         chart_data_lost_events =
 457         chart_lost_bytes =
 458         chart_backend_reconnects =
 459         chart_backend_latency = 0;
 460
 461         if(unlikely(netdata_exit)) break;
 462
 463         //fprintf(stderr, "\nBACKEND BEGIN:\n%s\nBACKEND END\n", buffer_tostring(b)); // FIXME
 464         //fprintf(stderr, "after = %lu, before = %lu\n", after, before);
 465
 466         // prepare for the next iteration
 467         // to add incrementally data to buffer
 468         after = before;
 469
 470         // ------------------------------------------------------------------------
 471         // if we are connected, receive a response, without blocking
 472
 473         if(likely(sock != -1)) {
 474             errno = 0;
 475
 476             // loop through to collect all data
 477             while(sock != -1 && errno != EWOULDBLOCK) {
 478                 buffer_need_bytes(response, 4096);
 479
 480                 ssize_t r = recv(sock, &response->buffer[response->len], response->size - response->len, MSG_DONTWAIT);
 481                 if(likely(r > 0)) {
 482                     // we received some data
 483                     response->len += r;
 484                     chart_received_bytes += r;
 485                     chart_receptions++;
 486                 }
 487                 else if(r == 0) {
 488                     error("Backend '%s' closed the socket", destination);
 489                     close(sock);
 490                     sock = -1;
 491                 }
 492                 else {
 493                     // failed to receive data
 494                     if(errno != EAGAIN && errno != EWOULDBLOCK) {
 495                         error("Cannot receive data from backend '%s'.", destination);
 496                     }
 497                 }
 498             }
 499
 500             // if we received data, process them
 501             if(buffer_strlen(response))
 502                 backend_response_checker(response);
 503         }
 504
 505         // ------------------------------------------------------------------------
 506         // if we are not connected, connect to a backend server
 507
 508         if(unlikely(sock == -1)) {
 509             usec_t start_ut = now_monotonic_usec();
 510             const char *s = destination;
 511             while(*s) {
 512                 const char *e = s;
 513
 514                 // skip separators, moving both s(tart) and e(nd)
 515                 while(isspace(*e) || *e == ',') s = ++e;
 516
 517                 // move e(nd) to the first separator
 518                 while(*e && !isspace(*e) && *e != ',') e++;
 519
 520                 // is there anything?
 521                 if(!*s || s == e) break;
 522
 523                 char buf[e - s + 1];
 524                 strncpyz(buf, s, e - s);
 525                 chart_backend_reconnects++;
 526                 sock = connect_to(buf, default_port, &timeout);
 527                 if(sock != -1) break;
 528                 s = e;
 529             }
 530             chart_backend_latency += now_monotonic_usec() - start_ut;
 531         }
 532
 533         if(unlikely(netdata_exit)) break;
 534
 535         // ------------------------------------------------------------------------
 536         // if we are connected, send our buffer to the backend server
 537
 538         if(likely(sock != -1)) {
 539             size_t len = buffer_strlen(b);
 540             usec_t start_ut = now_monotonic_usec();
 541             int flags = 0;
 542 #ifdef MSG_NOSIGNAL
 543             flags += MSG_NOSIGNAL;
 544 #endif
 545
 546             ssize_t written = send(sock, buffer_tostring(b), len, flags);
 547             chart_backend_latency += now_monotonic_usec() - start_ut;
 548             if(written != -1 && (size_t)written == len) {
 549                 // we sent the data successfully
 550                 chart_transmission_successes++;
 551                 chart_sent_bytes += written;
 552                 chart_sent_metrics = chart_buffered_metrics;
 553
 554                 // reset the failures count
 555                 failures = 0;
 556
 557                 // empty the buffer
 558                 buffer_flush(b);
 559             }
 560             else {
 561                 // oops! we couldn't send (all or some of the) data
 562                 error("Failed to write data to database backend '%s'. Willing to write %zu bytes, wrote %zd bytes. Will re-connect.", destination, len, written);
 563                 chart_transmission_failures++;
 564
 565                 if(written != -1)
 566                     chart_sent_bytes += written;
 567
 568                 // increment the counter we check for data loss
 569                 failures++;
 570
 571                 // close the socket - we will re-open it next time
 572                 close(sock);
 573                 sock = -1;
 574             }
 575         }
 576         else {
 577             error("Failed to update database backend '%s'", destination);
 578             chart_transmission_failures++;
 579
 580             // increment the counter we check for data loss
 581             failures++;
 582         }
 583
 584         if(failures > buffer_on_failures) {
 585             // too bad! we are going to lose data
 586             chart_lost_bytes += buffer_strlen(b);
 587             error("Reached %d backend failures. Flushing buffers to protect this host - this results in data loss on back-end server '%s'", failures, destination);
 588             buffer_flush(b);
 589             failures = 0;
 590             chart_data_lost_events++;
 591             chart_lost_metrics = chart_buffered_metrics;
 592         }
 593
 594         if(unlikely(netdata_exit)) break;
 595
 596         // ------------------------------------------------------------------------
 597         // update the monitoring charts
 598
 599         if(chart_ops->counter_done) rrdset_next(chart_ops);
 600         rrddim_set(chart_ops, "read",         chart_receptions);
 601         rrddim_set(chart_ops, "write",        chart_transmission_successes);
 602         rrddim_set(chart_ops, "discard",      chart_data_lost_events);
 603         rrddim_set(chart_ops, "failure",      chart_transmission_failures);
 604         rrddim_set(chart_ops, "reconnect",    chart_backend_reconnects);
 605         rrdset_done(chart_ops);
 606
 607         if(chart_metrics->counter_done) rrdset_next(chart_metrics);
 608         rrddim_set(chart_metrics, "buffered", chart_buffered_metrics);
 609         rrddim_set(chart_metrics, "lost",     chart_lost_metrics);
 610         rrddim_set(chart_metrics, "sent",     chart_sent_metrics);
 611         rrdset_done(chart_metrics);
 612
 613         if(chart_bytes->counter_done) rrdset_next(chart_bytes);
 614         rrddim_set(chart_bytes, "buffered",   chart_buffered_bytes);
 615         rrddim_set(chart_bytes, "lost",       chart_lost_bytes);
 616         rrddim_set(chart_bytes, "sent",       chart_sent_bytes);
 617         rrddim_set(chart_bytes, "received",   chart_received_bytes);
 618         rrdset_done(chart_bytes);
 619
 620         /*
 621         if(chart_latency->counter_done) rrdset_next(chart_latency);
 622         rrddim_set(chart_latency, "latency",  chart_backend_latency);
 623         rrdset_done(chart_latency);
 624         */
 625
 626         getrusage(RUSAGE_THREAD, &thread);
 627         if(chart_rusage->counter_done) rrdset_next(chart_rusage);
 628         rrddim_set(chart_rusage, "user",   thread.ru_utime.tv_sec * 1000000ULL + thread.ru_utime.tv_usec);
 629         rrddim_set(chart_rusage, "system", thread.ru_stime.tv_sec * 1000000ULL + thread.ru_stime.tv_usec);
 630         rrdset_done(chart_rusage);
 631
 632         if(likely(buffer_strlen(b) == 0))
 633             chart_buffered_metrics = 0;
 634
 635         if(unlikely(netdata_exit)) break;
 636     }
 637
 638 cleanup:
 639     if(sock != -1)
 640         close(sock);
 641
 642     buffer_free(b);
 643     buffer_free(response);
 644
 645     info("BACKEND thread exiting");
 646
 647     static_thread->enabled = 0;
 648     pthread_exit(NULL);
 649     return NULL;
 650 }