src/health.c

   1 #define NETDATA_HEALTH_INTERNALS
   2 #include "common.h"
   3
   4 int default_health_enabled = 1;
   5
   6 // ----------------------------------------------------------------------------
   7 // health initialization
   8
   9 inline char *health_config_dir(void) {
  10     char buffer[FILENAME_MAX + 1];
  11     snprintfz(buffer, FILENAME_MAX, "%s/health.d", netdata_configured_config_dir);
  12     return config_get("health", "health configuration directory", buffer);
  13 }
  14
  15 void health_init(void) {
  16     debug(D_HEALTH, "Health configuration initializing");
  17
  18     if(!central_netdata_to_push_data) {
  19         if(!(default_health_enabled = config_get_boolean("health", "enabled", 1))) {
  20             debug(D_HEALTH, "Health is disabled.");
  21             return;
  22         }
  23     }
  24     else {
  25         info("Health is disabled - setup alarms at the central netdata.");
  26         config_set_boolean("health", "enabled", 0);
  27         default_health_enabled = 0;
  28     }
  29
  30     char pathname[FILENAME_MAX + 1];
  31     snprintfz(pathname, FILENAME_MAX, "%s/health", netdata_configured_varlib_dir);
  32     if(mkdir(pathname, 0770) == -1 && errno != EEXIST)
  33         fatal("Cannot create directory '%s'.", pathname);
  34 }
  35
  36 // ----------------------------------------------------------------------------
  37 // re-load health configuration
  38
  39 void health_reload_host(RRDHOST *host) {
  40     char *path = health_config_dir();
  41
  42     // free all running alarms
  43     rrdhost_wrlock(host);
  44
  45     while(host->templates)
  46         rrdcalctemplate_free(host, host->templates);
  47
  48     while(host->alarms)
  49         rrdcalc_free(host, host->alarms);
  50
  51     rrdhost_unlock(host);
  52
  53     // invalidate all previous entries in the alarm log
  54     ALARM_ENTRY *t;
  55     for(t = host->health_log.alarms ; t ; t = t->next) {
  56         if(t->new_status != RRDCALC_STATUS_REMOVED)
  57             t->flags |= HEALTH_ENTRY_FLAG_UPDATED;
  58     }
  59
  60     // reset all thresholds to all charts
  61     RRDSET *st;
  62     for(st = host->rrdset_root; st ; st = st->next) {
  63         st->green = NAN;
  64         st->red = NAN;
  65     }
  66
  67     // load the new alarms
  68     rrdhost_wrlock(host);
  69     health_readdir(host, path);
  70     rrdhost_unlock(host);
  71
  72     // link the loaded alarms to their charts
  73     for(st = host->rrdset_root; st ; st = st->next) {
  74         rrdhost_wrlock(host);
  75
  76         rrdsetcalc_link_matching(st);
  77         rrdcalctemplate_link_matching(st);
  78
  79         rrdhost_unlock(host);
  80     }
  81 }
  82
  83 void health_reload(void) {
  84     RRDHOST *host;
  85
  86     for(host = localhost; host ; host = host->next)
  87         health_reload_host(host);
  88 }
  89
  90 // ----------------------------------------------------------------------------
  91 // health main thread and friends
  92
  93 static inline int rrdcalc_value2status(calculated_number n) {
  94     if(isnan(n) || isinf(n)) return RRDCALC_STATUS_UNDEFINED;
  95     if(n) return RRDCALC_STATUS_RAISED;
  96     return RRDCALC_STATUS_CLEAR;
  97 }
  98
  99 #define ALARM_EXEC_COMMAND_LENGTH 8192
 100
 101 static inline void health_alarm_execute(RRDHOST *host, ALARM_ENTRY *ae) {
 102     ae->flags |= HEALTH_ENTRY_FLAG_PROCESSED;
 103
 104     if(unlikely(ae->new_status < RRDCALC_STATUS_CLEAR)) {
 105         // do not send notifications for internal statuses
 106         debug(D_HEALTH, "Health not sending notification for alarm '%s.%s' status %s (internal statuses)", ae->chart, ae->name, rrdcalc_status2string(ae->new_status));
 107         goto done;
 108     }
 109
 110     if(unlikely(ae->new_status <= RRDCALC_STATUS_CLEAR && (ae->flags & HEALTH_ENTRY_FLAG_NO_CLEAR_NOTIFICATION))) {
 111         // do not send notifications for disabled statuses
 112         debug(D_HEALTH, "Health not sending notification for alarm '%s.%s' status %s (it has no-clear-notification enabled)", ae->chart, ae->name, rrdcalc_status2string(ae->new_status));
 113         // mark it as run, so that we will send the same alarm if it happens again
 114         goto done;
 115     }
 116
 117     // find the previous notification for the same alarm
 118     // which we have run the exec script
 119     // exception: alarms with HEALTH_ENTRY_FLAG_NO_CLEAR_NOTIFICATION set
 120     if(likely(!(ae->flags & HEALTH_ENTRY_FLAG_NO_CLEAR_NOTIFICATION))) {
 121         uint32_t id = ae->alarm_id;
 122         ALARM_ENTRY *t;
 123         for(t = ae->next; t ; t = t->next) {
 124             if(t->alarm_id == id && t->flags & HEALTH_ENTRY_FLAG_EXEC_RUN)
 125                 break;
 126         }
 127
 128         if(likely(t)) {
 129             // we have executed this alarm notification in the past
 130             if(t && t->new_status == ae->new_status) {
 131                 // don't send the notification for the same status again
 132                 debug(D_HEALTH, "Health not sending again notification for alarm '%s.%s' status %s", ae->chart, ae->name
 133                       , rrdcalc_status2string(ae->new_status));
 134                 goto done;
 135             }
 136         }
 137         else {
 138             // we have not executed this alarm notification in the past
 139             // so, don't send CLEAR notifications
 140             if(unlikely(ae->new_status == RRDCALC_STATUS_CLEAR)) {
 141                 debug(D_HEALTH, "Health not sending notification for first initialization of alarm '%s.%s' status %s"
 142                       , ae->chart, ae->name, rrdcalc_status2string(ae->new_status));
 143                 goto done;
 144             }
 145         }
 146     }
 147
 148     static char command_to_run[ALARM_EXEC_COMMAND_LENGTH + 1];
 149     pid_t command_pid;
 150
 151     const char *exec      = (ae->exec)      ? ae->exec      : host->health_default_exec;
 152     const char *recipient = (ae->recipient) ? ae->recipient : host->health_default_recipient;
 153
 154     snprintfz(command_to_run, ALARM_EXEC_COMMAND_LENGTH, "exec %s '%s' '%s' '%u' '%u' '%u' '%lu' '%s' '%s' '%s' '%s' '%s' '%0.0Lf' '%0.0Lf' '%s' '%u' '%u' '%s' '%s' '%s' '%s'",
 155               exec,
 156               recipient,
 157               host->hostname,
 158               ae->unique_id,
 159               ae->alarm_id,
 160               ae->alarm_event_id,
 161               (unsigned long)ae->when,
 162               ae->name,
 163               ae->chart?ae->chart:"NOCAHRT",
 164               ae->family?ae->family:"NOFAMILY",
 165               rrdcalc_status2string(ae->new_status),
 166               rrdcalc_status2string(ae->old_status),
 167               ae->new_value,
 168               ae->old_value,
 169               ae->source?ae->source:"UNKNOWN",
 170               (uint32_t)ae->duration,
 171               (uint32_t)ae->non_clear_duration,
 172               ae->units?ae->units:"",
 173               ae->info?ae->info:"",
 174               ae->new_value_string,
 175               ae->old_value_string
 176     );
 177
 178     ae->flags |= HEALTH_ENTRY_FLAG_EXEC_RUN;
 179     ae->exec_run_timestamp = now_realtime_sec();
 180
 181     debug(D_HEALTH, "executing command '%s'", command_to_run);
 182     FILE *fp = mypopen(command_to_run, &command_pid);
 183     if(!fp) {
 184         error("HEALTH: Cannot popen(\"%s\", \"r\").", command_to_run);
 185         goto done;
 186     }
 187     debug(D_HEALTH, "HEALTH reading from command");
 188     char *s = fgets(command_to_run, FILENAME_MAX, fp);
 189     (void)s;
 190     ae->exec_code = mypclose(fp, command_pid);
 191     debug(D_HEALTH, "done executing command - returned with code %d", ae->exec_code);
 192
 193     if(ae->exec_code != 0)
 194         ae->flags |= HEALTH_ENTRY_FLAG_EXEC_FAILED;
 195
 196 done:
 197     health_alarm_log_save(host, ae);
 198     return;
 199 }
 200
 201 static inline void health_process_notifications(RRDHOST *host, ALARM_ENTRY *ae) {
 202     debug(D_HEALTH, "Health alarm '%s.%s' = %0.2Lf - changed status from %s to %s",
 203          ae->chart?ae->chart:"NOCHART", ae->name,
 204          ae->new_value,
 205          rrdcalc_status2string(ae->old_status),
 206          rrdcalc_status2string(ae->new_status)
 207     );
 208
 209     health_alarm_execute(host, ae);
 210 }
 211
 212 static inline void health_alarm_log_process(RRDHOST *host) {
 213     static uint32_t stop_at_id = 0;
 214     uint32_t first_waiting = (host->health_log.alarms)?host->health_log.alarms->unique_id:0;
 215     time_t now = now_realtime_sec();
 216
 217     pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
 218
 219     ALARM_ENTRY *ae;
 220     for(ae = host->health_log.alarms; ae && ae->unique_id >= stop_at_id ; ae = ae->next) {
 221         if(unlikely(
 222             !(ae->flags & HEALTH_ENTRY_FLAG_PROCESSED) &&
 223             !(ae->flags & HEALTH_ENTRY_FLAG_UPDATED)
 224             )) {
 225
 226             if(unlikely(ae->unique_id < first_waiting))
 227                 first_waiting = ae->unique_id;
 228
 229             if(likely(now >= ae->delay_up_to_timestamp))
 230                 health_process_notifications(host, ae);
 231         }
 232     }
 233
 234     // remember this for the next iteration
 235     stop_at_id = first_waiting;
 236
 237     pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
 238
 239     if(host->health_log.count <= host->health_log.max)
 240         return;
 241
 242     // cleanup excess entries in the log
 243     pthread_rwlock_wrlock(&host->health_log.alarm_log_rwlock);
 244
 245     ALARM_ENTRY *last = NULL;
 246     unsigned int count = host->health_log.max * 2 / 3;
 247     for(ae = host->health_log.alarms; ae && count ; count--, last = ae, ae = ae->next) ;
 248
 249     if(ae && last && last->next == ae)
 250         last->next = NULL;
 251     else
 252         ae = NULL;
 253
 254     while(ae) {
 255         debug(D_HEALTH, "Health removing alarm log entry with id: %u", ae->unique_id);
 256
 257         ALARM_ENTRY *t = ae->next;
 258
 259         health_alarm_log_free_one_nochecks_nounlink(ae);
 260
 261         ae = t;
 262         host->health_log.count--;
 263     }
 264
 265     pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
 266 }
 267
 268 static inline int rrdcalc_isrunnable(RRDCALC *rc, time_t now, time_t *next_run) {
 269     if(unlikely(!rc->rrdset)) {
 270         debug(D_HEALTH, "Health not running alarm '%s.%s'. It is not linked to a chart.", rc->chart?rc->chart:"NOCHART", rc->name);
 271         return 0;
 272     }
 273
 274     if(unlikely(rc->next_update > now)) {
 275         if (unlikely(*next_run > rc->next_update)) {
 276             // update the next_run time of the main loop
 277             // to run this alarm precisely the time required
 278             *next_run = rc->next_update;
 279         }
 280
 281         debug(D_HEALTH, "Health not examining alarm '%s.%s' yet (will do in %d secs).", rc->chart?rc->chart:"NOCHART", rc->name, (int) (rc->next_update - now));
 282         return 0;
 283     }
 284
 285     if(unlikely(!rc->update_every)) {
 286         debug(D_HEALTH, "Health not running alarm '%s.%s'. It does not have an update frequency", rc->chart?rc->chart:"NOCHART", rc->name);
 287         return 0;
 288     }
 289
 290     if(unlikely(!rc->rrdset->last_collected_time.tv_sec || rc->rrdset->counter_done < 2)) {
 291         debug(D_HEALTH, "Health not running alarm '%s.%s'. Chart is not fully collected yet.", rc->chart?rc->chart:"NOCHART", rc->name);
 292         return 0;
 293     }
 294
 295     int update_every = rc->rrdset->update_every;
 296     time_t first = rrdset_first_entry_t(rc->rrdset);
 297     time_t last = rrdset_last_entry_t(rc->rrdset);
 298
 299     if(unlikely(now + update_every < first /* || now - update_every > last */)) {
 300         debug(D_HEALTH
 301               , "Health not examining alarm '%s.%s' yet (wanted time is out of bounds - we need %lu but got %lu - %lu)."
 302               , rc->chart ? rc->chart : "NOCHART", rc->name, (unsigned long) now, (unsigned long) first
 303               , (unsigned long) last);
 304         return 0;
 305     }
 306
 307     if(RRDCALC_HAS_DB_LOOKUP(rc)) {
 308         time_t needed = now + rc->before + rc->after;
 309
 310         if(needed + update_every < first || needed - update_every > last) {
 311             debug(D_HEALTH
 312                   , "Health not examining alarm '%s.%s' yet (not enough data yet - we need %lu but got %lu - %lu)."
 313                   , rc->chart ? rc->chart : "NOCHART", rc->name, (unsigned long) needed, (unsigned long) first
 314                   , (unsigned long) last);
 315             return 0;
 316         }
 317     }
 318
 319     return 1;
 320 }
 321
 322 void *health_main(void *ptr) {
 323     struct netdata_static_thread *static_thread = (struct netdata_static_thread *)ptr;
 324
 325     info("HEALTH thread created with task id %d", gettid());
 326
 327     if(pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, NULL) != 0)
 328         error("Cannot set pthread cancel type to DEFERRED.");
 329
 330     if(pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL) != 0)
 331         error("Cannot set pthread cancel state to ENABLE.");
 332
 333     int min_run_every = (int)config_get_number("health", "run at least every seconds", 10);
 334     if(min_run_every < 1) min_run_every = 1;
 335
 336     BUFFER *wb = buffer_create(100);
 337
 338     unsigned int loop = 0;
 339     while(!netdata_exit) {
 340         loop++;
 341         debug(D_HEALTH, "Health monitoring iteration no %u started", loop);
 342
 343         int oldstate, runnable = 0;
 344         time_t now = now_realtime_sec();
 345         time_t next_run = now + min_run_every;
 346         RRDCALC *rc;
 347
 348         if(unlikely(pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate) != 0))
 349             error("Cannot set pthread cancel state to DISABLE.");
 350
 351         RRDHOST *host;
 352         for(host = localhost; host ; host = host->next) {
 353             if(unlikely(!host->health_enabled)) continue;
 354
 355             rrdhost_rdlock(host);
 356
 357             // the first loop is to lookup values from the db
 358             for(rc = host->alarms; rc; rc = rc->next) {
 359                 if(unlikely(!rrdcalc_isrunnable(rc, now, &next_run))) {
 360                     if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_RUNNABLE))
 361                         rc->rrdcalc_flags &= ~RRDCALC_FLAG_RUNNABLE;
 362                     continue;
 363                 }
 364
 365                 runnable++;
 366                 rc->old_value = rc->value;
 367                 rc->rrdcalc_flags |= RRDCALC_FLAG_RUNNABLE;
 368
 369                 // 1. if there is database lookup, do it
 370                 // 2. if there is calculation expression, run it
 371
 372                 if(unlikely(RRDCALC_HAS_DB_LOOKUP(rc))) {
 373                     /* time_t old_db_timestamp = rc->db_before; */
 374                     int value_is_null = 0;
 375
 376                     int ret = rrd2value(rc->rrdset, wb, &rc->value, rc->dimensions, 1, rc->after, rc->before, rc->group, rc->options, &rc->db_after, &rc->db_before, &value_is_null);
 377
 378                     if(unlikely(ret != 200)) {
 379                         // database lookup failed
 380                         rc->value = NAN;
 381
 382                         debug(D_HEALTH, "Health on host '%s', alarm '%s.%s': database lookup returned error %d", host->hostname, rc->chart ? rc->chart : "NOCHART", rc->name, ret);
 383
 384                         if(unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_ERROR))) {
 385                             rc->rrdcalc_flags |= RRDCALC_FLAG_DB_ERROR;
 386                             error("Health on host '%s', alarm '%s.%s': database lookup returned error %d", host->hostname, rc->chart ? rc->chart : "NOCHART", rc->name, ret);
 387                         }
 388                     }
 389                     else if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_ERROR))
 390                         rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_ERROR;
 391
 392                     /* - RRDCALC_FLAG_DB_STALE not currently used
 393                     if (unlikely(old_db_timestamp == rc->db_before)) {
 394                         // database is stale
 395
 396                         debug(D_HEALTH, "Health on host '%s', alarm '%s.%s': database is stale", host->hostname, rc->chart?rc->chart:"NOCHART", rc->name);
 397
 398                         if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_STALE))) {
 399                             rc->rrdcalc_flags |= RRDCALC_FLAG_DB_STALE;
 400                             error("Health on host '%s', alarm '%s.%s': database is stale", host->hostname, rc->chart?rc->chart:"NOCHART", rc->name);
 401                         }
 402                     }
 403                     else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_STALE))
 404                         rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_STALE;
 405                     */
 406
 407                     if(unlikely(value_is_null)) {
 408                         // collected value is null
 409
 410                         rc->value = NAN;
 411
 412                         debug(D_HEALTH, "Health on host '%s', alarm '%s.%s': database lookup returned empty value (possibly value is not collected yet)", host->hostname, rc->chart ? rc->chart : "NOCHART", rc->name);
 413
 414                         if(unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_NAN))) {
 415                             rc->rrdcalc_flags |= RRDCALC_FLAG_DB_NAN;
 416                             error("Health on host '%s', alarm '%s.%s': database lookup returned empty value (possibly value is not collected yet)", host->hostname,  rc->chart ? rc->chart : "NOCHART", rc->name);
 417                         }
 418                     }
 419                     else if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_NAN))
 420                         rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_NAN;
 421
 422                     debug(D_HEALTH, "Health on host '%s', alarm '%s.%s': database lookup gave value " CALCULATED_NUMBER_FORMAT, host->hostname, rc->chart ? rc->chart : "NOCHART", rc->name, rc->value);
 423                 }
 424
 425                 if(unlikely(rc->calculation)) {
 426                     if(unlikely(!expression_evaluate(rc->calculation))) {
 427                         // calculation failed
 428
 429                         rc->value = NAN;
 430
 431                         debug(D_HEALTH, "Health on host '%s', alarm '%s.%s': expression '%s' failed: %s", host->hostname, rc->chart ? rc->chart : "NOCHART", rc->name, rc->calculation->parsed_as, buffer_tostring(rc->calculation->error_msg));
 432
 433                         if(unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_CALC_ERROR))) {
 434                             rc->rrdcalc_flags |= RRDCALC_FLAG_CALC_ERROR;
 435                             error("Health on host '%s', alarm '%s.%s': expression '%s' failed: %s", rc->chart ? rc->chart : "NOCHART", host->hostname,  rc->name, rc->calculation->parsed_as, buffer_tostring(rc->calculation->error_msg));
 436                         }
 437                     }
 438                     else {
 439                         if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_CALC_ERROR))
 440                             rc->rrdcalc_flags &= ~RRDCALC_FLAG_CALC_ERROR;
 441
 442                         debug(D_HEALTH, "Health on host '%s', alarm '%s.%s': expression '%s' gave value "
 443                                 CALCULATED_NUMBER_FORMAT
 444                                 ": %s (source: %s)", host->hostname, rc->chart ? rc->chart : "NOCHART", rc->name
 445                               , rc->calculation->parsed_as, rc->calculation->result,
 446                                 buffer_tostring(rc->calculation->error_msg), rc->source
 447                         );
 448
 449                         rc->value = rc->calculation->result;
 450                     }
 451                 }
 452             }
 453             rrdhost_unlock(host);
 454
 455             if(unlikely(runnable && !netdata_exit)) {
 456                 rrdhost_rdlock(host);
 457
 458                 for(rc = host->alarms; rc; rc = rc->next) {
 459                     if(unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_RUNNABLE)))
 460                         continue;
 461
 462                     int warning_status = RRDCALC_STATUS_UNDEFINED;
 463                     int critical_status = RRDCALC_STATUS_UNDEFINED;
 464
 465                     if(likely(rc->warning)) {
 466                         if(unlikely(!expression_evaluate(rc->warning))) {
 467                             // calculation failed
 468
 469                             debug(D_HEALTH, "Health on host '%s', alarm '%s.%s': warning expression failed with error: %s", host->hostname, rc->chart ? rc->chart : "NOCHART", rc->name, buffer_tostring(rc->warning->error_msg));
 470
 471                             if(unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_WARN_ERROR))) {
 472                                 rc->rrdcalc_flags |= RRDCALC_FLAG_WARN_ERROR;
 473                                 error("Health on host '%s', alarm '%s.%s': warning expression failed with error: %s", host->hostname, rc->chart ? rc->chart : "NOCHART", rc->name, buffer_tostring(rc->warning->error_msg));
 474                             }
 475                         }
 476                         else {
 477                             if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_WARN_ERROR))
 478                                 rc->rrdcalc_flags &= ~RRDCALC_FLAG_WARN_ERROR;
 479
 480                             debug(D_HEALTH, "Health on host '%s', alarm '%s.%s': warning expression gave value " CALCULATED_NUMBER_FORMAT ": %s (source: %s)", host->hostname, rc->chart ? rc->chart : "NOCHART", rc->name, rc->warning->result, buffer_tostring(rc->warning->error_msg), rc->source);
 481
 482                             warning_status = rrdcalc_value2status(rc->warning->result);
 483                         }
 484                     }
 485
 486                     if(likely(rc->critical)) {
 487                         if(unlikely(!expression_evaluate(rc->critical))) {
 488                             // calculation failed
 489
 490                             debug(D_HEALTH, "Health on host '%s', alarm '%s.%s': critical expression failed with error: %s", host->hostname, rc->chart ? rc->chart : "NOCHART", rc->name, buffer_tostring(rc->critical->error_msg));
 491
 492                             if(unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_CRIT_ERROR))) {
 493                                 rc->rrdcalc_flags |= RRDCALC_FLAG_CRIT_ERROR;
 494                                 error("Health on host '%s', alarm '%s.%s': critical expression failed with error: %s", host->hostname, rc->chart ? rc->chart : "NOCHART", rc->name, buffer_tostring(rc->critical->error_msg));
 495                             }
 496                         }
 497                         else {
 498                             if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_CRIT_ERROR))
 499                                 rc->rrdcalc_flags &= ~RRDCALC_FLAG_CRIT_ERROR;
 500
 501                             debug(D_HEALTH, "Health on host '%s', alarm '%s.%s': critical expression gave value " CALCULATED_NUMBER_FORMAT ": %s (source: %s)", host->hostname, rc->chart ? rc->chart : "NOCHART", rc->name , rc->critical->result, buffer_tostring(rc->critical->error_msg), rc->source);
 502
 503                             critical_status = rrdcalc_value2status(rc->critical->result);
 504                         }
 505                     }
 506
 507                     int status = RRDCALC_STATUS_UNDEFINED;
 508
 509                     switch(warning_status) {
 510                         case RRDCALC_STATUS_CLEAR:
 511                             status = RRDCALC_STATUS_CLEAR;
 512                             break;
 513
 514                         case RRDCALC_STATUS_RAISED:
 515                             status = RRDCALC_STATUS_WARNING;
 516                             break;
 517
 518                         default:
 519                             break;
 520                     }
 521
 522                     switch(critical_status) {
 523                         case RRDCALC_STATUS_CLEAR:
 524                             if(status == RRDCALC_STATUS_UNDEFINED)
 525                                 status = RRDCALC_STATUS_CLEAR;
 526                             break;
 527
 528                         case RRDCALC_STATUS_RAISED:
 529                             status = RRDCALC_STATUS_CRITICAL;
 530                             break;
 531
 532                         default:
 533                             break;
 534                     }
 535
 536                     if(status != rc->status) {
 537                         int delay = 0;
 538
 539                         if(now > rc->delay_up_to_timestamp) {
 540                             rc->delay_up_current = rc->delay_up_duration;
 541                             rc->delay_down_current = rc->delay_down_duration;
 542                             rc->delay_last = 0;
 543                             rc->delay_up_to_timestamp = 0;
 544                         }
 545                         else {
 546                             rc->delay_up_current = (int) (rc->delay_up_current * rc->delay_multiplier);
 547                             if(rc->delay_up_current > rc->delay_max_duration)
 548                                 rc->delay_up_current = rc->delay_max_duration;
 549
 550                             rc->delay_down_current = (int) (rc->delay_down_current * rc->delay_multiplier);
 551                             if(rc->delay_down_current > rc->delay_max_duration)
 552                                 rc->delay_down_current = rc->delay_max_duration;
 553                         }
 554
 555                         if(status > rc->status)
 556                             delay = rc->delay_up_current;
 557                         else
 558                             delay = rc->delay_down_current;
 559
 560                         // COMMENTED: because we do need to send raising alarms
 561                         // if(now + delay < rc->delay_up_to_timestamp)
 562                         //    delay = (int)(rc->delay_up_to_timestamp - now);
 563
 564                         rc->delay_last = delay;
 565                         rc->delay_up_to_timestamp = now + delay;
 566                         health_alarm_log(
 567                                 host, rc->id, rc->next_event_id++, now, rc->name, rc->rrdset->id
 568                                 , rc->rrdset->family, rc->exec, rc->recipient, now - rc->last_status_change
 569                                 , rc->old_value, rc->value, rc->status, status, rc->source, rc->units, rc->info
 570                                 , rc->delay_last, (rc->options & RRDCALC_FLAG_NO_CLEAR_NOTIFICATION)
 571                                                   ? HEALTH_ENTRY_FLAG_NO_CLEAR_NOTIFICATION : 0
 572                         );
 573                         rc->last_status_change = now;
 574                         rc->status = status;
 575                     }
 576
 577                     rc->last_updated = now;
 578                     rc->next_update = now + rc->update_every;
 579
 580                     if(next_run > rc->next_update)
 581                         next_run = rc->next_update;
 582                 }
 583
 584                 rrdhost_unlock(host);
 585             }
 586
 587             if(unlikely(netdata_exit))
 588                 break;
 589
 590             // execute notifications
 591             // and cleanup
 592             health_alarm_log_process(host);
 593
 594             if(unlikely(netdata_exit))
 595                 break;
 596
 597         } /* host loop */
 598
 599         if(unlikely(pthread_setcancelstate(oldstate, NULL) != 0))
 600             error("Cannot set pthread cancel state to RESTORE (%d).", oldstate);
 601
 602         if(unlikely(netdata_exit))
 603             break;
 604
 605         now = now_realtime_sec();
 606         if(now < next_run) {
 607             debug(D_HEALTH, "Health monitoring iteration no %u done. Next iteration in %d secs", loop, (int) (next_run - now));
 608             sleep_usec(USEC_PER_SEC * (usec_t) (next_run - now));
 609         }
 610         else
 611             debug(D_HEALTH, "Health monitoring iteration no %u done. Next iteration now", loop);
 612     }
 613
 614     buffer_free(wb);
 615
 616     info("HEALTH thread exiting");
 617
 618     static_thread->enabled = 0;
 619     pthread_exit(NULL);
 620     return NULL;
 621 }