3 #define RRDVAR_MAX_LENGTH 1024
5 struct health_options {
6 const char *health_default_exec;
7 const char *health_default_recipient;
8 const char *log_filename;
12 static struct health_options health = {
13 .health_default_exec = PLUGINS_DIR "/alarm-notify.sh",
14 .health_default_recipient = "root",
15 .log_filename = VARLIB_DIR "/health/alarm_log.db",
19 int health_enabled = 1;
21 // ----------------------------------------------------------------------------
22 // health alarm log load/save
23 // no need for locking - only one thread is reading / writing the alarms log
25 static inline int health_alarm_log_open(void) {
27 fclose(health.log_fp);
29 health.log_fp = fopen(health.log_filename, "a");
32 if (setvbuf(health.log_fp, NULL, _IOLBF, 0) != 0)
33 error("Cannot set line buffering on health log file.");
37 error("Cannot open health log file '%s'. Health data will be lost in case of netdata or server crash.", health.log_filename);
41 static inline void health_alarm_log_close(void) {
43 fclose(health.log_fp);
48 static inline void health_log_recreate(void) {
49 if(health.log_fp != NULL) {
50 health_alarm_log_close();
52 // open it with truncate
53 health.log_fp = fopen(health.log_filename, "w");
54 if(health.log_fp) fclose(health.log_fp);
55 else error("Cannot truncate health log '%s'", health.log_filename);
59 health_alarm_log_open();
63 static inline void health_alarm_log_save(RRDHOST *host, ALARM_ENTRY *ae) {
67 /* if(likely(health.log_fp)) {
68 if(unlikely(fprintf(health.log_fp, "A\t%s\t%08x\t%08x\t%08x\t%08x\t%08x\t%08x\t%s\t%s\t%s\t%s\t%s\t%08x\n",
74 (uint32_t)ae->duration,
75 (uint32_t)ae->non_clear_duration,
76 (uint32_t)ae->exec_run_timestamp,
83 error("Health: failed to save alarm log entry. Health data may be lost in case of abnormal restart.");
88 static inline void health_alarm_log_load(RRDHOST *host) {
93 // ----------------------------------------------------------------------------
94 // health alarm log management
96 static inline void health_alarm_log(RRDHOST *host,
97 uint32_t alarm_id, uint32_t alarm_event_id,
99 const char *name, const char *chart, const char *family,
100 const char *exec, const char *recipient, time_t duration,
101 calculated_number old_value, calculated_number new_value,
102 int old_status, int new_status,
108 debug(D_HEALTH, "Health adding alarm log entry with id: %u", host->health_log.next_log_id);
110 ALARM_ENTRY *ae = callocz(1, sizeof(ALARM_ENTRY));
111 ae->name = strdupz(name);
112 ae->hash_name = simple_hash(ae->name);
115 ae->chart = strdupz(chart);
116 ae->hash_chart = simple_hash(ae->chart);
120 ae->family = strdupz(family);
122 if(exec) ae->exec = strdupz(exec);
123 if(recipient) ae->recipient = strdupz(recipient);
124 if(source) ae->source = strdupz(source);
125 if(units) ae->units = strdupz(units);
126 if(info) ae->info = strdupz(info);
128 ae->unique_id = host->health_log.next_log_id++;
129 ae->alarm_id = alarm_id;
130 ae->alarm_event_id = alarm_event_id;
132 ae->old_value = old_value;
133 ae->new_value = new_value;
134 ae->old_status = old_status;
135 ae->new_status = new_status;
136 ae->duration = duration;
138 ae->delay_up_to_timestamp = when + delay;
140 if(ae->old_status == RRDCALC_STATUS_WARNING || ae->old_status == RRDCALC_STATUS_CRITICAL)
141 ae->non_clear_duration += ae->duration;
144 pthread_rwlock_wrlock(&host->health_log.alarm_log_rwlock);
145 ae->next = host->health_log.alarms;
146 host->health_log.alarms = ae;
147 host->health_log.count++;
148 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
150 // match previous alarms
151 pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
153 for(t = host->health_log.alarms ; t ; t = t->next) {
154 if(t != ae && t->alarm_id == ae->alarm_id) {
155 if(!(t->notifications & HEALTH_ENTRY_NOTIFICATIONS_UPDATED) && !t->updated_by) {
156 t->notifications |= HEALTH_ENTRY_NOTIFICATIONS_UPDATED;
159 if((t->new_status == RRDCALC_STATUS_WARNING || t->new_status == RRDCALC_STATUS_CRITICAL) &&
160 (t->old_status == RRDCALC_STATUS_WARNING || t->old_status == RRDCALC_STATUS_CRITICAL))
161 ae->non_clear_duration += t->non_clear_duration;
163 health_alarm_log_save(host, t);
166 // no need to continue
171 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
173 health_alarm_log_save(host, ae);
176 // ----------------------------------------------------------------------------
179 static inline int rrdvar_fix_name(char *variable) {
182 if (!isalnum(*variable) && *variable != '.' && *variable != '_') {
193 int rrdvar_compare(void* a, void* b) {
194 if(((RRDVAR *)a)->hash < ((RRDVAR *)b)->hash) return -1;
195 else if(((RRDVAR *)a)->hash > ((RRDVAR *)b)->hash) return 1;
196 else return strcmp(((RRDVAR *)a)->name, ((RRDVAR *)b)->name);
199 static inline RRDVAR *rrdvar_index_add(avl_tree_lock *tree, RRDVAR *rv) {
200 RRDVAR *ret = (RRDVAR *)avl_insert_lock(tree, (avl *)(rv));
202 debug(D_VARIABLES, "Request to insert RRDVAR '%s' into index failed. Already exists.", rv->name);
207 static inline RRDVAR *rrdvar_index_del(avl_tree_lock *tree, RRDVAR *rv) {
208 RRDVAR *ret = (RRDVAR *)avl_remove_lock(tree, (avl *)(rv));
210 error("Request to remove RRDVAR '%s' from index failed. Not Found.", rv->name);
215 static inline RRDVAR *rrdvar_index_find(avl_tree_lock *tree, const char *name, uint32_t hash) {
217 tmp.name = (char *)name;
218 tmp.hash = (hash)?hash:simple_hash(tmp.name);
220 return (RRDVAR *)avl_search_lock(tree, (avl *)&tmp);
223 static inline void rrdvar_free(RRDHOST *host, avl_tree_lock *tree, RRDVAR *rv) {
229 rrdvar_index_del(tree, rv);
235 static inline RRDVAR *rrdvar_create_and_index(const char *scope, avl_tree_lock *tree, const char *name, int type, calculated_number *value) {
236 char *variable = strdupz(name);
237 rrdvar_fix_name(variable);
238 uint32_t hash = simple_hash(variable);
240 RRDVAR *rv = rrdvar_index_find(tree, variable, hash);
242 debug(D_VARIABLES, "Variable '%s' not found in scope '%s'. Creating a new one.", variable, scope);
244 rv = callocz(1, sizeof(RRDVAR));
250 RRDVAR *ret = rrdvar_index_add(tree, rv);
251 if(unlikely(ret != rv)) {
252 debug(D_VARIABLES, "Variable '%s' in scope '%s' already exists", variable, scope);
253 rrdvar_free(NULL, NULL, rv);
257 debug(D_VARIABLES, "Variable '%s' created in scope '%s'", variable, scope);
268 // ----------------------------------------------------------------------------
271 calculated_number rrdvar2number(RRDVAR *rv) {
273 case RRDVAR_TYPE_CALCULATED: {
274 calculated_number *n = (calculated_number *)rv->value;
278 case RRDVAR_TYPE_TIME_T: {
279 time_t *n = (time_t *)rv->value;
283 case RRDVAR_TYPE_COLLECTED: {
284 collected_number *n = (collected_number *)rv->value;
288 case RRDVAR_TYPE_TOTAL: {
289 total_number *n = (total_number *)rv->value;
293 case RRDVAR_TYPE_INT: {
294 int *n = (int *)rv->value;
299 error("I don't know how to convert RRDVAR type %d to calculated_number", rv->type);
304 void dump_variable(void *data) {
305 RRDVAR *rv = (RRDVAR *)data;
306 debug(D_HEALTH, "%50s : %20.5Lf", rv->name, rrdvar2number(rv));
309 int health_variable_lookup(const char *variable, uint32_t hash, RRDCALC *rc, calculated_number *result) {
310 RRDSET *st = rc->rrdset;
315 rv = rrdvar_index_find(&st->variables_root_index, variable, hash);
317 *result = rrdvar2number(rv);
321 rv = rrdvar_index_find(&st->rrdfamily->variables_root_index, variable, hash);
323 *result = rrdvar2number(rv);
327 rv = rrdvar_index_find(&st->rrdhost->variables_root_index, variable, hash);
329 *result = rrdvar2number(rv);
333 debug(D_HEALTH, "Available local chart '%s' variables:", st->id);
334 avl_traverse_lock(&st->variables_root_index, dump_variable);
336 debug(D_HEALTH, "Available family '%s' variables:", st->rrdfamily->family);
337 avl_traverse_lock(&st->rrdfamily->variables_root_index, dump_variable);
339 debug(D_HEALTH, "Available host '%s' variables:", st->rrdhost->hostname);
340 avl_traverse_lock(&st->rrdhost->variables_root_index, dump_variable);
345 // ----------------------------------------------------------------------------
346 // RRDSETVAR management
348 RRDSETVAR *rrdsetvar_create(RRDSET *st, const char *variable, int type, void *value, uint32_t options) {
349 debug(D_VARIABLES, "RRDVARSET create for chart id '%s' name '%s' with variable name '%s'", st->id, st->name, variable);
350 RRDSETVAR *rs = (RRDSETVAR *)callocz(1, sizeof(RRDSETVAR));
352 char buffer[RRDVAR_MAX_LENGTH + 1];
353 snprintfz(buffer, RRDVAR_MAX_LENGTH, "%s.%s", st->id, variable);
354 rs->fullid = strdupz(buffer);
356 snprintfz(buffer, RRDVAR_MAX_LENGTH, "%s.%s", st->name, variable);
357 rs->fullname = strdupz(buffer);
359 rs->variable = strdupz(variable);
363 rs->options = options;
366 rs->local = rrdvar_create_and_index("local", &st->variables_root_index, rs->variable, rs->type, rs->value);
367 rs->family = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->fullid, rs->type, rs->value);
368 rs->host = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullid, rs->type, rs->value);
369 rs->family_name = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->fullname, rs->type, rs->value);
370 rs->host_name = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullname, rs->type, rs->value);
372 rs->next = st->variables;
378 void rrdsetvar_rename_all(RRDSET *st) {
379 debug(D_VARIABLES, "RRDSETVAR rename for chart id '%s' name '%s'", st->id, st->name);
381 // only these 2 can change name
385 char buffer[RRDVAR_MAX_LENGTH + 1];
386 RRDSETVAR *rs, *next = st->variables;
390 snprintfz(buffer, RRDVAR_MAX_LENGTH, "%s.%s", st->name, rs->variable);
392 if (strcmp(buffer, rs->fullname)) {
394 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family_name);
395 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_name);
398 rs->fullname = strdupz(st->name);
399 rs->family_name = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->fullname, rs->type, rs->value);
400 rs->host_name = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullname, rs->type, rs->value);
404 rrdsetcalc_link_matching(st);
407 void rrdsetvar_free(RRDSETVAR *rs) {
408 RRDSET *st = rs->rrdset;
409 debug(D_VARIABLES, "RRDSETVAR free for chart id '%s' name '%s', variable '%s'", st->id, st->name, rs->variable);
411 rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local);
412 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family);
413 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host);
414 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family_name);
415 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_name);
417 if(st->variables == rs) {
418 st->variables = rs->next;
422 for (t = st->variables; t && t->next != rs; t = t->next);
423 if(!t) error("RRDSETVAR '%s' not found in chart '%s' variables linked list", rs->fullname, st->id);
424 else t->next = rs->next;
433 // ----------------------------------------------------------------------------
434 // RRDDIMVAR management
436 #define RRDDIMVAR_ID_MAX 1024
438 RRDDIMVAR *rrddimvar_create(RRDDIM *rd, int type, const char *prefix, const char *suffix, void *value, uint32_t options) {
439 RRDSET *st = rd->rrdset;
441 debug(D_VARIABLES, "RRDDIMSET create for chart id '%s' name '%s', dimension id '%s', name '%s%s%s'", st->id, st->name, rd->id, (prefix)?prefix:"", rd->name, (suffix)?suffix:"");
443 if(!prefix) prefix = "";
444 if(!suffix) suffix = "";
446 char buffer[RRDDIMVAR_ID_MAX + 1];
447 RRDDIMVAR *rs = (RRDDIMVAR *)callocz(1, sizeof(RRDDIMVAR));
449 rs->prefix = strdupz(prefix);
450 rs->suffix = strdupz(suffix);
452 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s%s%s", rs->prefix, rd->id, rs->suffix);
453 rs->id = strdupz(buffer);
455 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s%s%s", rs->prefix, rd->name, rs->suffix);
456 rs->name = strdupz(buffer);
458 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->id, rs->id);
459 rs->fullidid = strdupz(buffer);
461 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->id, rs->name);
462 rs->fullidname = strdupz(buffer);
464 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->name, rs->id);
465 rs->fullnameid = strdupz(buffer);
467 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->name, rs->name);
468 rs->fullnamename = strdupz(buffer);
472 rs->options = options;
475 rs->local_id = rrdvar_create_and_index("local", &st->variables_root_index, rs->id, rs->type, rs->value);
476 rs->local_name = rrdvar_create_and_index("local", &st->variables_root_index, rs->name, rs->type, rs->value);
478 rs->family_id = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->id, rs->type, rs->value);
479 rs->family_name = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->name, rs->type, rs->value);
481 rs->host_fullidid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullidid, rs->type, rs->value);
482 rs->host_fullidname = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullidname, rs->type, rs->value);
483 rs->host_fullnameid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullnameid, rs->type, rs->value);
484 rs->host_fullnamename = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullnamename, rs->type, rs->value);
486 rs->next = rd->variables;
492 void rrddimvar_rename_all(RRDDIM *rd) {
493 RRDSET *st = rd->rrdset;
494 debug(D_VARIABLES, "RRDDIMSET rename for chart id '%s' name '%s', dimension id '%s', name '%s'", st->id, st->name, rd->id, rd->name);
496 RRDDIMVAR *rs, *next = rd->variables;
500 if (strcmp(rd->name, rs->name)) {
501 char buffer[RRDDIMVAR_ID_MAX + 1];
505 rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local_name);
507 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s%s%s", rs->prefix, rd->name, rs->suffix);
508 rs->name = strdupz(buffer);
509 rs->local_name = rrdvar_create_and_index("local", &st->variables_root_index, rs->name, rs->type, rs->value);
511 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullidname);
512 freez(rs->fullidname);
513 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->id, rs->name);
514 rs->fullidname = strdupz(buffer);
515 rs->host_fullidname = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index,
516 rs->fullidname, rs->type, rs->value);
519 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnameid);
520 freez(rs->fullnameid);
521 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->name, rs->id);
522 rs->fullnameid = strdupz(buffer);
523 rs->host_fullnameid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index,
524 rs->fullnameid, rs->type, rs->value);
527 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnamename);
528 freez(rs->fullnamename);
529 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->name, rs->name);
530 rs->fullnamename = strdupz(buffer);
531 rs->host_fullnamename = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index,
532 rs->fullnamename, rs->type, rs->value);
537 void rrddimvar_free(RRDDIMVAR *rs) {
538 RRDDIM *rd = rs->rrddim;
539 RRDSET *st = rd->rrdset;
540 debug(D_VARIABLES, "RRDDIMSET free for chart id '%s' name '%s', dimension id '%s', name '%s', prefix='%s', suffix='%s'", st->id, st->name, rd->id, rd->name, rs->prefix, rs->suffix);
542 rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local_id);
543 rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local_name);
545 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family_id);
546 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family_name);
548 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullidid);
549 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullidname);
550 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnameid);
551 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnamename);
553 if(rd->variables == rs) {
554 debug(D_VARIABLES, "RRDDIMSET removing first entry for chart id '%s' name '%s', dimension id '%s', name '%s'", st->id, st->name, rd->id, rd->name);
555 rd->variables = rs->next;
558 debug(D_VARIABLES, "RRDDIMSET removing non-first entry for chart id '%s' name '%s', dimension id '%s', name '%s'", st->id, st->name, rd->id, rd->name);
560 for (t = rd->variables; t && t->next != rs; t = t->next) ;
561 if(!t) error("RRDDIMVAR '%s' not found in dimension '%s/%s' variables linked list", rs->name, st->id, rd->id);
562 else t->next = rs->next;
570 freez(rs->fullidname);
571 freez(rs->fullnameid);
572 freez(rs->fullnamename);
576 // ----------------------------------------------------------------------------
577 // RRDCALC management
579 static inline const char *rrdcalc_status2string(int status) {
581 case RRDCALC_STATUS_REMOVED:
584 case RRDCALC_STATUS_UNDEFINED:
587 case RRDCALC_STATUS_UNINITIALIZED:
588 return "UNINITIALIZED";
590 case RRDCALC_STATUS_CLEAR:
593 case RRDCALC_STATUS_RAISED:
596 case RRDCALC_STATUS_WARNING:
599 case RRDCALC_STATUS_CRITICAL:
603 error("Unknown alarm status %d", status);
608 static void rrdsetcalc_link(RRDSET *st, RRDCALC *rc) {
609 debug(D_HEALTH, "Health linking alarm '%s.%s' to chart '%s' of host '%s'", rc->chart?rc->chart:"NOCHART", rc->name, st->id, st->rrdhost->hostname);
611 rc->last_status_change = time(NULL);
614 rc->rrdset_next = st->alarms;
615 rc->rrdset_prev = NULL;
618 rc->rrdset_next->rrdset_prev = rc;
622 if(rc->update_every < rc->rrdset->update_every) {
623 error("Health alarm '%s.%s' has update every %d, less than chart update every %d. Setting alarm update frequency to %d.", rc->rrdset->id, rc->name, rc->update_every, rc->rrdset->update_every, rc->rrdset->update_every);
624 rc->update_every = rc->rrdset->update_every;
627 if(!isnan(rc->green) && isnan(st->green)) {
628 debug(D_HEALTH, "Health alarm '%s.%s' green threshold set from %Lf to %Lf.", rc->rrdset->id, rc->name, rc->rrdset->green, rc->green);
629 st->green = rc->green;
632 if(!isnan(rc->red) && isnan(st->red)) {
633 debug(D_HEALTH, "Health alarm '%s.%s' red threshold set from %Lf to %Lf.", rc->rrdset->id, rc->name, rc->rrdset->red, rc->red);
637 rc->local = rrdvar_create_and_index("local", &st->variables_root_index, rc->name, RRDVAR_TYPE_CALCULATED, &rc->value);
638 rc->family = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rc->name, RRDVAR_TYPE_CALCULATED, &rc->value);
640 char fullname[RRDVAR_MAX_LENGTH + 1];
641 snprintfz(fullname, RRDVAR_MAX_LENGTH, "%s.%s", st->id, rc->name);
642 rc->hostid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, fullname, RRDVAR_TYPE_CALCULATED, &rc->value);
644 snprintfz(fullname, RRDVAR_MAX_LENGTH, "%s.%s", st->name, rc->name);
645 rc->hostname = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, fullname, RRDVAR_TYPE_CALCULATED, &rc->value);
647 if(!rc->units) rc->units = strdupz(st->units);
650 time_t now = time(NULL);
651 health_alarm_log(st->rrdhost, rc->id, rc->next_event_id++, now, rc->name, rc->rrdset->id, rc->rrdset->family, rc->exec, rc->recipient, now - rc->last_status_change, rc->old_value, rc->value, rc->status, RRDCALC_STATUS_UNINITIALIZED, rc->source, rc->units, rc->info, 0);
655 static inline int rrdcalc_is_matching_this_rrdset(RRDCALC *rc, RRDSET *st) {
656 if( (rc->hash_chart == st->hash && !strcmp(rc->chart, st->id)) ||
657 (rc->hash_chart == st->hash_name && !strcmp(rc->chart, st->name)))
663 // this has to be called while the RRDHOST is locked
664 inline void rrdsetcalc_link_matching(RRDSET *st) {
665 // debug(D_HEALTH, "find matching alarms for chart '%s'", st->id);
668 for(rc = st->rrdhost->alarms; rc ; rc = rc->next) {
669 if(unlikely(rc->rrdset))
672 if(unlikely(rrdcalc_is_matching_this_rrdset(rc, st)))
673 rrdsetcalc_link(st, rc);
677 // this has to be called while the RRDHOST is locked
678 inline void rrdsetcalc_unlink(RRDCALC *rc) {
679 RRDSET *st = rc->rrdset;
682 debug(D_HEALTH, "Requested to unlink RRDCALC '%s.%s' which is not linked to any RRDSET", rc->chart?rc->chart:"NOCHART", rc->name);
683 error("Requested to unlink RRDCALC '%s.%s' which is not linked to any RRDSET", rc->chart?rc->chart:"NOCHART", rc->name);
688 time_t now = time(NULL);
689 health_alarm_log(st->rrdhost, rc->id, rc->next_event_id++, now, rc->name, rc->rrdset->id, rc->rrdset->family, rc->exec, rc->recipient, now - rc->last_status_change, rc->old_value, rc->value, rc->status, RRDCALC_STATUS_REMOVED, rc->source, rc->units, rc->info, 0);
692 RRDHOST *host = st->rrdhost;
694 debug(D_HEALTH, "Health unlinking alarm '%s.%s' from chart '%s' of host '%s'", rc->chart?rc->chart:"NOCHART", rc->name, st->id, host->hostname);
698 rc->rrdset_prev->rrdset_next = rc->rrdset_next;
701 rc->rrdset_next->rrdset_prev = rc->rrdset_prev;
704 st->alarms = rc->rrdset_next;
706 rc->rrdset_prev = rc->rrdset_next = NULL;
708 rrdvar_free(st->rrdhost, &st->variables_root_index, rc->local);
711 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rc->family);
714 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rc->hostid);
717 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rc->hostname);
722 // RRDCALC will remain in RRDHOST
723 // so that if the matching chart is found in the future
724 // it will be applied automatically
727 RRDCALC *rrdcalc_find(RRDSET *st, const char *name) {
729 uint32_t hash = simple_hash(name);
731 for( rc = st->alarms; rc ; rc = rc->rrdset_next ) {
732 if(unlikely(rc->hash == hash && !strcmp(rc->name, name)))
739 static inline int rrdcalc_exists(RRDHOST *host, const char *chart, const char *name, uint32_t hash_chart, uint32_t hash_name) {
742 if(unlikely(!chart)) {
743 error("attempt to find RRDCALC '%s' without giving a chart name", name);
747 if(unlikely(!hash_chart)) hash_chart = simple_hash(chart);
748 if(unlikely(!hash_name)) hash_name = simple_hash(name);
750 // make sure it does not already exist
751 for(rc = host->alarms; rc ; rc = rc->next) {
752 if (unlikely(rc->chart && rc->hash == hash_name && rc->hash_chart == hash_chart && !strcmp(name, rc->name) && !strcmp(chart, rc->chart))) {
753 debug(D_HEALTH, "Health alarm '%s.%s' already exists in host '%s'.", chart, name, host->hostname);
754 error("Health alarm '%s.%s' already exists in host '%s'.", chart, name, host->hostname);
762 static inline uint32_t rrdcalc_get_unique_id(RRDHOST *host, const char *chart, const char *name, uint32_t *next_event_id) {
764 uint32_t hash_chart = simple_hash(chart);
765 uint32_t hash_name = simple_hash(name);
767 // re-use old IDs, by looking them up in the alarm log
769 for(ae = host->health_log.alarms; ae ;ae = ae->next) {
770 if(unlikely(ae->hash_name == hash_name && ae->hash_chart == hash_chart && !strcmp(name, ae->name) && !strcmp(chart, ae->chart))) {
771 if(next_event_id) *next_event_id = ae->alarm_event_id + 1;
777 return host->health_log.next_alarm_id++;
780 static inline void rrdcalc_create_part2(RRDHOST *host, RRDCALC *rc) {
781 rrdhost_check_rdlock(host);
783 if(rc->calculation) {
784 rc->calculation->status = &rc->status;
785 rc->calculation->this = &rc->value;
786 rc->calculation->after = &rc->db_after;
787 rc->calculation->before = &rc->db_before;
788 rc->calculation->rrdcalc = rc;
792 rc->warning->status = &rc->status;
793 rc->warning->this = &rc->value;
794 rc->warning->after = &rc->db_after;
795 rc->warning->before = &rc->db_before;
796 rc->warning->rrdcalc = rc;
800 rc->critical->status = &rc->status;
801 rc->critical->this = &rc->value;
802 rc->critical->after = &rc->db_after;
803 rc->critical->before = &rc->db_before;
804 rc->critical->rrdcalc = rc;
807 // link it to the host
808 if(likely(host->alarms)) {
811 for(t = host->alarms; t && t->next ; t = t->next) ;
818 // link it to its chart
820 for(st = host->rrdset_root; st ; st = st->next) {
821 if(rrdcalc_is_matching_this_rrdset(rc, st)) {
822 rrdsetcalc_link(st, rc);
828 static inline RRDCALC *rrdcalc_create(RRDHOST *host, RRDCALCTEMPLATE *rt, const char *chart) {
830 debug(D_HEALTH, "Health creating dynamic alarm (from template) '%s.%s'", chart, rt->name);
832 if(rrdcalc_exists(host, chart, rt->name, 0, 0))
835 RRDCALC *rc = callocz(1, sizeof(RRDCALC));
836 rc->next_event_id = 1;
837 rc->id = rrdcalc_get_unique_id(host, chart, rt->name, &rc->next_event_id);
838 rc->name = strdupz(rt->name);
839 rc->hash = simple_hash(rc->name);
840 rc->chart = strdupz(chart);
841 rc->hash_chart = simple_hash(rc->chart);
843 if(rt->dimensions) rc->dimensions = strdupz(rt->dimensions);
845 rc->green = rt->green;
850 rc->delay_up_duration = rt->delay_up_duration;
851 rc->delay_down_duration = rt->delay_down_duration;
852 rc->delay_max_duration = rt->delay_max_duration;
853 rc->delay_multiplier = rt->delay_multiplier;
855 rc->group = rt->group;
856 rc->after = rt->after;
857 rc->before = rt->before;
858 rc->update_every = rt->update_every;
859 rc->options = rt->options;
861 if(rt->exec) rc->exec = strdupz(rt->exec);
862 if(rt->recipient) rc->recipient = strdupz(rt->recipient);
863 if(rt->source) rc->source = strdupz(rt->source);
864 if(rt->units) rc->units = strdupz(rt->units);
865 if(rt->info) rc->info = strdupz(rt->info);
867 if(rt->calculation) {
868 rc->calculation = expression_parse(rt->calculation->source, NULL, NULL);
870 error("Health alarm '%s.%s': failed to parse calculation expression '%s'", chart, rt->name, rt->calculation->source);
873 rc->warning = expression_parse(rt->warning->source, NULL, NULL);
875 error("Health alarm '%s.%s': failed to re-parse warning expression '%s'", chart, rt->name, rt->warning->source);
878 rc->critical = expression_parse(rt->critical->source, NULL, NULL);
880 error("Health alarm '%s.%s': failed to re-parse critical expression '%s'", chart, rt->name, rt->critical->source);
883 debug(D_HEALTH, "Health runtime added alarm '%s.%s': exec '%s', recipient '%s', green %Lf, red %Lf, lookup: group %d, after %d, before %d, options %u, dimensions '%s', update every %d, calculation '%s', warning '%s', critical '%s', source '%s', delay up %d, delay down %d, delay max %d, delay_multiplier %f",
884 (rc->chart)?rc->chart:"NOCHART",
886 (rc->exec)?rc->exec:"DEFAULT",
887 (rc->recipient)?rc->recipient:"DEFAULT",
894 (rc->dimensions)?rc->dimensions:"NONE",
896 (rc->calculation)?rc->calculation->parsed_as:"NONE",
897 (rc->warning)?rc->warning->parsed_as:"NONE",
898 (rc->critical)?rc->critical->parsed_as:"NONE",
900 rc->delay_up_duration,
901 rc->delay_down_duration,
902 rc->delay_max_duration,
906 rrdcalc_create_part2(host, rc);
910 void rrdcalc_free(RRDHOST *host, RRDCALC *rc) {
913 debug(D_HEALTH, "Health removing alarm '%s.%s' of host '%s'", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);
915 // unlink it from RRDSET
916 if(rc->rrdset) rrdsetcalc_unlink(rc);
918 // unlink it from RRDHOST
919 if(unlikely(rc == host->alarms))
920 host->alarms = rc->next;
922 else if(likely(host->alarms)) {
923 RRDCALC *t, *last = host->alarms;
924 for(t = last->next; t && t != rc; last = t, t = t->next) ;
926 last->next = rc->next;
928 error("Cannot unlink alarm '%s.%s' from host '%s': not found", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);
931 error("Cannot unlink unlink '%s.%s' from host '%s': This host does not have any calculations", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);
933 expression_free(rc->calculation);
934 expression_free(rc->warning);
935 expression_free(rc->critical);
940 freez(rc->dimensions);
942 freez(rc->recipient);
949 // ----------------------------------------------------------------------------
950 // RRDCALCTEMPLATE management
952 void rrdcalctemplate_link_matching(RRDSET *st) {
955 for(rt = st->rrdhost->templates; rt ; rt = rt->next) {
956 if(rt->hash_context == st->hash_context && !strcmp(rt->context, st->context)) {
957 RRDCALC *rc = rrdcalc_create(st->rrdhost, rt, st->id);
959 error("Health tried to create alarm from template '%s', but it failed", rt->name);
961 #ifdef NETDATA_INTERNAL_CHECKS
962 else if(rc->rrdset != st)
963 error("Health alarm '%s.%s' should be linked to chart '%s', but it is not", rc->chart?rc->chart:"NOCHART", rc->name, st->id);
969 static inline void rrdcalctemplate_free(RRDHOST *host, RRDCALCTEMPLATE *rt) {
970 debug(D_HEALTH, "Health removing template '%s' of host '%s'", rt->name, host->hostname);
972 if(host->templates) {
973 if(host->templates == rt) {
974 host->templates = rt->next;
977 RRDCALCTEMPLATE *t, *last = host->templates;
978 for (t = last->next; t && t != rt; last = t, t = t->next ) ;
979 if(last && last->next == rt) {
980 last->next = rt->next;
984 error("Cannot find RRDCALCTEMPLATE '%s' linked in host '%s'", rt->name, host->hostname);
988 expression_free(rt->calculation);
989 expression_free(rt->warning);
990 expression_free(rt->critical);
994 freez(rt->recipient);
999 freez(rt->dimensions);
1003 // ----------------------------------------------------------------------------
1004 // load health configuration
1006 #define HEALTH_CONF_MAX_LINE 4096
1008 #define HEALTH_ALARM_KEY "alarm"
1009 #define HEALTH_TEMPLATE_KEY "template"
1010 #define HEALTH_ON_KEY "on"
1011 #define HEALTH_LOOKUP_KEY "lookup"
1012 #define HEALTH_CALC_KEY "calc"
1013 #define HEALTH_EVERY_KEY "every"
1014 #define HEALTH_GREEN_KEY "green"
1015 #define HEALTH_RED_KEY "red"
1016 #define HEALTH_WARN_KEY "warn"
1017 #define HEALTH_CRIT_KEY "crit"
1018 #define HEALTH_EXEC_KEY "exec"
1019 #define HEALTH_RECIPIENT_KEY "to"
1020 #define HEALTH_UNITS_KEY "units"
1021 #define HEALTH_INFO_KEY "info"
1022 #define HEALTH_DELAY_KEY "delay"
1024 static inline int rrdcalc_add_alarm_from_config(RRDHOST *host, RRDCALC *rc) {
1026 error("Health configuration for alarm '%s' does not have a chart", rc->name);
1030 if(!rc->update_every) {
1031 error("Health configuration for alarm '%s.%s' has no frequency (parameter 'every'). Ignoring it.", rc->chart?rc->chart:"NOCHART", rc->name);
1035 if(!RRDCALC_HAS_DB_LOOKUP(rc) && !rc->warning && !rc->critical) {
1036 error("Health configuration for alarm '%s.%s' is useless (no calculation, no warning and no critical evaluation)", rc->chart?rc->chart:"NOCHART", rc->name);
1040 if (rrdcalc_exists(host, rc->chart, rc->name, rc->hash_chart, rc->hash))
1043 rc->id = rrdcalc_get_unique_id(&localhost, rc->chart, rc->name, &rc->next_event_id);
1045 debug(D_HEALTH, "Health configuration adding alarm '%s.%s' (%u): exec '%s', recipient '%s', green %Lf, red %Lf, lookup: group %d, after %d, before %d, options %u, dimensions '%s', update every %d, calculation '%s', warning '%s', critical '%s', source '%s', delay up %d, delay down %d, delay max %d, delay_multiplier %f",
1046 rc->chart?rc->chart:"NOCHART",
1049 (rc->exec)?rc->exec:"DEFAULT",
1050 (rc->recipient)?rc->recipient:"DEFAULT",
1057 (rc->dimensions)?rc->dimensions:"NONE",
1059 (rc->calculation)?rc->calculation->parsed_as:"NONE",
1060 (rc->warning)?rc->warning->parsed_as:"NONE",
1061 (rc->critical)?rc->critical->parsed_as:"NONE",
1063 rc->delay_up_duration,
1064 rc->delay_down_duration,
1065 rc->delay_max_duration,
1066 rc->delay_multiplier
1069 rrdcalc_create_part2(host, rc);
1073 static inline int rrdcalctemplate_add_template_from_config(RRDHOST *host, RRDCALCTEMPLATE *rt) {
1074 if(unlikely(!rt->context)) {
1075 error("Health configuration for template '%s' does not have a context", rt->name);
1079 if(unlikely(!rt->update_every)) {
1080 error("Health configuration for template '%s' has no frequency (parameter 'every'). Ignoring it.", rt->name);
1084 if(unlikely(!RRDCALCTEMPLATE_HAS_CALCULATION(rt) && !rt->warning && !rt->critical)) {
1085 error("Health configuration for template '%s' is useless (no calculation, no warning and no critical evaluation)", rt->name);
1089 RRDCALCTEMPLATE *t, *last = NULL;
1090 for (t = host->templates; t ; last = t, t = t->next) {
1091 if(unlikely(t->hash_name == rt->hash_name && !strcmp(t->name, rt->name))) {
1092 error("Health configuration template '%s' already exists for host '%s'.", rt->name, host->hostname);
1097 debug(D_HEALTH, "Health configuration adding template '%s': context '%s', exec '%s', recipient '%s', green %Lf, red %Lf, lookup: group %d, after %d, before %d, options %u, dimensions '%s', update every %d, calculation '%s', warning '%s', critical '%s', source '%s', delay up %d, delay down %d, delay max %d, delay_multiplier %f",
1099 (rt->context)?rt->context:"NONE",
1100 (rt->exec)?rt->exec:"DEFAULT",
1101 (rt->recipient)?rt->recipient:"DEFAULT",
1108 (rt->dimensions)?rt->dimensions:"NONE",
1110 (rt->calculation)?rt->calculation->parsed_as:"NONE",
1111 (rt->warning)?rt->warning->parsed_as:"NONE",
1112 (rt->critical)?rt->critical->parsed_as:"NONE",
1114 rt->delay_up_duration,
1115 rt->delay_down_duration,
1116 rt->delay_max_duration,
1117 rt->delay_multiplier
1124 rt->next = host->templates;
1125 host->templates = rt;
1131 static inline int health_parse_duration(char *string, int *result) {
1132 // make sure it is a number
1133 if(!*string || !(isdigit(*string) || *string == '+' || *string == '-')) {
1139 calculated_number n = strtold(string, &e);
1143 *result = (int) (n * 86400 * 365);
1146 *result = (int) (n * 86400 * 30);
1149 *result = (int) (n * 86400 * 7);
1152 *result = (int) (n * 86400);
1155 *result = (int) (n * 3600);
1158 *result = (int) (n * 60);
1163 *result = (int) (n);
1173 static inline int health_parse_delay(
1174 size_t line, const char *path, const char *file, char *string,
1175 int *delay_up_duration,
1176 int *delay_down_duration,
1177 int *delay_max_duration,
1178 float *delay_multiplier) {
1181 char given_down = 0;
1183 char given_multiplier = 0;
1189 while(*s && !isspace(*s)) s++;
1190 while(*s && isspace(*s)) *s++ = '\0';
1195 while(*s && !isspace(*s)) s++;
1196 while(*s && isspace(*s)) *s++ = '\0';
1198 if(!strcasecmp(key, "up")) {
1199 if (!health_parse_duration(value, delay_up_duration)) {
1200 error("Health configuration at line %zu of file '%s/%s': invalid value '%s' for '%s' keyword",
1201 line, path, file, value, key);
1205 else if(!strcasecmp(key, "down")) {
1206 if (!health_parse_duration(value, delay_down_duration)) {
1207 error("Health configuration at line %zu of file '%s/%s': invalid value '%s' for '%s' keyword",
1208 line, path, file, value, key);
1210 else given_down = 1;
1212 else if(!strcasecmp(key, "multiplier")) {
1213 *delay_multiplier = strtof(value, NULL);
1214 if(isnan(*delay_multiplier) || isinf(*delay_multiplier) || islessequal(*delay_multiplier, 0)) {
1215 error("Health configuration at line %zu of file '%s/%s': invalid value '%s' for '%s' keyword",
1216 line, path, file, value, key);
1218 else given_multiplier = 1;
1220 else if(!strcasecmp(key, "max")) {
1221 if (!health_parse_duration(value, delay_max_duration)) {
1222 error("Health configuration at line %zu of file '%s/%s': invalid value '%s' for '%s' keyword",
1223 line, path, file, value, key);
1228 error("Health configuration at line %zu of file '%s/%s': unknown keyword '%s'",
1229 line, path, file, key);
1234 *delay_up_duration = 0;
1237 *delay_down_duration = 0;
1239 if(!given_multiplier)
1240 *delay_multiplier = 1.0;
1243 if((*delay_max_duration) < (*delay_up_duration) * (*delay_multiplier))
1244 *delay_max_duration = (*delay_up_duration) * (*delay_multiplier);
1246 if((*delay_max_duration) < (*delay_down_duration) * (*delay_multiplier))
1247 *delay_max_duration = (*delay_down_duration) * (*delay_multiplier);
1253 static inline int health_parse_db_lookup(
1254 size_t line, const char *path, const char *file, char *string,
1255 int *group_method, int *after, int *before, int *every,
1256 uint32_t *options, char **dimensions
1258 debug(D_HEALTH, "Health configuration parsing database lookup %zu@%s/%s: %s", line, path, file, string);
1260 if(*dimensions) freez(*dimensions);
1267 char *s = string, *key;
1269 // first is the group method
1271 while(*s && !isspace(*s)) s++;
1272 while(*s && isspace(*s)) *s++ = '\0';
1274 error("Health configuration invalid chart calculation at line %zu of file '%s/%s': expected group method followed by the 'after' time, but got '%s'",
1275 line, path, file, key);
1279 if((*group_method = web_client_api_request_v1_data_group(key, -1)) == -1) {
1280 error("Health configuration at line %zu of file '%s/%s': invalid group method '%s'",
1281 line, path, file, key);
1285 // then is the 'after' time
1287 while(*s && !isspace(*s)) s++;
1288 while(*s && isspace(*s)) *s++ = '\0';
1290 if(!health_parse_duration(key, after)) {
1291 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' after group method",
1292 line, path, file, key);
1297 *every = abs(*after);
1299 // now we may have optional parameters
1302 while(*s && !isspace(*s)) s++;
1303 while(*s && isspace(*s)) *s++ = '\0';
1306 if(!strcasecmp(key, "at")) {
1308 while(*s && !isspace(*s)) s++;
1309 while(*s && isspace(*s)) *s++ = '\0';
1311 if (!health_parse_duration(value, before)) {
1312 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' for '%s' keyword",
1313 line, path, file, value, key);
1316 else if(!strcasecmp(key, HEALTH_EVERY_KEY)) {
1318 while(*s && !isspace(*s)) s++;
1319 while(*s && isspace(*s)) *s++ = '\0';
1321 if (!health_parse_duration(value, every)) {
1322 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' for '%s' keyword",
1323 line, path, file, value, key);
1326 else if(!strcasecmp(key, "absolute") || !strcasecmp(key, "abs") || !strcasecmp(key, "absolute_sum")) {
1327 *options |= RRDR_OPTION_ABSOLUTE;
1329 else if(!strcasecmp(key, "min2max")) {
1330 *options |= RRDR_OPTION_MIN2MAX;
1332 else if(!strcasecmp(key, "null2zero")) {
1333 *options |= RRDR_OPTION_NULL2ZERO;
1335 else if(!strcasecmp(key, "percentage")) {
1336 *options |= RRDR_OPTION_PERCENTAGE;
1338 else if(!strcasecmp(key, "unaligned")) {
1339 *options |= RRDR_OPTION_NOT_ALIGNED;
1341 else if(!strcasecmp(key, "of")) {
1342 if(*s && strcasecmp(s, "all"))
1343 *dimensions = strdupz(s);
1347 error("Health configuration at line %zu of file '%s/%s': unknown keyword '%s'",
1348 line, path, file, key);
1355 static inline char *health_source_file(size_t line, const char *path, const char *filename) {
1356 char buffer[FILENAME_MAX + 1];
1357 snprintfz(buffer, FILENAME_MAX, "%zu@%s/%s", line, path, filename);
1358 return strdupz(buffer);
1361 static inline void strip_quotes(char *s) {
1363 if(*s == '\'' || *s == '"') *s = ' ';
1368 int health_readfile(const char *path, const char *filename) {
1369 debug(D_HEALTH, "Health configuration reading file '%s/%s'", path, filename);
1371 static uint32_t hash_alarm = 0, hash_template = 0, hash_on = 0, hash_calc = 0, hash_green = 0, hash_red = 0, hash_warn = 0, hash_crit = 0, hash_exec = 0, hash_every = 0, hash_lookup = 0, hash_units = 0, hash_info = 0, hash_recipient = 0, hash_delay = 0;
1372 char buffer[HEALTH_CONF_MAX_LINE + 1];
1374 if(unlikely(!hash_alarm)) {
1375 hash_alarm = simple_uhash(HEALTH_ALARM_KEY);
1376 hash_template = simple_uhash(HEALTH_TEMPLATE_KEY);
1377 hash_on = simple_uhash(HEALTH_ON_KEY);
1378 hash_calc = simple_uhash(HEALTH_CALC_KEY);
1379 hash_lookup = simple_uhash(HEALTH_LOOKUP_KEY);
1380 hash_green = simple_uhash(HEALTH_GREEN_KEY);
1381 hash_red = simple_uhash(HEALTH_RED_KEY);
1382 hash_warn = simple_uhash(HEALTH_WARN_KEY);
1383 hash_crit = simple_uhash(HEALTH_CRIT_KEY);
1384 hash_exec = simple_uhash(HEALTH_EXEC_KEY);
1385 hash_every = simple_uhash(HEALTH_EVERY_KEY);
1386 hash_units = simple_hash(HEALTH_UNITS_KEY);
1387 hash_info = simple_hash(HEALTH_INFO_KEY);
1388 hash_recipient = simple_hash(HEALTH_RECIPIENT_KEY);
1389 hash_delay = simple_uhash(HEALTH_DELAY_KEY);
1392 snprintfz(buffer, HEALTH_CONF_MAX_LINE, "%s/%s", path, filename);
1393 FILE *fp = fopen(buffer, "r");
1395 error("Health configuration cannot read file '%s'.", buffer);
1400 RRDCALCTEMPLATE *rt = NULL;
1402 size_t line = 0, append = 0;
1404 while((s = fgets(&buffer[append], (int)(HEALTH_CONF_MAX_LINE - append), fp)) || append) {
1405 int stop_appending = !s;
1407 // info("Line %zu of file '%s/%s': '%s'", line, path, filename, s);
1410 // info("Trimmed line %zu of file '%s/%s': '%s'", line, path, filename, s);
1413 if(!stop_appending && s[append - 1] == '\\') {
1414 s[append - 1] = ' ';
1415 append = &s[append] - buffer;
1416 if(append < HEALTH_CONF_MAX_LINE)
1419 error("Health configuration has too long muli-line at line %zu of file '%s/%s'.", line, path, filename);
1425 while(*s && *s != ':') s++;
1427 error("Health configuration has invalid line %zu of file '%s/%s'. It does not contain a ':'. Ignoring it.", line, path, filename);
1435 value = trim(value);
1438 error("Health configuration has invalid line %zu of file '%s/%s'. Keyword is empty. Ignoring it.", line, path, filename);
1443 error("Health configuration has invalid line %zu of file '%s/%s'. value is empty. Ignoring it.", line, path, filename);
1447 // info("Health file '%s/%s', key '%s', value '%s'", path, filename, key, value);
1448 uint32_t hash = simple_uhash(key);
1450 if(hash == hash_alarm && !strcasecmp(key, HEALTH_ALARM_KEY)) {
1451 if(rc && !rrdcalc_add_alarm_from_config(&localhost, rc))
1452 rrdcalc_free(&localhost, rc);
1455 if (!rrdcalctemplate_add_template_from_config(&localhost, rt))
1456 rrdcalctemplate_free(&localhost, rt);
1460 rc = callocz(1, sizeof(RRDCALC));
1461 rc->next_event_id = 1;
1462 rc->name = strdupz(value);
1463 rc->hash = simple_hash(rc->name);
1464 rc->source = health_source_file(line, path, filename);
1468 rc->old_value = NAN;
1469 rc->delay_multiplier = 1.0;
1471 if(rrdvar_fix_name(rc->name))
1472 error("Health configuration renamed alarm '%s' to '%s'", value, rc->name);
1474 else if(hash == hash_template && !strcasecmp(key, HEALTH_TEMPLATE_KEY)) {
1476 if(!rrdcalc_add_alarm_from_config(&localhost, rc))
1477 rrdcalc_free(&localhost, rc);
1481 if(rt && !rrdcalctemplate_add_template_from_config(&localhost, rt))
1482 rrdcalctemplate_free(&localhost, rt);
1484 rt = callocz(1, sizeof(RRDCALCTEMPLATE));
1485 rt->name = strdupz(value);
1486 rt->hash_name = simple_hash(rt->name);
1487 rt->source = health_source_file(line, path, filename);
1490 rt->delay_multiplier = 1.0;
1492 if(rrdvar_fix_name(rt->name))
1493 error("Health configuration renamed template '%s' to '%s'", value, rt->name);
1496 if(hash == hash_on && !strcasecmp(key, HEALTH_ON_KEY)) {
1498 if(strcmp(rc->chart, value))
1499 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' has key '%s' twice, once with value '%s' and later with value '%s'. Using ('%s').",
1500 line, path, filename, rc->name, key, rc->chart, value, value);
1504 rc->chart = strdupz(value);
1505 rc->hash_chart = simple_hash(rc->chart);
1507 else if(hash == hash_lookup && !strcasecmp(key, HEALTH_LOOKUP_KEY)) {
1508 health_parse_db_lookup(line, path, filename, value, &rc->group, &rc->after, &rc->before,
1510 &rc->options, &rc->dimensions);
1512 else if(hash == hash_every && !strcasecmp(key, HEALTH_EVERY_KEY)) {
1513 if(!health_parse_duration(value, &rc->update_every))
1514 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' cannot parse duration: '%s'.",
1515 line, path, filename, rc->name, key, value);
1517 else if(hash == hash_green && !strcasecmp(key, HEALTH_GREEN_KEY)) {
1519 rc->green = strtold(value, &e);
1521 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' leaves this string unmatched: '%s'.",
1522 line, path, filename, rc->name, key, e);
1525 else if(hash == hash_red && !strcasecmp(key, HEALTH_RED_KEY)) {
1527 rc->red = strtold(value, &e);
1529 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' leaves this string unmatched: '%s'.",
1530 line, path, filename, rc->name, key, e);
1533 else if(hash == hash_calc && !strcasecmp(key, HEALTH_CALC_KEY)) {
1534 const char *failed_at = NULL;
1536 rc->calculation = expression_parse(value, &failed_at, &error);
1537 if(!rc->calculation) {
1538 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1539 line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
1542 else if(hash == hash_warn && !strcasecmp(key, HEALTH_WARN_KEY)) {
1543 const char *failed_at = NULL;
1545 rc->warning = expression_parse(value, &failed_at, &error);
1547 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1548 line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
1551 else if(hash == hash_crit && !strcasecmp(key, HEALTH_CRIT_KEY)) {
1552 const char *failed_at = NULL;
1554 rc->critical = expression_parse(value, &failed_at, &error);
1556 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1557 line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
1560 else if(hash == hash_exec && !strcasecmp(key, HEALTH_EXEC_KEY)) {
1562 if(strcmp(rc->exec, value))
1563 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' has key '%s' twice, once with value '%s' and later with value '%s'. Using ('%s').",
1564 line, path, filename, rc->name, key, rc->exec, value, value);
1568 rc->exec = strdupz(value);
1570 else if(hash == hash_recipient && !strcasecmp(key, HEALTH_RECIPIENT_KEY)) {
1572 if(strcmp(rc->recipient, value))
1573 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' has key '%s' twice, once with value '%s' and later with value '%s'. Using ('%s').",
1574 line, path, filename, rc->name, key, rc->recipient, value, value);
1576 freez(rc->recipient);
1578 rc->recipient = strdupz(value);
1580 else if(hash == hash_units && !strcasecmp(key, HEALTH_UNITS_KEY)) {
1582 if(strcmp(rc->units, value))
1583 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' has key '%s' twice, once with value '%s' and later with value '%s'. Using ('%s').",
1584 line, path, filename, rc->name, key, rc->units, value, value);
1588 rc->units = strdupz(value);
1589 strip_quotes(rc->units);
1591 else if(hash == hash_info && !strcasecmp(key, HEALTH_INFO_KEY)) {
1593 if(strcmp(rc->info, value))
1594 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' has key '%s' twice, once with value '%s' and later with value '%s'. Using ('%s').",
1595 line, path, filename, rc->name, key, rc->info, value, value);
1599 rc->info = strdupz(value);
1600 strip_quotes(rc->info);
1602 else if(hash == hash_delay && !strcasecmp(key, HEALTH_DELAY_KEY)) {
1603 health_parse_delay(line, path, filename, value, &rc->delay_up_duration, &rc->delay_down_duration, &rc->delay_max_duration, &rc->delay_multiplier);
1606 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' has unknown key '%s'.",
1607 line, path, filename, rc->name, key);
1611 if(hash == hash_on && !strcasecmp(key, HEALTH_ON_KEY)) {
1613 if(strcmp(rt->context, value))
1614 info("Health configuration at line %zu of file '%s/%s' for template '%s' has key '%s' twice, once with value '%s' and later with value '%s'. Using ('%s').",
1615 line, path, filename, rt->name, key, rt->context, value, value);
1619 rt->context = strdupz(value);
1620 rt->hash_context = simple_hash(rt->context);
1622 else if(hash == hash_lookup && !strcasecmp(key, HEALTH_LOOKUP_KEY)) {
1623 health_parse_db_lookup(line, path, filename, value, &rt->group, &rt->after, &rt->before,
1625 &rt->options, &rt->dimensions);
1627 else if(hash == hash_every && !strcasecmp(key, HEALTH_EVERY_KEY)) {
1628 if(!health_parse_duration(value, &rt->update_every))
1629 info("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' cannot parse duration: '%s'.",
1630 line, path, filename, rt->name, key, value);
1632 else if(hash == hash_green && !strcasecmp(key, HEALTH_GREEN_KEY)) {
1634 rt->green = strtold(value, &e);
1636 info("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' leaves this string unmatched: '%s'.",
1637 line, path, filename, rt->name, key, e);
1640 else if(hash == hash_red && !strcasecmp(key, HEALTH_RED_KEY)) {
1642 rt->red = strtold(value, &e);
1644 info("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' leaves this string unmatched: '%s'.",
1645 line, path, filename, rt->name, key, e);
1648 else if(hash == hash_calc && !strcasecmp(key, HEALTH_CALC_KEY)) {
1649 const char *failed_at = NULL;
1651 rt->calculation = expression_parse(value, &failed_at, &error);
1652 if(!rt->calculation) {
1653 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1654 line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
1657 else if(hash == hash_warn && !strcasecmp(key, HEALTH_WARN_KEY)) {
1658 const char *failed_at = NULL;
1660 rt->warning = expression_parse(value, &failed_at, &error);
1662 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1663 line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
1666 else if(hash == hash_crit && !strcasecmp(key, HEALTH_CRIT_KEY)) {
1667 const char *failed_at = NULL;
1669 rt->critical = expression_parse(value, &failed_at, &error);
1671 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1672 line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
1675 else if(hash == hash_exec && !strcasecmp(key, HEALTH_EXEC_KEY)) {
1677 if(strcmp(rt->exec, value))
1678 info("Health configuration at line %zu of file '%s/%s' for template '%s' has key '%s' twice, once with value '%s' and later with value '%s'. Using ('%s').",
1679 line, path, filename, rt->name, key, rt->exec, value, value);
1683 rt->exec = strdupz(value);
1685 else if(hash == hash_recipient && !strcasecmp(key, HEALTH_RECIPIENT_KEY)) {
1687 if(strcmp(rt->recipient, value))
1688 info("Health configuration at line %zu of file '%s/%s' for template '%s' has key '%s' twice, once with value '%s' and later with value '%s'. Using ('%s').",
1689 line, path, filename, rt->name, key, rt->recipient, value, value);
1691 freez(rt->recipient);
1693 rt->recipient = strdupz(value);
1695 else if(hash == hash_units && !strcasecmp(key, HEALTH_UNITS_KEY)) {
1697 if(strcmp(rt->units, value))
1698 info("Health configuration at line %zu of file '%s/%s' for template '%s' has key '%s' twice, once with value '%s' and later with value '%s'. Using ('%s').",
1699 line, path, filename, rt->name, key, rt->units, value, value);
1703 rt->units = strdupz(value);
1704 strip_quotes(rt->units);
1706 else if(hash == hash_info && !strcasecmp(key, HEALTH_INFO_KEY)) {
1708 if(strcmp(rt->info, value))
1709 info("Health configuration at line %zu of file '%s/%s' for template '%s' has key '%s' twice, once with value '%s' and later with value '%s'. Using ('%s').",
1710 line, path, filename, rt->name, key, rt->info, value, value);
1714 rt->info = strdupz(value);
1715 strip_quotes(rt->info);
1717 else if(hash == hash_delay && !strcasecmp(key, HEALTH_DELAY_KEY)) {
1718 health_parse_delay(line, path, filename, value, &rt->delay_up_duration, &rt->delay_down_duration, &rt->delay_max_duration, &rt->delay_multiplier);
1721 error("Health configuration at line %zu of file '%s/%s' for template '%s' has unknown key '%s'.",
1722 line, path, filename, rt->name, key);
1726 error("Health configuration at line %zu of file '%s/%s' has unknown key '%s'. Expected either '" HEALTH_ALARM_KEY "' or '" HEALTH_TEMPLATE_KEY "'.",
1727 line, path, filename, key);
1731 if(rc && !rrdcalc_add_alarm_from_config(&localhost, rc))
1732 rrdcalc_free(&localhost, rc);
1734 if(rt && !rrdcalctemplate_add_template_from_config(&localhost, rt))
1735 rrdcalctemplate_free(&localhost, rt);
1741 void health_readdir(const char *path) {
1742 size_t pathlen = strlen(path);
1744 debug(D_HEALTH, "Health configuration reading directory '%s'", path);
1746 DIR *dir = opendir(path);
1748 error("Health configuration cannot open directory '%s'.", path);
1752 struct dirent *de = NULL;
1753 while ((de = readdir(dir))) {
1754 size_t len = strlen(de->d_name);
1756 if(de->d_type == DT_DIR
1758 (de->d_name[0] == '.' && de->d_name[1] == '\0')
1759 || (de->d_name[0] == '.' && de->d_name[1] == '.' && de->d_name[2] == '\0')
1761 debug(D_HEALTH, "Ignoring directory '%s'", de->d_name);
1765 else if(de->d_type == DT_DIR) {
1766 char *s = mallocz(pathlen + strlen(de->d_name) + 2);
1769 strcat(s, de->d_name);
1775 else if((de->d_type == DT_LNK || de->d_type == DT_REG || de->d_type == DT_UNKNOWN) &&
1776 len > 5 && !strcmp(&de->d_name[len - 5], ".conf")) {
1777 health_readfile(path, de->d_name);
1780 else debug(D_HEALTH, "Ignoring file '%s'", de->d_name);
1786 static inline char *health_config_dir(void) {
1787 char buffer[FILENAME_MAX + 1];
1788 snprintfz(buffer, FILENAME_MAX, "%s/health.d", config_get("global", "config directory", CONFIG_DIR));
1789 return config_get("health", "health configuration directory", buffer);
1792 void health_init(void) {
1793 debug(D_HEALTH, "Health configuration initializing");
1795 if(!(health_enabled = config_get_boolean("health", "enabled", 1))) {
1796 debug(D_HEALTH, "Health is disabled.");
1800 health_alarm_log_load(&localhost);
1802 char *path = health_config_dir();
1805 char buffer[FILENAME_MAX + 1];
1806 snprintfz(buffer, FILENAME_MAX, "%s/alarm-notify.sh", config_get("global", "plugins directory", PLUGINS_DIR));
1807 health.health_default_exec = config_get("health", "script to execute on alarm", buffer);
1810 long n = config_get_number("health", "in memory max health log entries", (long)localhost.health_log.max);
1812 error("Health configuration has invalid max log entries %ld. Using default %u", n, localhost.health_log.max);
1813 config_set_number("health", "in memory max health log entries", (long)localhost.health_log.max);
1815 else localhost.health_log.max = (unsigned int)n;
1817 rrdhost_rwlock(&localhost);
1818 health_readdir(path);
1819 rrdhost_unlock(&localhost);
1822 // ----------------------------------------------------------------------------
1825 static inline void health_string2json(BUFFER *wb, const char *prefix, const char *label, const char *value, const char *suffix) {
1827 buffer_sprintf(wb, "%s\"%s\":\"%s\"%s", prefix, label, value, suffix);
1829 buffer_sprintf(wb, "%s\"%s\":null%s", prefix, label, suffix);
1832 static inline void health_alarm_entry2json_nolock(BUFFER *wb, ALARM_ENTRY *ae, RRDHOST *host) {
1833 buffer_sprintf(wb, "\n\t{\n"
1834 "\t\t\"hostname\": \"%s\",\n"
1835 "\t\t\"unique_id\": %u,\n"
1836 "\t\t\"alarm_id\": %u,\n"
1837 "\t\t\"alarm_event_id\": %u,\n"
1838 "\t\t\"name\": \"%s\",\n"
1839 "\t\t\"chart\": \"%s\",\n"
1840 "\t\t\"family\": \"%s\",\n"
1841 "\t\t\"processed\": %s,\n"
1842 "\t\t\"updated\": %s,\n"
1843 "\t\t\"exec_run\": %lu,\n"
1844 "\t\t\"exec_failed\": %s,\n"
1845 "\t\t\"exec\": \"%s\",\n"
1846 "\t\t\"recipient\": \"%s\",\n"
1847 "\t\t\"exec_code\": %d,\n"
1848 "\t\t\"source\": \"%s\",\n"
1849 "\t\t\"units\": \"%s\",\n"
1850 "\t\t\"info\": \"%s\",\n"
1851 "\t\t\"when\": %lu,\n"
1852 "\t\t\"duration\": %lu,\n"
1853 "\t\t\"non_clear_duration\": %lu,\n"
1854 "\t\t\"status\": \"%s\",\n"
1855 "\t\t\"old_status\": \"%s\",\n"
1856 "\t\t\"delay\": %d,\n"
1857 "\t\t\"delay_up_to_timestamp\": %lu,\n",
1865 (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_PROCESSED)?"true":"false",
1866 (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_UPDATED)?"true":"false",
1867 (unsigned long)ae->exec_run_timestamp,
1868 (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_EXEC_FAILED)?"true":"false",
1869 ae->exec?ae->exec:health.health_default_exec,
1870 ae->recipient?ae->recipient:health.health_default_recipient,
1873 ae->units?ae->units:"",
1874 ae->info?ae->info:"",
1875 (unsigned long)ae->when,
1876 (unsigned long)ae->duration,
1877 (unsigned long)ae->non_clear_duration,
1878 rrdcalc_status2string(ae->new_status),
1879 rrdcalc_status2string(ae->old_status),
1881 (unsigned long)ae->delay_up_to_timestamp
1884 buffer_strcat(wb, "\t\t\"value\":");
1885 buffer_rrd_value(wb, ae->new_value);
1886 buffer_strcat(wb, ",\n");
1888 buffer_strcat(wb, "\t\t\"old_value\":");
1889 buffer_rrd_value(wb, ae->old_value);
1890 buffer_strcat(wb, "\n");
1892 buffer_strcat(wb, "\t}");
1895 void health_alarm_log2json(RRDHOST *host, BUFFER *wb, uint32_t after) {
1896 pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
1898 buffer_strcat(wb, "[");
1900 unsigned int max = host->health_log.max;
1901 unsigned int count = 0;
1903 for(ae = host->health_log.alarms; ae && count < max ; count++, ae = ae->next) {
1904 if(ae->unique_id > after) {
1905 if(likely(count)) buffer_strcat(wb, ",");
1906 health_alarm_entry2json_nolock(wb, ae, host);
1910 buffer_strcat(wb, "\n]\n");
1912 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1915 static inline void health_rrdcalc2json_nolock(BUFFER *wb, RRDCALC *rc) {
1917 "\t\t\"%s.%s\": {\n"
1918 "\t\t\t\"id\": %lu,\n"
1919 "\t\t\t\"name\": \"%s\",\n"
1920 "\t\t\t\"chart\": \"%s\",\n"
1921 "\t\t\t\"family\": \"%s\",\n"
1922 "\t\t\t\"active\": %s,\n"
1923 "\t\t\t\"exec\": \"%s\",\n"
1924 "\t\t\t\"recipient\": \"%s\",\n"
1925 "\t\t\t\"source\": \"%s\",\n"
1926 "\t\t\t\"units\": \"%s\",\n"
1927 "\t\t\t\"info\": \"%s\",\n"
1928 "\t\t\t\"status\": \"%s\",\n"
1929 "\t\t\t\"last_status_change\": %lu,\n"
1930 "\t\t\t\"last_updated\": %lu,\n"
1931 "\t\t\t\"next_update\": %lu,\n"
1932 "\t\t\t\"update_every\": %d,\n"
1933 "\t\t\t\"delay_up_duration\": %d,\n"
1934 "\t\t\t\"delay_down_duration\": %d,\n"
1935 "\t\t\t\"delay_max_duration\": %d,\n"
1936 "\t\t\t\"delay_multiplier\": %f,\n"
1937 "\t\t\t\"delay\": %d,\n"
1938 "\t\t\t\"delay_up_to_timestamp\": %lu,\n"
1939 , rc->chart, rc->name
1940 , (unsigned long)rc->id
1943 , (rc->rrdset && rc->rrdset->family)?rc->rrdset->family:""
1944 , (rc->rrdset)?"true":"false"
1945 , rc->exec?rc->exec:health.health_default_exec
1946 , rc->recipient?rc->recipient:health.health_default_recipient
1948 , rc->units?rc->units:""
1949 , rc->info?rc->info:""
1950 , rrdcalc_status2string(rc->status)
1951 , (unsigned long)rc->last_status_change
1952 , (unsigned long)rc->last_updated
1953 , (unsigned long)rc->next_update
1955 , rc->delay_up_duration
1956 , rc->delay_down_duration
1957 , rc->delay_max_duration
1958 , rc->delay_multiplier
1960 , (unsigned long)rc->delay_up_to_timestamp
1963 if(RRDCALC_HAS_DB_LOOKUP(rc)) {
1964 if(rc->dimensions && *rc->dimensions)
1965 health_string2json(wb, "\t\t\t", "lookup_dimensions", rc->dimensions, ",\n");
1968 "\t\t\t\"db_after\": %lu,\n"
1969 "\t\t\t\"db_before\": %lu,\n"
1970 "\t\t\t\"lookup_method\": \"%s\",\n"
1971 "\t\t\t\"lookup_after\": %d,\n"
1972 "\t\t\t\"lookup_before\": %d,\n"
1973 "\t\t\t\"lookup_options\": \"",
1974 (unsigned long) rc->db_after,
1975 (unsigned long) rc->db_before,
1976 group_method2string(rc->group),
1980 buffer_data_options2string(wb, rc->options);
1981 buffer_strcat(wb, "\",\n");
1984 if(rc->calculation) {
1985 health_string2json(wb, "\t\t\t", "calc", rc->calculation->source, ",\n");
1986 health_string2json(wb, "\t\t\t", "calc_parsed", rc->calculation->parsed_as, ",\n");
1990 health_string2json(wb, "\t\t\t", "warn", rc->warning->source, ",\n");
1991 health_string2json(wb, "\t\t\t", "warn_parsed", rc->warning->parsed_as, ",\n");
1995 health_string2json(wb, "\t\t\t", "crit", rc->critical->source, ",\n");
1996 health_string2json(wb, "\t\t\t", "crit_parsed", rc->critical->parsed_as, ",\n");
1999 buffer_strcat(wb, "\t\t\t\"green\":");
2000 buffer_rrd_value(wb, rc->green);
2001 buffer_strcat(wb, ",\n");
2003 buffer_strcat(wb, "\t\t\t\"red\":");
2004 buffer_rrd_value(wb, rc->red);
2005 buffer_strcat(wb, ",\n");
2007 buffer_strcat(wb, "\t\t\t\"value\":");
2008 buffer_rrd_value(wb, rc->value);
2009 buffer_strcat(wb, "\n");
2011 buffer_strcat(wb, "\t\t}");
2014 //void health_rrdcalctemplate2json_nolock(BUFFER *wb, RRDCALCTEMPLATE *rt) {
2018 void health_alarms2json(RRDHOST *host, BUFFER *wb, int all) {
2021 rrdhost_rdlock(&localhost);
2022 buffer_sprintf(wb, "{\n\t\"hostname\": \"%s\","
2023 "\n\t\"latest_alarm_log_unique_id\": %u,"
2024 "\n\t\"status\": %s,"
2026 "\n\t\"alarms\": {\n",
2028 (host->health_log.next_log_id > 0)?(host->health_log.next_log_id - 1):0,
2029 health_enabled?"true":"false",
2030 (unsigned long)time(NULL));
2033 for(i = 0, rc = host->alarms; rc ; rc = rc->next) {
2034 if(unlikely(!rc->rrdset || !rc->rrdset->last_collected_time.tv_sec))
2037 if(likely(!all && !(rc->status == RRDCALC_STATUS_WARNING || rc->status == RRDCALC_STATUS_CRITICAL)))
2040 if(likely(i)) buffer_strcat(wb, ",\n");
2041 health_rrdcalc2json_nolock(wb, rc);
2045 // buffer_strcat(wb, "\n\t},\n\t\"templates\": {");
2046 // RRDCALCTEMPLATE *rt;
2047 // for(rt = host->templates; rt ; rt = rt->next)
2048 // health_rrdcalctemplate2json_nolock(wb, rt);
2050 buffer_strcat(wb, "\n\t}\n}\n");
2051 rrdhost_unlock(&localhost);
2055 // ----------------------------------------------------------------------------
2056 // re-load health configuration
2058 static inline void health_free_all_nolock(RRDHOST *host) {
2059 while(host->templates)
2060 rrdcalctemplate_free(host, host->templates);
2063 rrdcalc_free(host, host->alarms);
2066 void health_reload(void) {
2067 if(!health_enabled) {
2068 error("Health reload is requested, but health is not enabled.");
2072 char *path = health_config_dir();
2074 // free all running alarms
2075 rrdhost_rwlock(&localhost);
2076 health_free_all_nolock(&localhost);
2077 rrdhost_unlock(&localhost);
2079 // invalidate all previous entries in the alarm log
2081 for(t = localhost.health_log.alarms ; t ; t = t->next) {
2082 if(t->new_status != RRDCALC_STATUS_REMOVED)
2083 t->notifications |= HEALTH_ENTRY_NOTIFICATIONS_UPDATED;
2086 // reset all thresholds to all charts
2088 for(st = localhost.rrdset_root; st ; st = st->next) {
2093 // load the new alarms
2094 rrdhost_rwlock(&localhost);
2095 health_readdir(path);
2096 rrdhost_unlock(&localhost);
2098 // link the loaded alarms to their charts
2099 for(st = localhost.rrdset_root; st ; st = st->next) {
2100 rrdhost_rwlock(&localhost);
2102 rrdsetcalc_link_matching(st);
2103 rrdcalctemplate_link_matching(st);
2105 rrdhost_unlock(&localhost);
2109 // ----------------------------------------------------------------------------
2110 // health main thread and friends
2112 static inline int rrdcalc_value2status(calculated_number n) {
2113 if(isnan(n)) return RRDCALC_STATUS_UNDEFINED;
2114 if(n) return RRDCALC_STATUS_RAISED;
2115 return RRDCALC_STATUS_CLEAR;
2118 static inline void health_alarm_execute(RRDHOST *host, ALARM_ENTRY *ae) {
2119 ae->notifications |= HEALTH_ENTRY_NOTIFICATIONS_PROCESSED;
2121 // find the previous notification for the same alarm
2123 for(t = ae->next; t ;t = t->next) {
2124 if(t->alarm_id == ae->alarm_id && t->notifications & HEALTH_ENTRY_NOTIFICATIONS_EXEC_RUN)
2128 if(t && t->new_status == ae->new_status) {
2129 // don't send the same notification again
2130 info("Health not sending again notification for alarm '%s.%s' status %s", ae->chart, ae->name, rrdcalc_status2string(ae->new_status));
2134 if((ae->old_status == RRDCALC_STATUS_UNDEFINED && ae->new_status == RRDCALC_STATUS_UNINITIALIZED)
2135 || (ae->old_status == RRDCALC_STATUS_UNINITIALIZED && ae->new_status == RRDCALC_STATUS_CLEAR)) {
2136 info("Health not sending notification for first initialization of alarm '%s.%s' status %s", ae->chart, ae->name, rrdcalc_status2string(ae->new_status));
2140 char buffer[FILENAME_MAX + 1];
2143 const char *exec = ae->exec;
2144 if(!exec) exec = health.health_default_exec;
2146 const char *recipient = ae->recipient;
2147 if(!recipient) recipient = health.health_default_recipient;
2149 snprintfz(buffer, FILENAME_MAX, "exec %s '%s' '%s' '%u' '%u' '%u' '%lu' '%s' '%s' '%s' '%s' '%s' '%0.0Lf' '%0.0Lf' '%s' '%u' '%u' '%s' '%s'",
2156 (unsigned long)ae->when,
2158 ae->chart?ae->chart:"NOCAHRT",
2159 ae->family?ae->family:"NOFAMILY",
2160 rrdcalc_status2string(ae->new_status),
2161 rrdcalc_status2string(ae->old_status),
2164 ae->source?ae->source:"UNKNOWN",
2165 (uint32_t)ae->duration,
2166 (uint32_t)ae->non_clear_duration,
2167 ae->units?ae->units:"",
2168 ae->info?ae->info:""
2171 ae->notifications |= HEALTH_ENTRY_NOTIFICATIONS_EXEC_RUN;
2172 ae->exec_run_timestamp = time(NULL);
2174 debug(D_HEALTH, "executing command '%s'", buffer);
2175 FILE *fp = mypopen(buffer, &command_pid);
2177 error("HEALTH: Cannot popen(\"%s\", \"r\").", buffer);
2180 debug(D_HEALTH, "HEALTH reading from command");
2181 char *s = fgets(buffer, FILENAME_MAX, fp);
2183 ae->exec_code = mypclose(fp, command_pid);
2184 debug(D_HEALTH, "done executing command - returned with code %d", ae->exec_code);
2186 if(ae->exec_code != 0)
2187 ae->notifications |= HEALTH_ENTRY_NOTIFICATIONS_EXEC_FAILED;
2190 health_alarm_log_save(host, ae);
2194 static inline void health_process_notifications(RRDHOST *host, ALARM_ENTRY *ae) {
2195 info("Health alarm '%s.%s' = %0.2Lf - changed status from %s to %s",
2196 ae->chart?ae->chart:"NOCHART", ae->name,
2198 rrdcalc_status2string(ae->old_status),
2199 rrdcalc_status2string(ae->new_status)
2202 health_alarm_execute(host, ae);
2205 static inline void health_alarm_log_process(RRDHOST *host) {
2206 static uint32_t stop_at_id = 0;
2207 uint32_t first_waiting = (host->health_log.alarms)?host->health_log.alarms->unique_id:0;
2208 time_t now = time(NULL);
2210 pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
2213 for(ae = host->health_log.alarms; ae && ae->unique_id >= stop_at_id ; ae = ae->next) {
2215 !(ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_PROCESSED) &&
2216 !(ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_UPDATED)
2219 if(unlikely(ae->unique_id < first_waiting))
2220 first_waiting = ae->unique_id;
2222 if(likely(now >= ae->delay_up_to_timestamp))
2223 health_process_notifications(host, ae);
2227 // remember this for the next iteration
2228 stop_at_id = first_waiting;
2230 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
2232 if(host->health_log.count <= host->health_log.max)
2235 // cleanup excess entries in the log
2236 pthread_rwlock_wrlock(&host->health_log.alarm_log_rwlock);
2238 ALARM_ENTRY *last = NULL;
2239 unsigned int count = host->health_log.max * 2 / 3;
2240 for(ae = host->health_log.alarms; ae && count ; count--, last = ae, ae = ae->next) ;
2242 if(ae && last && last->next == ae)
2248 debug(D_HEALTH, "Health removing alarm log entry with id: %u", ae->unique_id);
2250 ALARM_ENTRY *t = ae->next;
2256 freez(ae->recipient);
2263 host->health_log.count--;
2266 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
2269 static inline int rrdcalc_isrunnable(RRDCALC *rc, time_t now, time_t *next_run) {
2270 if (unlikely(!rc->rrdset)) {
2271 debug(D_HEALTH, "Health not running alarm '%s.%s'. It is not linked to a chart.", rc->chart?rc->chart:"NOCHART", rc->name);
2275 if (unlikely(!rc->rrdset->last_collected_time.tv_sec)) {
2276 debug(D_HEALTH, "Health not running alarm '%s.%s'. Chart is not yet collected.", rc->chart?rc->chart:"NOCHART", rc->name);
2280 if (unlikely(!rc->update_every)) {
2281 debug(D_HEALTH, "Health not running alarm '%s.%s'. It does not have an update frequency", rc->chart?rc->chart:"NOCHART", rc->name);
2285 if (unlikely(rc->next_update > now)) {
2286 if (unlikely(*next_run > rc->next_update))
2287 *next_run = rc->next_update;
2289 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));
2294 // we should check that the DB lookup is possible
2296 // - the duration of the chart includes the required timeframe
2297 // we SHOULD NOT check the dimensions - there might be alarms that refer non-existing dimensions (e.g. cpu steal)
2302 void *health_main(void *ptr) {
2305 info("HEALTH thread created with task id %d", gettid());
2307 if(pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, NULL) != 0)
2308 error("Cannot set pthread cancel type to DEFERRED.");
2310 if(pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL) != 0)
2311 error("Cannot set pthread cancel state to ENABLE.");
2313 int min_run_every = (int)config_get_number("health", "run at least every seconds", 10);
2314 if(min_run_every < 1) min_run_every = 1;
2316 BUFFER *wb = buffer_create(100);
2318 unsigned int loop = 0;
2319 while(health_enabled && !netdata_exit) {
2321 debug(D_HEALTH, "Health monitoring iteration no %u started", loop);
2323 int oldstate, runnable = 0;
2324 time_t now = time(NULL);
2325 time_t next_run = now + min_run_every;
2328 if (unlikely(pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate) != 0))
2329 error("Cannot set pthread cancel state to DISABLE.");
2331 rrdhost_rdlock(&localhost);
2333 // the first loop is to lookup values from the db
2334 for (rc = localhost.alarms; rc; rc = rc->next) {
2335 if (unlikely(!rrdcalc_isrunnable(rc, now, &next_run)))
2339 rc->old_value = rc->value;
2341 // 1. if there is database lookup, do it
2342 // 2. if there is calculation expression, run it
2344 if (unlikely(RRDCALC_HAS_DB_LOOKUP(rc))) {
2345 time_t old_db_timestamp = rc->db_before;
2346 int value_is_null = 0;
2348 int ret = rrd2value(rc->rrdset, wb, &rc->value,
2349 rc->dimensions, 1, rc->after, rc->before, rc->group,
2350 rc->options, &rc->db_after, &rc->db_before, &value_is_null);
2352 if (unlikely(ret != 200)) {
2353 // database lookup failed
2356 debug(D_HEALTH, "Health alarm '%s.%s': database lookup returned error %d", rc->chart?rc->chart:"NOCHART", rc->name, ret);
2358 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_ERROR))) {
2359 rc->rrdcalc_flags |= RRDCALC_FLAG_DB_ERROR;
2360 error("Health alarm '%s.%s': database lookup returned error %d", rc->chart?rc->chart:"NOCHART", rc->name, ret);
2363 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_ERROR))
2364 rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_ERROR;
2366 if (unlikely(old_db_timestamp == rc->db_before)) {
2367 // database is stale
2369 debug(D_HEALTH, "Health alarm '%s.%s': database is stale", rc->chart?rc->chart:"NOCHART", rc->name);
2371 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_STALE))) {
2372 rc->rrdcalc_flags |= RRDCALC_FLAG_DB_STALE;
2373 error("Health alarm '%s.%s': database is stale", rc->chart?rc->chart:"NOCHART", rc->name);
2376 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_STALE))
2377 rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_STALE;
2379 if (unlikely(value_is_null)) {
2380 // collected value is null
2384 debug(D_HEALTH, "Health alarm '%s.%s': database lookup returned empty value (possibly value is not collected yet)",
2385 rc->chart?rc->chart:"NOCHART", rc->name);
2387 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_NAN))) {
2388 rc->rrdcalc_flags |= RRDCALC_FLAG_DB_NAN;
2389 error("Health alarm '%s.%s': database lookup returned empty value (possibly value is not collected yet)",
2390 rc->chart?rc->chart:"NOCHART", rc->name);
2393 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_NAN))
2394 rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_NAN;
2396 debug(D_HEALTH, "Health alarm '%s.%s': database lookup gave value "
2397 CALCULATED_NUMBER_FORMAT, rc->chart?rc->chart:"NOCHART", rc->name, rc->value);
2400 if(unlikely(rc->calculation)) {
2401 if (unlikely(!expression_evaluate(rc->calculation))) {
2402 // calculation failed
2406 debug(D_HEALTH, "Health alarm '%s.%s': failed to evaluate calculation with error: %s",
2407 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->calculation->error_msg));
2409 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_CALC_ERROR))) {
2410 rc->rrdcalc_flags |= RRDCALC_FLAG_CALC_ERROR;
2411 error("Health alarm '%s.%s': failed to evaluate calculation with error: %s",
2412 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->calculation->error_msg));
2416 if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_CALC_ERROR))
2417 rc->rrdcalc_flags &= ~RRDCALC_FLAG_CALC_ERROR;
2419 debug(D_HEALTH, "Health alarm '%s.%s': calculation expression gave value "
2420 CALCULATED_NUMBER_FORMAT
2421 ": %s (source: %s)",
2422 rc->chart?rc->chart:"NOCHART", rc->name,
2423 rc->calculation->result,
2424 buffer_tostring(rc->calculation->error_msg),
2428 rc->value = rc->calculation->result;
2432 rrdhost_unlock(&localhost);
2434 if (unlikely(runnable && !netdata_exit)) {
2435 rrdhost_rdlock(&localhost);
2437 for (rc = localhost.alarms; rc; rc = rc->next) {
2438 if (unlikely(!rrdcalc_isrunnable(rc, now, &next_run)))
2441 int warning_status = RRDCALC_STATUS_UNDEFINED;
2442 int critical_status = RRDCALC_STATUS_UNDEFINED;
2444 if(likely(rc->warning)) {
2445 if(unlikely(!expression_evaluate(rc->warning))) {
2446 // calculation failed
2448 debug(D_HEALTH, "Health alarm '%s.%s': warning expression failed with error: %s",
2449 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->warning->error_msg));
2451 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_WARN_ERROR))) {
2452 rc->rrdcalc_flags |= RRDCALC_FLAG_WARN_ERROR;
2453 error("Health alarm '%s.%s': warning expression failed with error: %s",
2454 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->warning->error_msg));
2458 if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_WARN_ERROR))
2459 rc->rrdcalc_flags &= ~RRDCALC_FLAG_WARN_ERROR;
2461 debug(D_HEALTH, "Health alarm '%s.%s': warning expression gave value "
2462 CALCULATED_NUMBER_FORMAT
2463 ": %s (source: %s)",
2464 rc->chart?rc->chart:"NOCHART", rc->name,
2465 rc->warning->result,
2466 buffer_tostring(rc->warning->error_msg),
2470 warning_status = rrdcalc_value2status(rc->warning->result);
2474 if(likely(rc->critical)) {
2475 if(unlikely(!expression_evaluate(rc->critical))) {
2476 // calculation failed
2478 debug(D_HEALTH, "Health alarm '%s.%s': critical expression failed with error: %s",
2479 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->critical->error_msg));
2481 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_CRIT_ERROR))) {
2482 rc->rrdcalc_flags |= RRDCALC_FLAG_CRIT_ERROR;
2483 error("Health alarm '%s.%s': critical expression failed with error: %s",
2484 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->critical->error_msg));
2488 if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_CRIT_ERROR))
2489 rc->rrdcalc_flags &= ~RRDCALC_FLAG_CRIT_ERROR;
2491 debug(D_HEALTH, "Health alarm '%s.%s': critical expression gave value "
2492 CALCULATED_NUMBER_FORMAT
2493 ": %s (source: %s)",
2494 rc->chart?rc->chart:"NOCHART", rc->name,
2495 rc->critical->result,
2496 buffer_tostring(rc->critical->error_msg),
2500 critical_status = rrdcalc_value2status(rc->critical->result);
2504 int status = RRDCALC_STATUS_UNDEFINED;
2506 switch(warning_status) {
2507 case RRDCALC_STATUS_CLEAR:
2508 status = RRDCALC_STATUS_CLEAR;
2511 case RRDCALC_STATUS_RAISED:
2512 status = RRDCALC_STATUS_WARNING;
2519 switch(critical_status) {
2520 case RRDCALC_STATUS_CLEAR:
2521 if(status == RRDCALC_STATUS_UNDEFINED)
2522 status = RRDCALC_STATUS_CLEAR;
2525 case RRDCALC_STATUS_RAISED:
2526 status = RRDCALC_STATUS_CRITICAL;
2533 if(status != rc->status) {
2536 if(now > rc->delay_up_to_timestamp) {
2537 rc->delay_up_current = rc->delay_up_duration;
2538 rc->delay_down_current = rc->delay_down_duration;
2540 rc->delay_up_to_timestamp = 0;
2543 rc->delay_up_current = (int)(rc->delay_up_current * rc->delay_multiplier);
2544 if(rc->delay_up_current > rc->delay_max_duration) rc->delay_up_current = rc->delay_max_duration;
2546 rc->delay_down_current = (int)(rc->delay_down_current * rc->delay_multiplier);
2547 if(rc->delay_down_current > rc->delay_max_duration) rc->delay_down_current = rc->delay_max_duration;
2550 if(status > rc->status)
2551 delay = rc->delay_up_current;
2553 delay = rc->delay_down_current;
2555 // COMMENTED: because we do need to send raising alarms
2556 // if(now + delay < rc->delay_up_to_timestamp)
2557 // delay = (int)(rc->delay_up_to_timestamp - now);
2559 rc->delay_last = delay;
2560 rc->delay_up_to_timestamp = now + delay;
2561 health_alarm_log(&localhost, rc->id, rc->next_event_id++, now, rc->name, rc->rrdset->id, rc->rrdset->family, rc->exec, rc->recipient, now - rc->last_status_change, rc->old_value, rc->value, rc->status, status, rc->source, rc->units, rc->info, rc->delay_last);
2562 rc->last_status_change = now;
2563 rc->status = status;
2566 rc->last_updated = now;
2567 rc->next_update = now + rc->update_every;
2569 if (next_run > rc->next_update)
2570 next_run = rc->next_update;
2573 rrdhost_unlock(&localhost);
2576 if (unlikely(pthread_setcancelstate(oldstate, NULL) != 0))
2577 error("Cannot set pthread cancel state to RESTORE (%d).", oldstate);
2579 if(unlikely(netdata_exit))
2582 // execute notifications
2584 health_alarm_log_process(&localhost);
2586 if(unlikely(netdata_exit))
2590 if(now < next_run) {
2591 debug(D_HEALTH, "Health monitoring iteration no %u done. Next iteration in %d secs",
2592 loop, (int) (next_run - now));
2593 sleep_usec(1000000 * (unsigned long long) (next_run - now));
2596 debug(D_HEALTH, "Health monitoring iteration no %u done. Next iteration now", loop);
2602 info("HEALTH thread exiting");