3 #define RRDVAR_MAX_LENGTH 1024
5 static const char *health_default_exec = PLUGINS_DIR "/alarm-email.sh";
6 static const char *health_default_recipient = "root";
7 int health_enabled = 1;
9 // ----------------------------------------------------------------------------
12 static inline int rrdvar_fix_name(char *variable) {
15 if (!isalnum(*variable) && *variable != '.' && *variable != '_') {
26 int rrdvar_compare(void* a, void* b) {
27 if(((RRDVAR *)a)->hash < ((RRDVAR *)b)->hash) return -1;
28 else if(((RRDVAR *)a)->hash > ((RRDVAR *)b)->hash) return 1;
29 else return strcmp(((RRDVAR *)a)->name, ((RRDVAR *)b)->name);
32 static inline RRDVAR *rrdvar_index_add(avl_tree_lock *tree, RRDVAR *rv) {
33 RRDVAR *ret = (RRDVAR *)avl_insert_lock(tree, (avl *)(rv));
35 debug(D_VARIABLES, "Request to insert RRDVAR '%s' into index failed. Already exists.", rv->name);
40 static inline RRDVAR *rrdvar_index_del(avl_tree_lock *tree, RRDVAR *rv) {
41 RRDVAR *ret = (RRDVAR *)avl_remove_lock(tree, (avl *)(rv));
43 error("Request to remove RRDVAR '%s' from index failed. Not Found.", rv->name);
48 static inline RRDVAR *rrdvar_index_find(avl_tree_lock *tree, const char *name, uint32_t hash) {
50 tmp.name = (char *)name;
51 tmp.hash = (hash)?hash:simple_hash(tmp.name);
53 return (RRDVAR *)avl_search_lock(tree, (avl *)&tmp);
56 static inline void rrdvar_free(RRDHOST *host, avl_tree_lock *tree, RRDVAR *rv) {
62 rrdvar_index_del(tree, rv);
68 static inline RRDVAR *rrdvar_create_and_index(const char *scope, avl_tree_lock *tree, const char *name, int type, calculated_number *value) {
69 char *variable = strdupz(name);
70 rrdvar_fix_name(variable);
71 uint32_t hash = simple_hash(variable);
73 RRDVAR *rv = rrdvar_index_find(tree, variable, hash);
75 debug(D_VARIABLES, "Variable '%s' not found in scope '%s'. Creating a new one.", variable, scope);
77 rv = callocz(1, sizeof(RRDVAR));
83 RRDVAR *ret = rrdvar_index_add(tree, rv);
84 if(unlikely(ret != rv)) {
85 debug(D_VARIABLES, "Variable '%s' in scope '%s' already exists", variable, scope);
86 rrdvar_free(NULL, NULL, rv);
90 debug(D_VARIABLES, "Variable '%s' created in scope '%s'", variable, scope);
101 // ----------------------------------------------------------------------------
104 calculated_number rrdvar2number(RRDVAR *rv) {
106 case RRDVAR_TYPE_CALCULATED: {
107 calculated_number *n = (calculated_number *)rv->value;
111 case RRDVAR_TYPE_TIME_T: {
112 time_t *n = (time_t *)rv->value;
116 case RRDVAR_TYPE_COLLECTED: {
117 collected_number *n = (collected_number *)rv->value;
121 case RRDVAR_TYPE_TOTAL: {
122 total_number *n = (total_number *)rv->value;
126 case RRDVAR_TYPE_INT: {
127 int *n = (int *)rv->value;
132 error("I don't know how to convert RRDVAR type %d to calculated_number", rv->type);
137 void dump_variable(void *data) {
138 RRDVAR *rv = (RRDVAR *)data;
139 debug(D_HEALTH, "%50s : %20.5Lf", rv->name, rrdvar2number(rv));
142 int health_variable_lookup(const char *variable, uint32_t hash, RRDCALC *rc, calculated_number *result) {
143 RRDSET *st = rc->rrdset;
148 rv = rrdvar_index_find(&st->variables_root_index, variable, hash);
150 *result = rrdvar2number(rv);
154 rv = rrdvar_index_find(&st->rrdfamily->variables_root_index, variable, hash);
156 *result = rrdvar2number(rv);
160 rv = rrdvar_index_find(&st->rrdhost->variables_root_index, variable, hash);
162 *result = rrdvar2number(rv);
166 debug(D_HEALTH, "Available local chart '%s' variables:", st->id);
167 avl_traverse_lock(&st->variables_root_index, dump_variable);
169 debug(D_HEALTH, "Available family '%s' variables:", st->rrdfamily->family);
170 avl_traverse_lock(&st->rrdfamily->variables_root_index, dump_variable);
172 debug(D_HEALTH, "Available host '%s' variables:", st->rrdhost->hostname);
173 avl_traverse_lock(&st->rrdhost->variables_root_index, dump_variable);
178 // ----------------------------------------------------------------------------
179 // RRDSETVAR management
181 RRDSETVAR *rrdsetvar_create(RRDSET *st, const char *variable, int type, void *value, uint32_t options) {
182 debug(D_VARIABLES, "RRDVARSET create for chart id '%s' name '%s' with variable name '%s'", st->id, st->name, variable);
183 RRDSETVAR *rs = (RRDSETVAR *)callocz(1, sizeof(RRDSETVAR));
185 char buffer[RRDVAR_MAX_LENGTH + 1];
186 snprintfz(buffer, RRDVAR_MAX_LENGTH, "%s.%s", st->id, variable);
187 rs->fullid = strdupz(buffer);
189 snprintfz(buffer, RRDVAR_MAX_LENGTH, "%s.%s", st->name, variable);
190 rs->fullname = strdupz(buffer);
192 rs->variable = strdupz(variable);
196 rs->options = options;
199 rs->local = rrdvar_create_and_index("local", &st->variables_root_index, rs->variable, rs->type, rs->value);
200 rs->family = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->fullid, rs->type, rs->value);
201 rs->host = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullid, rs->type, rs->value);
202 rs->family_name = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->fullname, rs->type, rs->value);
203 rs->host_name = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullname, rs->type, rs->value);
205 rs->next = st->variables;
211 void rrdsetvar_rename_all(RRDSET *st) {
212 debug(D_VARIABLES, "RRDSETVAR rename for chart id '%s' name '%s'", st->id, st->name);
214 // only these 2 can change name
218 char buffer[RRDVAR_MAX_LENGTH + 1];
219 RRDSETVAR *rs, *next = st->variables;
223 snprintfz(buffer, RRDVAR_MAX_LENGTH, "%s.%s", st->name, rs->variable);
225 if (strcmp(buffer, rs->fullname)) {
227 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family_name);
228 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_name);
231 rs->fullname = strdupz(st->name);
232 rs->family_name = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->fullname, rs->type, rs->value);
233 rs->host_name = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullname, rs->type, rs->value);
237 rrdsetcalc_link_matching(st);
240 void rrdsetvar_free(RRDSETVAR *rs) {
241 RRDSET *st = rs->rrdset;
242 debug(D_VARIABLES, "RRDSETVAR free for chart id '%s' name '%s', variable '%s'", st->id, st->name, rs->variable);
244 rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local);
245 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family);
246 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host);
247 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family_name);
248 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_name);
250 if(st->variables == rs) {
251 st->variables = rs->next;
255 for (t = st->variables; t && t->next != rs; t = t->next);
256 if(!t) error("RRDSETVAR '%s' not found in chart '%s' variables linked list", rs->fullname, st->id);
257 else t->next = rs->next;
266 // ----------------------------------------------------------------------------
267 // RRDDIMVAR management
269 #define RRDDIMVAR_ID_MAX 1024
271 RRDDIMVAR *rrddimvar_create(RRDDIM *rd, int type, const char *prefix, const char *suffix, void *value, uint32_t options) {
272 RRDSET *st = rd->rrdset;
274 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:"");
276 if(!prefix) prefix = "";
277 if(!suffix) suffix = "";
279 char buffer[RRDDIMVAR_ID_MAX + 1];
280 RRDDIMVAR *rs = (RRDDIMVAR *)callocz(1, sizeof(RRDDIMVAR));
282 rs->prefix = strdupz(prefix);
283 rs->suffix = strdupz(suffix);
285 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s%s%s", rs->prefix, rd->id, rs->suffix);
286 rs->id = strdupz(buffer);
288 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s%s%s", rs->prefix, rd->name, rs->suffix);
289 rs->name = strdupz(buffer);
291 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->id, rs->id);
292 rs->fullidid = strdupz(buffer);
294 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->id, rs->name);
295 rs->fullidname = strdupz(buffer);
297 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->name, rs->id);
298 rs->fullnameid = strdupz(buffer);
300 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->name, rs->name);
301 rs->fullnamename = strdupz(buffer);
305 rs->options = options;
308 rs->local_id = rrdvar_create_and_index("local", &st->variables_root_index, rs->id, rs->type, rs->value);
309 rs->local_name = rrdvar_create_and_index("local", &st->variables_root_index, rs->name, rs->type, rs->value);
311 rs->family_id = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->id, rs->type, rs->value);
312 rs->family_name = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->name, rs->type, rs->value);
314 rs->host_fullidid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullidid, rs->type, rs->value);
315 rs->host_fullidname = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullidname, rs->type, rs->value);
316 rs->host_fullnameid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullnameid, rs->type, rs->value);
317 rs->host_fullnamename = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullnamename, rs->type, rs->value);
319 rs->next = rd->variables;
325 void rrddimvar_rename_all(RRDDIM *rd) {
326 RRDSET *st = rd->rrdset;
327 debug(D_VARIABLES, "RRDDIMSET rename for chart id '%s' name '%s', dimension id '%s', name '%s'", st->id, st->name, rd->id, rd->name);
329 RRDDIMVAR *rs, *next = rd->variables;
333 if (strcmp(rd->name, rs->name)) {
334 char buffer[RRDDIMVAR_ID_MAX + 1];
338 rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local_name);
340 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s%s%s", rs->prefix, rd->name, rs->suffix);
341 rs->name = strdupz(buffer);
342 rs->local_name = rrdvar_create_and_index("local", &st->variables_root_index, rs->name, rs->type, rs->value);
344 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullidname);
345 freez(rs->fullidname);
346 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->id, rs->name);
347 rs->fullidname = strdupz(buffer);
348 rs->host_fullidname = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index,
349 rs->fullidname, rs->type, rs->value);
352 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnameid);
353 freez(rs->fullnameid);
354 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->name, rs->id);
355 rs->fullnameid = strdupz(buffer);
356 rs->host_fullnameid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index,
357 rs->fullnameid, rs->type, rs->value);
360 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnamename);
361 freez(rs->fullnamename);
362 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->name, rs->name);
363 rs->fullnamename = strdupz(buffer);
364 rs->host_fullnamename = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index,
365 rs->fullnamename, rs->type, rs->value);
370 void rrddimvar_free(RRDDIMVAR *rs) {
371 RRDDIM *rd = rs->rrddim;
372 RRDSET *st = rd->rrdset;
373 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);
375 rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local_id);
376 rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local_name);
378 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family_id);
379 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family_name);
381 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullidid);
382 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullidname);
383 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnameid);
384 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnamename);
386 if(rd->variables == rs) {
387 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);
388 rd->variables = rs->next;
391 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);
393 for (t = rd->variables; t && t->next != rs; t = t->next) ;
394 if(!t) error("RRDDIMVAR '%s' not found in dimension '%s/%s' variables linked list", rs->name, st->id, rd->id);
395 else t->next = rs->next;
403 freez(rs->fullidname);
404 freez(rs->fullnameid);
405 freez(rs->fullnamename);
409 // ----------------------------------------------------------------------------
410 // RRDCALC management
412 static inline const char *rrdcalc_status2string(int status) {
414 case RRDCALC_STATUS_UNINITIALIZED:
415 return "UNINITIALIZED";
417 case RRDCALC_STATUS_UNDEFINED:
420 case RRDCALC_STATUS_CLEAR:
423 case RRDCALC_STATUS_RAISED:
426 case RRDCALC_STATUS_WARNING:
429 case RRDCALC_STATUS_CRITICAL:
437 static void rrdsetcalc_link(RRDSET *st, RRDCALC *rc) {
438 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);
440 rc->last_status_change = time(NULL);
443 rc->rrdset_next = st->alarms;
444 rc->rrdset_prev = NULL;
447 if(rc->update_every < rc->rrdset->update_every) {
448 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);
449 rc->update_every = rc->rrdset->update_every;
452 if(!isnan(rc->green) && isnan(st->green)) {
453 debug(D_HEALTH, "Health alarm '%s.%s' green threshold set from %Lf to %Lf.", rc->rrdset->id, rc->name, rc->rrdset->green, rc->green);
454 st->green = rc->green;
457 if(!isnan(rc->red) && isnan(st->red)) {
458 debug(D_HEALTH, "Health alarm '%s.%s' red threshold set from %Lf to %Lf.", rc->rrdset->id, rc->name, rc->rrdset->red, rc->red);
462 rc->local = rrdvar_create_and_index("local", &st->variables_root_index, rc->name, RRDVAR_TYPE_CALCULATED, &rc->value);
463 rc->family = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rc->name, RRDVAR_TYPE_CALCULATED, &rc->value);
465 char fullname[RRDVAR_MAX_LENGTH + 1];
466 snprintfz(fullname, RRDVAR_MAX_LENGTH, "%s.%s", st->id, rc->name);
467 rc->hostid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, fullname, RRDVAR_TYPE_CALCULATED, &rc->value);
469 snprintfz(fullname, RRDVAR_MAX_LENGTH, "%s.%s", st->name, rc->name);
470 rc->hostname = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, fullname, RRDVAR_TYPE_CALCULATED, &rc->value);
472 if(!rc->units) rc->units = strdupz(st->units);
475 static inline int rrdcalc_is_matching_this_rrdset(RRDCALC *rc, RRDSET *st) {
476 if( (rc->hash_chart == st->hash && !strcmp(rc->chart, st->id)) ||
477 (rc->hash_chart == st->hash_name && !strcmp(rc->chart, st->name)))
483 // this has to be called while the RRDHOST is locked
484 inline void rrdsetcalc_link_matching(RRDSET *st) {
485 // debug(D_HEALTH, "find matching alarms for chart '%s'", st->id);
488 for(rc = st->rrdhost->alarms; rc ; rc = rc->next) {
489 if(unlikely(rc->rrdset))
492 if(unlikely(rrdcalc_is_matching_this_rrdset(rc, st)))
493 rrdsetcalc_link(st, rc);
497 // this has to be called while the RRDHOST is locked
498 inline void rrdsetcalc_unlink(RRDCALC *rc) {
499 RRDSET *st = rc->rrdset;
502 error("Requested to unlink RRDCALC '%s.%s' which is not linked to any RRDSET", rc->chart?rc->chart:"NOCHART", rc->name);
506 RRDHOST *host = st->rrdhost;
508 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);
512 rc->rrdset_prev->rrdset_next = rc->rrdset_next;
515 rc->rrdset_next->rrdset_prev = rc->rrdset_prev;
518 st->alarms = rc->rrdset_next;
520 rc->rrdset_prev = rc->rrdset_next = NULL;
522 rrdvar_free(st->rrdhost, &st->variables_root_index, rc->local);
525 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rc->family);
528 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rc->hostid);
531 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rc->hostname);
536 // RRDCALC will remain in RRDHOST
537 // so that if the matching chart is found in the future
538 // it will be applied automatically
541 RRDCALC *rrdcalc_find(RRDSET *st, const char *name) {
543 uint32_t hash = simple_hash(name);
545 for( rc = st->alarms; rc ; rc = rc->rrdset_next ) {
546 if(unlikely(rc->hash == hash && !strcmp(rc->name, name)))
553 static inline int rrdcalc_exists(RRDHOST *host, const char *name, uint32_t hash) {
556 // make sure it does not already exist
557 for(rc = host->alarms; rc ; rc = rc->next) {
558 if (unlikely(rc->hash == hash && !strcmp(name, rc->name))) {
559 error("Health alarm '%s' already exists in host '%s'.", name, host->hostname);
567 static inline void rrdcalc_create_part2(RRDHOST *host, RRDCALC *rc) {
568 rrdhost_check_rdlock(host);
570 if(rc->calculation) {
571 rc->calculation->this = &rc->value;
572 rc->calculation->after = &rc->db_after;
573 rc->calculation->before = &rc->db_before;
574 rc->calculation->rrdcalc = rc;
578 rc->warning->this = &rc->value;
579 rc->warning->after = &rc->db_after;
580 rc->warning->before = &rc->db_before;
581 rc->warning->rrdcalc = rc;
585 rc->critical->this = &rc->value;
586 rc->critical->after = &rc->db_after;
587 rc->critical->before = &rc->db_before;
588 rc->critical->rrdcalc = rc;
591 // link it to the host
592 if(likely(host->alarms)) {
595 for(t = host->alarms; t && t->next ; t = t->next) ;
602 // link it to its chart
604 for(st = host->rrdset_root; st ; st = st->next) {
605 if(rrdcalc_is_matching_this_rrdset(rc, st)) {
606 rrdsetcalc_link(st, rc);
612 static inline uint32_t rrdcalc_fullname(char *fullname, size_t len, const char *chart, const char *name) {
613 snprintfz(fullname, len - 1, "%s%s%s", chart?chart:"", chart?".":"", name);
614 rrdvar_fix_name(fullname);
615 return simple_hash(fullname);
618 static inline RRDCALC *rrdcalc_create(RRDHOST *host, const char *name, const char *chart, const char *dimensions,
619 const char *units, const char *info,
620 int group_method, int after, int before, int update_every, uint32_t options,
621 calculated_number green, calculated_number red,
622 const char *exec, const char *recipient, const char *source,
623 const char *calc, const char *warn, const char *crit) {
625 char fullname[RRDVAR_MAX_LENGTH + 1];
626 uint32_t hash = rrdcalc_fullname(fullname, RRDVAR_MAX_LENGTH + 1, chart, name);
628 if(rrdcalc_exists(host, fullname, hash))
631 RRDCALC *rc = callocz(1, sizeof(RRDCALC));
632 rc->id = host->health_log.next_alarm_id++;
633 rc->next_event_id = 1;
634 rc->name = strdupz(name);
635 rc->hash = simple_hash(rc->name);
636 rc->chart = strdupz(chart);
637 rc->hash_chart = simple_hash(rc->chart);
639 if(dimensions) rc->dimensions = strdupz(dimensions);
646 rc->group = group_method;
649 rc->update_every = update_every;
650 rc->options = options;
652 if(exec) rc->exec = strdupz(exec);
653 if(recipient) rc->recipient = strdupz(recipient);
654 if(source) rc->source = strdupz(source);
655 if(units) rc->units = strdupz(units);
656 if(info) rc->info = strdupz(info);
659 rc->calculation = expression_parse(calc, NULL, NULL);
661 error("Health alarm '%s.%s': failed to parse calculation expression '%s'", chart, name, calc);
664 rc->warning = expression_parse(warn, NULL, NULL);
666 error("Health alarm '%s.%s': failed to re-parse warning expression '%s'", chart, name, warn);
669 rc->critical = expression_parse(crit, NULL, NULL);
671 error("Health alarm '%s.%s': failed to re-parse critical expression '%s'", chart, name, crit);
674 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",
675 (rc->chart)?rc->chart:"NOCHART",
677 (rc->exec)?rc->exec:"DEFAULT",
678 (rc->recipient)?rc->recipient:"DEFAULT",
685 (rc->dimensions)?rc->dimensions:"NONE",
687 (rc->calculation)?rc->calculation->parsed_as:"NONE",
688 (rc->warning)?rc->warning->parsed_as:"NONE",
689 (rc->critical)?rc->critical->parsed_as:"NONE",
693 rrdcalc_create_part2(host, rc);
697 void rrdcalc_free(RRDHOST *host, RRDCALC *rc) {
700 debug(D_HEALTH, "Health removing alarm '%s.%s' of host '%s'", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);
702 // unlink it from RRDSET
703 if(rc->rrdset) rrdsetcalc_unlink(rc);
705 // unlink it from RRDHOST
706 if(unlikely(rc == host->alarms))
707 host->alarms = rc->next;
709 else if(likely(host->alarms)) {
710 RRDCALC *t, *last = host->alarms;
712 for(t = last->next; t && t != rc; last = t, t = t->next) ;
713 if(last && last->next == rc)
714 last->next = rc->next;
716 error("Cannot unlink alarm '%s.%s' from host '%s': not found", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);
719 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);
721 expression_free(rc->calculation);
722 expression_free(rc->warning);
723 expression_free(rc->critical);
728 freez(rc->dimensions);
730 freez(rc->recipient);
737 // ----------------------------------------------------------------------------
738 // RRDCALCTEMPLATE management
740 void rrdcalctemplate_link_matching(RRDSET *st) {
743 for(rt = st->rrdhost->templates; rt ; rt = rt->next) {
744 if(rt->hash_context == st->hash_context && !strcmp(rt->context, st->context)) {
745 RRDCALC *rc = rrdcalc_create(st->rrdhost, rt->name, st->id,
746 rt->dimensions, rt->units, rt->info, rt->group, rt->after, rt->before, rt->update_every, rt->options,
748 rt->exec, rt->recipient, rt->source,
749 (rt->calculation)?rt->calculation->source:NULL,
750 (rt->warning)?rt->warning->source:NULL,
751 (rt->critical)?rt->critical->source:NULL);
754 error("Health tried to create alarm from template '%s', but it failed", rt->name);
756 #ifdef NETDATA_INTERNAL_CHECKS
757 else if(rc->rrdset != st)
758 error("Health alarm '%s.%s' should be linked to chart '%s', but it is not", rc->chart?rc->chart:"NOCHART", rc->name, st->id);
764 static inline void rrdcalctemplate_free(RRDHOST *host, RRDCALCTEMPLATE *rt) {
765 debug(D_HEALTH, "Health removing template '%s' of host '%s'", rt->name, host->hostname);
767 if(host->templates) {
768 if(host->templates == rt) {
769 host->templates = rt->next;
772 RRDCALCTEMPLATE *t, *last = host->templates;
773 for (t = last->next; t && t != rt; last = t, t = t->next ) ;
774 if(last && last->next == rt) {
775 last->next = rt->next;
779 error("Cannot find RRDCALCTEMPLATE '%s' linked in host '%s'", rt->name, host->hostname);
783 expression_free(rt->calculation);
784 expression_free(rt->warning);
785 expression_free(rt->critical);
789 freez(rt->recipient);
794 freez(rt->dimensions);
798 // ----------------------------------------------------------------------------
799 // load health configuration
801 #define HEALTH_CONF_MAX_LINE 4096
803 #define HEALTH_ALARM_KEY "alarm"
804 #define HEALTH_TEMPLATE_KEY "template"
805 #define HEALTH_ON_KEY "on"
806 #define HEALTH_LOOKUP_KEY "lookup"
807 #define HEALTH_CALC_KEY "calc"
808 #define HEALTH_EVERY_KEY "every"
809 #define HEALTH_GREEN_KEY "green"
810 #define HEALTH_RED_KEY "red"
811 #define HEALTH_WARN_KEY "warn"
812 #define HEALTH_CRIT_KEY "crit"
813 #define HEALTH_EXEC_KEY "exec"
814 #define HEALTH_RECIPIENT_KEY "to"
815 #define HEALTH_UNITS_KEY "units"
816 #define HEALTH_INFO_KEY "info"
818 static inline int rrdcalc_add_alarm_from_config(RRDHOST *host, RRDCALC *rc) {
820 char fullname[RRDVAR_MAX_LENGTH + 1];
821 uint32_t hash = rrdcalc_fullname(fullname, RRDVAR_MAX_LENGTH + 1, rc->chart, rc->name);
823 if (rrdcalc_exists(host, fullname, hash))
828 error("Health configuration for alarm '%s' does not have a chart", rc->name);
832 if(!rc->update_every) {
833 error("Health configuration for alarm '%s.%s' has no frequency (parameter 'every'). Ignoring it.", rc->chart?rc->chart:"NOCHART", rc->name);
837 if(!RRDCALC_HAS_DB_LOOKUP(rc) && !rc->warning && !rc->critical) {
838 error("Health configuration for alarm '%s.%s' is useless (no calculation, no warning and no critical evaluation)", rc->chart?rc->chart:"NOCHART", rc->name);
842 debug(D_HEALTH, "Health configuration adding 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",
843 rc->chart?rc->chart:"NOCHART",
845 (rc->exec)?rc->exec:"DEFAULT",
846 (rc->recipient)?rc->recipient:"DEFAULT",
853 (rc->dimensions)?rc->dimensions:"NONE",
855 (rc->calculation)?rc->calculation->parsed_as:"NONE",
856 (rc->warning)?rc->warning->parsed_as:"NONE",
857 (rc->critical)?rc->critical->parsed_as:"NONE",
861 rrdcalc_create_part2(host, rc);
865 static inline int rrdcalctemplate_add_template_from_config(RRDHOST *host, RRDCALCTEMPLATE *rt) {
866 if(unlikely(!rt->context)) {
867 error("Health configuration for template '%s' does not have a context", rt->name);
871 if(unlikely(!rt->update_every)) {
872 error("Health configuration for template '%s' has no frequency (parameter 'every'). Ignoring it.", rt->name);
876 if(unlikely(!RRDCALCTEMPLATE_HAS_CALCULATION(rt) && !rt->warning && !rt->critical)) {
877 error("Health configuration for template '%s' is useless (no calculation, no warning and no critical evaluation)", rt->name);
881 RRDCALCTEMPLATE *t, *last = NULL;
882 for (t = host->templates; t ; last = t, t = t->next) {
883 if(unlikely(t->hash_name == rt->hash_name && !strcmp(t->name, rt->name))) {
884 error("Health configuration template '%s' already exists for host '%s'.", rt->name, host->hostname);
889 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'",
891 (rt->context)?rt->context:"NONE",
892 (rt->exec)?rt->exec:"DEFAULT",
893 (rt->recipient)?rt->recipient:"DEFAULT",
900 (rt->dimensions)?rt->dimensions:"NONE",
902 (rt->calculation)?rt->calculation->parsed_as:"NONE",
903 (rt->warning)?rt->warning->parsed_as:"NONE",
904 (rt->critical)?rt->critical->parsed_as:"NONE",
912 rt->next = host->templates;
913 host->templates = rt;
919 static inline int health_parse_duration(char *string, int *result) {
920 // make sure it is a number
921 if(!*string || !(isdigit(*string) || *string == '+' || *string == '-')) {
927 calculated_number n = strtold(string, &e);
931 *result = (int) (n * 86400 * 365);
934 *result = (int) (n * 86400 * 30);
937 *result = (int) (n * 86400 * 7);
940 *result = (int) (n * 86400);
943 *result = (int) (n * 3600);
946 *result = (int) (n * 60);
961 static inline int health_parse_db_lookup(
962 size_t line, const char *path, const char *file, char *string,
963 int *group_method, int *after, int *before, int *every,
964 uint32_t *options, char **dimensions
966 debug(D_HEALTH, "Health configuration parsing database lookup %zu@%s/%s: %s", line, path, file, string);
968 if(*dimensions) freez(*dimensions);
975 char *s = string, *key;
977 // first is the group method
979 while(*s && !isspace(*s)) s++;
980 while(*s && isspace(*s)) *s++ = '\0';
982 error("Health configuration invalid chart calculation at line %zu of file '%s/%s': expected group method followed by the 'after' time, but got '%s'",
983 line, path, file, key);
987 if((*group_method = web_client_api_request_v1_data_group(key, -1)) == -1) {
988 error("Health configuration at line %zu of file '%s/%s': invalid group method '%s'",
989 line, path, file, key);
993 // then is the 'after' time
995 while(*s && !isspace(*s)) s++;
996 while(*s && isspace(*s)) *s++ = '\0';
998 if(!health_parse_duration(key, after)) {
999 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' after group method",
1000 line, path, file, key);
1005 *every = abs(*after);
1007 // now we may have optional parameters
1010 while(*s && !isspace(*s)) s++;
1011 while(*s && isspace(*s)) *s++ = '\0';
1014 if(!strcasecmp(key, "at")) {
1016 while(*s && !isspace(*s)) s++;
1017 while(*s && isspace(*s)) *s++ = '\0';
1019 if (!health_parse_duration(value, before)) {
1020 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' for '%s' keyword",
1021 line, path, file, value, key);
1024 else if(!strcasecmp(key, HEALTH_EVERY_KEY)) {
1026 while(*s && !isspace(*s)) s++;
1027 while(*s && isspace(*s)) *s++ = '\0';
1029 if (!health_parse_duration(value, every)) {
1030 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' for '%s' keyword",
1031 line, path, file, value, key);
1034 else if(!strcasecmp(key, "absolute") || !strcasecmp(key, "abs") || !strcasecmp(key, "absolute_sum")) {
1035 *options |= RRDR_OPTION_ABSOLUTE;
1037 else if(!strcasecmp(key, "min2max")) {
1038 *options |= RRDR_OPTION_MIN2MAX;
1040 else if(!strcasecmp(key, "null2zero")) {
1041 *options |= RRDR_OPTION_NULL2ZERO;
1043 else if(!strcasecmp(key, "percentage")) {
1044 *options |= RRDR_OPTION_PERCENTAGE;
1046 else if(!strcasecmp(key, "unaligned")) {
1047 *options |= RRDR_OPTION_NOT_ALIGNED;
1049 else if(!strcasecmp(key, "of")) {
1050 if(*s && strcasecmp(s, "all"))
1051 *dimensions = strdupz(s);
1055 error("Health configuration at line %zu of file '%s/%s': unknown keyword '%s'",
1056 line, path, file, key);
1063 static inline char *health_source_file(size_t line, const char *path, const char *filename) {
1064 char buffer[FILENAME_MAX + 1];
1065 snprintfz(buffer, FILENAME_MAX, "%zu@%s/%s", line, path, filename);
1066 return strdupz(buffer);
1069 static inline void strip_quotes(char *s) {
1071 if(*s == '\'' || *s == '"') *s = ' ';
1076 int health_readfile(const char *path, const char *filename) {
1077 debug(D_HEALTH, "Health configuration reading file '%s/%s'", path, filename);
1079 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;
1080 char buffer[HEALTH_CONF_MAX_LINE + 1];
1082 if(unlikely(!hash_alarm)) {
1083 hash_alarm = simple_uhash(HEALTH_ALARM_KEY);
1084 hash_template = simple_uhash(HEALTH_TEMPLATE_KEY);
1085 hash_on = simple_uhash(HEALTH_ON_KEY);
1086 hash_calc = simple_uhash(HEALTH_CALC_KEY);
1087 hash_lookup = simple_uhash(HEALTH_LOOKUP_KEY);
1088 hash_green = simple_uhash(HEALTH_GREEN_KEY);
1089 hash_red = simple_uhash(HEALTH_RED_KEY);
1090 hash_warn = simple_uhash(HEALTH_WARN_KEY);
1091 hash_crit = simple_uhash(HEALTH_CRIT_KEY);
1092 hash_exec = simple_uhash(HEALTH_EXEC_KEY);
1093 hash_every = simple_uhash(HEALTH_EVERY_KEY);
1094 hash_units = simple_hash(HEALTH_UNITS_KEY);
1095 hash_info = simple_hash(HEALTH_INFO_KEY);
1096 hash_recipient = simple_hash(HEALTH_RECIPIENT_KEY);
1099 snprintfz(buffer, HEALTH_CONF_MAX_LINE, "%s/%s", path, filename);
1100 FILE *fp = fopen(buffer, "r");
1102 error("Health configuration cannot read file '%s'.", buffer);
1107 RRDCALCTEMPLATE *rt = NULL;
1109 size_t line = 0, append = 0;
1111 while((s = fgets(&buffer[append], (int)(HEALTH_CONF_MAX_LINE - append), fp)) || append) {
1112 int stop_appending = !s;
1114 // info("Line %zu of file '%s/%s': '%s'", line, path, filename, s);
1117 // info("Trimmed line %zu of file '%s/%s': '%s'", line, path, filename, s);
1120 if(!stop_appending && s[append - 1] == '\\') {
1121 s[append - 1] = ' ';
1122 append = &s[append] - buffer;
1123 if(append < HEALTH_CONF_MAX_LINE)
1126 error("Health configuration has too long muli-line at line %zu of file '%s/%s'.", line, path, filename);
1132 while(*s && *s != ':') s++;
1134 error("Health configuration has invalid line %zu of file '%s/%s'. It does not contain a ':'. Ignoring it.", line, path, filename);
1142 value = trim(value);
1145 error("Health configuration has invalid line %zu of file '%s/%s'. Keyword is empty. Ignoring it.", line, path, filename);
1150 error("Health configuration has invalid line %zu of file '%s/%s'. value is empty. Ignoring it.", line, path, filename);
1154 // info("Health file '%s/%s', key '%s', value '%s'", path, filename, key, value);
1155 uint32_t hash = simple_uhash(key);
1157 if(hash == hash_alarm && !strcasecmp(key, HEALTH_ALARM_KEY)) {
1158 if(rc && !rrdcalc_add_alarm_from_config(&localhost, rc))
1159 rrdcalc_free(&localhost, rc);
1162 if (!rrdcalctemplate_add_template_from_config(&localhost, rt))
1163 rrdcalctemplate_free(&localhost, rt);
1167 rc = callocz(1, sizeof(RRDCALC));
1168 rc->id = localhost.health_log.next_alarm_id++;
1169 rc->next_event_id = 1;
1170 rc->name = strdupz(value);
1171 rc->hash = simple_hash(rc->name);
1172 rc->source = health_source_file(line, path, filename);
1176 rc->old_value = NAN;
1178 if(rrdvar_fix_name(rc->name))
1179 error("Health configuration renamed alarm '%s' to '%s'", value, rc->name);
1181 else if(hash == hash_template && !strcasecmp(key, HEALTH_TEMPLATE_KEY)) {
1183 if(!rrdcalc_add_alarm_from_config(&localhost, rc))
1184 rrdcalc_free(&localhost, rc);
1188 if(rt && !rrdcalctemplate_add_template_from_config(&localhost, rt))
1189 rrdcalctemplate_free(&localhost, rt);
1191 rt = callocz(1, sizeof(RRDCALCTEMPLATE));
1192 rt->name = strdupz(value);
1193 rt->hash_name = simple_hash(rt->name);
1194 rt->source = health_source_file(line, path, filename);
1198 if(rrdvar_fix_name(rt->name))
1199 error("Health configuration renamed template '%s' to '%s'", value, rt->name);
1202 if(hash == hash_on && !strcasecmp(key, HEALTH_ON_KEY)) {
1204 if(strcmp(rc->chart, value))
1205 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').",
1206 line, path, filename, rc->name, key, rc->chart, value, value);
1210 rc->chart = strdupz(value);
1211 rc->hash_chart = simple_hash(rc->chart);
1213 else if(hash == hash_lookup && !strcasecmp(key, HEALTH_LOOKUP_KEY)) {
1214 health_parse_db_lookup(line, path, filename, value, &rc->group, &rc->after, &rc->before,
1216 &rc->options, &rc->dimensions);
1218 else if(hash == hash_every && !strcasecmp(key, HEALTH_EVERY_KEY)) {
1219 if(!health_parse_duration(value, &rc->update_every))
1220 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' cannot parse duration: '%s'.",
1221 line, path, filename, rc->name, key, value);
1223 else if(hash == hash_green && !strcasecmp(key, HEALTH_GREEN_KEY)) {
1225 rc->green = strtold(value, &e);
1227 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' leaves this string unmatched: '%s'.",
1228 line, path, filename, rc->name, key, e);
1231 else if(hash == hash_red && !strcasecmp(key, HEALTH_RED_KEY)) {
1233 rc->red = strtold(value, &e);
1235 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' leaves this string unmatched: '%s'.",
1236 line, path, filename, rc->name, key, e);
1239 else if(hash == hash_calc && !strcasecmp(key, HEALTH_CALC_KEY)) {
1240 const char *failed_at = NULL;
1242 rc->calculation = expression_parse(value, &failed_at, &error);
1243 if(!rc->calculation) {
1244 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1245 line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
1248 else if(hash == hash_warn && !strcasecmp(key, HEALTH_WARN_KEY)) {
1249 const char *failed_at = NULL;
1251 rc->warning = expression_parse(value, &failed_at, &error);
1253 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1254 line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
1257 else if(hash == hash_crit && !strcasecmp(key, HEALTH_CRIT_KEY)) {
1258 const char *failed_at = NULL;
1260 rc->critical = expression_parse(value, &failed_at, &error);
1262 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1263 line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
1266 else if(hash == hash_exec && !strcasecmp(key, HEALTH_EXEC_KEY)) {
1268 if(strcmp(rc->exec, value))
1269 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').",
1270 line, path, filename, rc->name, key, rc->exec, value, value);
1274 rc->exec = strdupz(value);
1276 else if(hash == hash_recipient && !strcasecmp(key, HEALTH_RECIPIENT_KEY)) {
1278 if(strcmp(rc->recipient, value))
1279 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').",
1280 line, path, filename, rc->name, key, rc->recipient, value, value);
1282 freez(rc->recipient);
1284 rc->recipient = strdupz(value);
1286 else if(hash == hash_units && !strcasecmp(key, HEALTH_UNITS_KEY)) {
1288 if(strcmp(rc->units, value))
1289 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').",
1290 line, path, filename, rc->name, key, rc->units, value, value);
1294 rc->units = strdupz(value);
1295 strip_quotes(rc->units);
1297 else if(hash == hash_info && !strcasecmp(key, HEALTH_INFO_KEY)) {
1299 if(strcmp(rc->info, value))
1300 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').",
1301 line, path, filename, rc->name, key, rc->info, value, value);
1305 rc->info = strdupz(value);
1306 strip_quotes(rc->info);
1309 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' has unknown key '%s'.",
1310 line, path, filename, rc->name, key);
1314 if(hash == hash_on && !strcasecmp(key, HEALTH_ON_KEY)) {
1316 if(strcmp(rt->context, value))
1317 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').",
1318 line, path, filename, rt->name, key, rt->context, value, value);
1322 rt->context = strdupz(value);
1323 rt->hash_context = simple_hash(rt->context);
1325 else if(hash == hash_lookup && !strcasecmp(key, HEALTH_LOOKUP_KEY)) {
1326 health_parse_db_lookup(line, path, filename, value, &rt->group, &rt->after, &rt->before,
1328 &rt->options, &rt->dimensions);
1330 else if(hash == hash_every && !strcasecmp(key, HEALTH_EVERY_KEY)) {
1331 if(!health_parse_duration(value, &rt->update_every))
1332 info("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' cannot parse duration: '%s'.",
1333 line, path, filename, rt->name, key, value);
1335 else if(hash == hash_green && !strcasecmp(key, HEALTH_GREEN_KEY)) {
1337 rt->green = strtold(value, &e);
1339 info("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' leaves this string unmatched: '%s'.",
1340 line, path, filename, rt->name, key, e);
1343 else if(hash == hash_red && !strcasecmp(key, HEALTH_RED_KEY)) {
1345 rt->red = strtold(value, &e);
1347 info("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' leaves this string unmatched: '%s'.",
1348 line, path, filename, rt->name, key, e);
1351 else if(hash == hash_calc && !strcasecmp(key, HEALTH_CALC_KEY)) {
1352 const char *failed_at = NULL;
1354 rt->calculation = expression_parse(value, &failed_at, &error);
1355 if(!rt->calculation) {
1356 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1357 line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
1360 else if(hash == hash_warn && !strcasecmp(key, HEALTH_WARN_KEY)) {
1361 const char *failed_at = NULL;
1363 rt->warning = expression_parse(value, &failed_at, &error);
1365 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1366 line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
1369 else if(hash == hash_crit && !strcasecmp(key, HEALTH_CRIT_KEY)) {
1370 const char *failed_at = NULL;
1372 rt->critical = expression_parse(value, &failed_at, &error);
1374 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1375 line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
1378 else if(hash == hash_exec && !strcasecmp(key, HEALTH_EXEC_KEY)) {
1380 if(strcmp(rt->exec, value))
1381 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').",
1382 line, path, filename, rt->name, key, rt->exec, value, value);
1386 rt->exec = strdupz(value);
1388 else if(hash == hash_recipient && !strcasecmp(key, HEALTH_RECIPIENT_KEY)) {
1390 if(strcmp(rt->recipient, value))
1391 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').",
1392 line, path, filename, rt->name, key, rt->recipient, value, value);
1394 freez(rt->recipient);
1396 rt->recipient = strdupz(value);
1398 else if(hash == hash_units && !strcasecmp(key, HEALTH_UNITS_KEY)) {
1400 if(strcmp(rt->units, value))
1401 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').",
1402 line, path, filename, rt->name, key, rt->units, value, value);
1406 rt->units = strdupz(value);
1407 strip_quotes(rt->units);
1409 else if(hash == hash_info && !strcasecmp(key, HEALTH_INFO_KEY)) {
1411 if(strcmp(rt->info, value))
1412 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').",
1413 line, path, filename, rt->name, key, rt->info, value, value);
1417 rt->info = strdupz(value);
1418 strip_quotes(rt->info);
1421 error("Health configuration at line %zu of file '%s/%s' for template '%s' has unknown key '%s'.",
1422 line, path, filename, rt->name, key);
1426 error("Health configuration at line %zu of file '%s/%s' has unknown key '%s'. Expected either '" HEALTH_ALARM_KEY "' or '" HEALTH_TEMPLATE_KEY "'.",
1427 line, path, filename, key);
1431 if(rc && !rrdcalc_add_alarm_from_config(&localhost, rc))
1432 rrdcalc_free(&localhost, rc);
1434 if(rt && !rrdcalctemplate_add_template_from_config(&localhost, rt))
1435 rrdcalctemplate_free(&localhost, rt);
1441 void health_readdir(const char *path) {
1442 size_t pathlen = strlen(path);
1444 debug(D_HEALTH, "Health configuration reading directory '%s'", path);
1446 DIR *dir = opendir(path);
1448 error("Health configuration cannot open directory '%s'.", path);
1452 struct dirent *de = NULL;
1453 while ((de = readdir(dir))) {
1454 size_t len = strlen(de->d_name);
1456 if(de->d_type == DT_DIR
1458 (de->d_name[0] == '.' && de->d_name[1] == '\0')
1459 || (de->d_name[0] == '.' && de->d_name[1] == '.' && de->d_name[2] == '\0')
1463 else if(de->d_type == DT_DIR) {
1464 char *s = mallocz(pathlen + strlen(de->d_name) + 2);
1467 strcat(s, de->d_name);
1473 else if((de->d_type == DT_LNK || de->d_type == DT_REG) &&
1474 len > 5 && !strcmp(&de->d_name[len - 5], ".conf")) {
1475 health_readfile(path, de->d_name);
1482 static inline char *health_config_dir(void) {
1483 char buffer[FILENAME_MAX + 1];
1484 snprintfz(buffer, FILENAME_MAX, "%s/health.d", config_get("global", "config directory", CONFIG_DIR));
1485 return config_get("health", "health configuration directory", buffer);
1488 void health_init(void) {
1489 debug(D_HEALTH, "Health configuration initializing");
1491 if(!(health_enabled = config_get_boolean("health", "enabled", 1))) {
1492 debug(D_HEALTH, "Health is disabled.");
1496 char *path = health_config_dir();
1499 char buffer[FILENAME_MAX + 1];
1500 snprintfz(buffer, FILENAME_MAX, "%s/alarm-email.sh", config_get("global", "plugins directory", PLUGINS_DIR));
1501 health_default_exec = config_get("health", "script to execute on alarm", buffer);
1504 long n = config_get_number("health", "in memory max health log entries", (long)localhost.health_log.max);
1506 error("Health configuration has invalid max log entries %ld. Using default %u", n, localhost.health_log.max);
1507 config_set_number("health", "in memory max health log entries", (long)localhost.health_log.max);
1509 else localhost.health_log.max = (unsigned int)n;
1511 rrdhost_rwlock(&localhost);
1512 health_readdir(path);
1513 rrdhost_unlock(&localhost);
1516 // ----------------------------------------------------------------------------
1519 static inline void health_string2json(BUFFER *wb, const char *prefix, const char *label, const char *value, const char *suffix) {
1521 buffer_sprintf(wb, "%s\"%s\":\"%s\"%s", prefix, label, value, suffix);
1523 buffer_sprintf(wb, "%s\"%s\":null%s", prefix, label, suffix);
1526 static inline void health_alarm_entry2json_nolock(BUFFER *wb, ALARM_ENTRY *ae) {
1527 buffer_sprintf(wb, "\n\t{\n"
1528 "\t\t\"unique_id\":%u,\n"
1529 "\t\t\"alarm_id\":%u,\n"
1530 "\t\t\"alarm_event_id\":%u,\n"
1531 "\t\t\"name\":\"%s\",\n"
1532 "\t\t\"chart\":\"%s\",\n"
1533 "\t\t\"family\":\"%s\",\n"
1534 "\t\t\"processed\":%s,\n"
1535 "\t\t\"updated\":%s,\n"
1536 "\t\t\"exec_run\":%s,\n"
1537 "\t\t\"exec_failed\":%s,\n"
1538 "\t\t\"exec\":\"%s\",\n"
1539 "\t\t\"recipient\":\"%s\",\n"
1540 "\t\t\"exec_code\":%d,\n"
1541 "\t\t\"source\":\"%s\",\n"
1542 "\t\t\"units\":\"%s\",\n"
1543 "\t\t\"info\":\"%s\",\n"
1544 "\t\t\"when\":%lu,\n"
1545 "\t\t\"duration\":%lu,\n"
1546 "\t\t\"non_clear_duration\":%lu,\n"
1547 "\t\t\"status\":\"%s\",\n"
1548 "\t\t\"old_status\":\"%s\",\n",
1555 (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_PROCESSED)?"true":"false",
1556 (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_UPDATED)?"true":"false",
1557 (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_EXEC_RUN)?"true":"false",
1558 (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_EXEC_FAILED)?"true":"false",
1559 ae->exec?ae->exec:health_default_exec,
1560 ae->recipient?ae->recipient:health_default_recipient,
1563 ae->units?ae->units:"",
1564 ae->info?ae->info:"",
1565 (unsigned long)ae->when,
1566 (unsigned long)ae->duration,
1567 (unsigned long)ae->non_clear_duration,
1568 rrdcalc_status2string(ae->new_status),
1569 rrdcalc_status2string(ae->old_status)
1572 buffer_strcat(wb, "\t\t\"value\":");
1573 buffer_rrd_value(wb, ae->new_value);
1574 buffer_strcat(wb, ",\n");
1576 buffer_strcat(wb, "\t\t\"old_value\":");
1577 buffer_rrd_value(wb, ae->old_value);
1578 buffer_strcat(wb, "\n");
1580 buffer_strcat(wb, "\t}");
1583 void health_alarm_log2json(RRDHOST *host, BUFFER *wb) {
1584 pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
1586 buffer_strcat(wb, "[");
1588 unsigned int max = host->health_log.max;
1589 unsigned int count = 0;
1591 for(ae = host->health_log.alarms; ae && count < max ; count++, ae = ae->next) {
1592 if(likely(count)) buffer_strcat(wb, ",");
1593 health_alarm_entry2json_nolock(wb, ae);
1596 buffer_strcat(wb, "\n]\n");
1598 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1601 static inline void health_rrdcalc2json_nolock(BUFFER *wb, RRDCALC *rc) {
1603 "\t\t\"%s.%s\": {\n"
1604 "\t\t\t\"name\": \"%s\",\n"
1605 "\t\t\t\"chart\": \"%s\",\n"
1606 "\t\t\t\"family\": \"%s\",\n"
1607 "\t\t\t\"active\": %s,\n"
1608 "\t\t\t\"exec\": \"%s\",\n"
1609 "\t\t\t\"recipient\": \"%s\",\n"
1610 "\t\t\t\"source\": \"%s\",\n"
1611 "\t\t\t\"units\": \"%s\",\n"
1612 "\t\t\t\"info\": \"%s\",\n"
1613 "\t\t\t\"status\": \"%s\",\n"
1614 "\t\t\t\"last_status_change\": %lu,\n"
1615 "\t\t\t\"last_updated\": %lu,\n"
1616 "\t\t\t\"next_update\": %lu,\n"
1617 "\t\t\t\"update_every\": %d,\n"
1618 , rc->chart, rc->name
1621 , (rc->rrdset && rc->rrdset->family)?rc->rrdset->family:""
1622 , (rc->rrdset)?"true":"false"
1623 , rc->exec?rc->exec:health_default_exec
1624 , rc->recipient?rc->recipient:health_default_recipient
1626 , rc->units?rc->units:""
1627 , rc->info?rc->info:""
1628 , rrdcalc_status2string(rc->status)
1629 , (unsigned long)rc->last_status_change
1630 , (unsigned long)rc->last_updated
1631 , (unsigned long)rc->next_update
1635 if(RRDCALC_HAS_DB_LOOKUP(rc)) {
1636 if(rc->dimensions && *rc->dimensions)
1637 health_string2json(wb, "\t\t\t", "lookup_dimensions", rc->dimensions, ",\n");
1640 "\t\t\t\"db_after\": %lu,\n"
1641 "\t\t\t\"db_before\": %lu,\n"
1642 "\t\t\t\"lookup_method\": \"%s\",\n"
1643 "\t\t\t\"lookup_after\": %d,\n"
1644 "\t\t\t\"lookup_before\": %d,\n"
1645 "\t\t\t\"lookup_options\": \"",
1646 (unsigned long) rc->db_after,
1647 (unsigned long) rc->db_before,
1648 group_method2string(rc->group),
1652 buffer_data_options2string(wb, rc->options);
1653 buffer_strcat(wb, "\",\n");
1656 if(rc->calculation) {
1657 health_string2json(wb, "\t\t\t", "calc", rc->calculation->source, ",\n");
1658 health_string2json(wb, "\t\t\t", "calc_parsed", rc->calculation->parsed_as, ",\n");
1662 health_string2json(wb, "\t\t\t", "warn", rc->warning->source, ",\n");
1663 health_string2json(wb, "\t\t\t", "warn_parsed", rc->warning->parsed_as, ",\n");
1667 health_string2json(wb, "\t\t\t", "crit", rc->critical->source, ",\n");
1668 health_string2json(wb, "\t\t\t", "crit_parsed", rc->critical->parsed_as, ",\n");
1671 buffer_strcat(wb, "\t\t\t\"green\":");
1672 buffer_rrd_value(wb, rc->green);
1673 buffer_strcat(wb, ",\n");
1675 buffer_strcat(wb, "\t\t\t\"red\":");
1676 buffer_rrd_value(wb, rc->red);
1677 buffer_strcat(wb, ",\n");
1679 buffer_strcat(wb, "\t\t\t\"value\":");
1680 buffer_rrd_value(wb, rc->value);
1681 buffer_strcat(wb, "\n");
1683 buffer_strcat(wb, "\t\t}");
1686 //void health_rrdcalctemplate2json_nolock(BUFFER *wb, RRDCALCTEMPLATE *rt) {
1690 void health_alarms2json(RRDHOST *host, BUFFER *wb, int all) {
1692 rrdhost_rdlock(&localhost);
1694 buffer_strcat(wb, "{\n\t\"alarms\": {\n");
1696 for(i = 0, rc = host->alarms; rc ; rc = rc->next) {
1697 if(unlikely(!rc->rrdset || !rc->rrdset->last_collected_time.tv_sec))
1700 if(likely(!all && !(rc->status == RRDCALC_STATUS_WARNING || rc->status == RRDCALC_STATUS_CRITICAL)))
1703 if(likely(i)) buffer_strcat(wb, ",\n");
1704 health_rrdcalc2json_nolock(wb, rc);
1708 // buffer_strcat(wb, "\n\t},\n\t\"templates\": {");
1710 // RRDCALCTEMPLATE *rt;
1711 // for(rt = host->templates; rt ; rt = rt->next)
1712 // health_rrdcalctemplate2json_nolock(wb, rt);
1714 buffer_sprintf(wb, "\n\t},\n\t\"now\": %lu\n}\n", (unsigned long)time(NULL));
1715 rrdhost_unlock(&localhost);
1719 // ----------------------------------------------------------------------------
1720 // re-load health configuration
1722 static inline void health_free_all_nolock(RRDHOST *host) {
1723 while(host->templates)
1724 rrdcalctemplate_free(host, host->templates);
1727 rrdcalc_free(host, host->alarms);
1730 void health_reload(void) {
1731 if(!health_enabled) {
1732 error("Health reload is requested, but health is not enabled.");
1736 char *path = health_config_dir();
1738 rrdhost_rwlock(&localhost);
1739 health_free_all_nolock(&localhost);
1740 rrdhost_unlock(&localhost);
1743 for(st = localhost.rrdset_root; st ; st = st->next) {
1748 rrdhost_rwlock(&localhost);
1749 health_readdir(path);
1750 rrdhost_unlock(&localhost);
1752 for(st = localhost.rrdset_root; st ; st = st->next) {
1753 rrdhost_rwlock(&localhost);
1755 rrdsetcalc_link_matching(st);
1756 rrdcalctemplate_link_matching(st);
1758 rrdhost_unlock(&localhost);
1763 // ----------------------------------------------------------------------------
1764 // health main thread and friends
1766 static inline int rrdcalc_value2status(calculated_number n) {
1767 if(isnan(n)) return RRDCALC_STATUS_UNDEFINED;
1768 if(n) return RRDCALC_STATUS_RAISED;
1769 return RRDCALC_STATUS_CLEAR;
1772 static inline void health_alarm_execute(RRDHOST *host, ALARM_ENTRY *ae) {
1773 if(ae->old_status == RRDCALC_STATUS_UNINITIALIZED && ae->new_status == RRDCALC_STATUS_CLEAR)
1776 char buffer[FILENAME_MAX + 1];
1779 const char *exec = ae->exec;
1780 if(!exec) exec = health_default_exec;
1782 const char *recipient = ae->recipient;
1783 if(!recipient) recipient = health_default_recipient;
1785 snprintfz(buffer, FILENAME_MAX, "exec %s '%s' '%s' '%u' '%u' '%u' '%zu' '%s' '%s' '%s' '%s' '%s' '%0.0Lf' '%0.0Lf' '%s' '%u' '%u' '%s' '%s'",
1794 ae->chart?ae->chart:"NOCAHRT",
1795 ae->family?ae->family:"NOFAMILY",
1796 rrdcalc_status2string(ae->new_status),
1797 rrdcalc_status2string(ae->old_status),
1800 ae->source?ae->source:"UNKNOWN",
1801 (uint32_t)ae->duration,
1802 (uint32_t)ae->non_clear_duration,
1803 ae->units?ae->units:"",
1804 ae->info?ae->info:""
1807 ae->notifications |= HEALTH_ENTRY_NOTIFICATIONS_EXEC_RUN;
1809 debug(D_HEALTH, "executing command '%s'", buffer);
1810 FILE *fp = mypopen(buffer, &command_pid);
1812 error("HEALTH: Cannot popen(\"%s\", \"r\").", buffer);
1815 debug(D_HEALTH, "HEALTH reading from command");
1816 char *s = fgets(buffer, FILENAME_MAX, fp);
1818 debug(D_HEALTH, "HEALTH closing command");
1819 ae->exec_code = mypclose(fp, command_pid);
1820 debug(D_HEALTH, "done executing command - returned with code %d", ae->exec_code);
1822 if(ae->exec_code != 0)
1823 ae->notifications |= HEALTH_ENTRY_NOTIFICATIONS_EXEC_FAILED;
1826 static inline void health_process_notifications(RRDHOST *host, ALARM_ENTRY *ae) {
1827 info("Health alarm '%s.%s' = %0.2Lf - changed status from %s to %s",
1828 ae->chart?ae->chart:"NOCHART", ae->name,
1830 rrdcalc_status2string(ae->old_status),
1831 rrdcalc_status2string(ae->new_status)
1834 health_alarm_execute(host, ae);
1837 static inline void health_alarm_log(RRDHOST *host,
1838 uint32_t alarm_id, uint32_t alarm_event_id,
1840 const char *name, const char *chart, const char *family,
1841 const char *exec, const char *recipient, time_t duration,
1842 calculated_number old_value, calculated_number new_value,
1843 int old_status, int new_status,
1848 ALARM_ENTRY *ae = callocz(1, sizeof(ALARM_ENTRY));
1849 ae->name = strdupz(name);
1850 ae->hash_name = simple_hash(ae->name);
1853 ae->chart = strdupz(chart);
1854 ae->hash_chart = simple_hash(ae->chart);
1858 ae->family = strdupz(family);
1860 if(exec) ae->exec = strdupz(exec);
1861 if(recipient) ae->recipient = strdupz(recipient);
1862 if(source) ae->source = strdupz(source);
1863 if(units) ae->units = strdupz(units);
1864 if(info) ae->info = strdupz(info);
1866 ae->unique_id = host->health_log.next_log_id++;
1867 ae->alarm_id = alarm_id;
1868 ae->alarm_event_id = alarm_event_id;
1870 ae->old_value = old_value;
1871 ae->new_value = new_value;
1872 ae->old_status = old_status;
1873 ae->new_status = new_status;
1874 ae->duration = duration;
1876 if(ae->old_status == RRDCALC_STATUS_WARNING || ae->old_status == RRDCALC_STATUS_CRITICAL)
1877 ae->non_clear_duration += ae->duration;
1880 pthread_rwlock_wrlock(&host->health_log.alarm_log_rwlock);
1881 ae->next = host->health_log.alarms;
1882 host->health_log.alarms = ae;
1883 host->health_log.count++;
1884 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1886 // match previous alarms
1887 pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
1889 for(t = host->health_log.alarms ; t ; t = t->next) {
1891 t->hash_name == ae->hash_name &&
1892 t->hash_chart == ae->hash_chart &&
1893 !strcmp(t->name, ae->name) &&
1894 t->chart && ae->chart && !strcmp(t->chart, ae->chart)) {
1896 if(!(t->notifications & HEALTH_ENTRY_NOTIFICATIONS_UPDATED) && !t->updated_by) {
1897 t->notifications |= HEALTH_ENTRY_NOTIFICATIONS_UPDATED;
1900 if((t->new_status == RRDCALC_STATUS_WARNING || t->new_status == RRDCALC_STATUS_CRITICAL) &&
1901 (t->old_status == RRDCALC_STATUS_WARNING || t->old_status == RRDCALC_STATUS_CRITICAL))
1902 ae->non_clear_duration += t->non_clear_duration;
1905 // no need to continue
1910 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1913 static inline void health_alarm_log_process(RRDHOST *host) {
1914 static uint32_t last_processed = 0;
1917 pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
1919 for(ae = host->health_log.alarms; ae ;ae = ae->next) {
1920 if(last_processed >= ae->unique_id) break;
1922 if(!(ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_PROCESSED) &&
1923 !(ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_UPDATED)) {
1924 ae->notifications |= HEALTH_ENTRY_NOTIFICATIONS_PROCESSED;
1925 health_process_notifications(host, ae);
1929 if(host->health_log.alarms)
1930 last_processed = host->health_log.alarms->unique_id;
1932 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1934 if(host->health_log.count <= host->health_log.max)
1937 // cleanup excess entries in the log
1938 pthread_rwlock_wrlock(&host->health_log.alarm_log_rwlock);
1940 ALARM_ENTRY *last = NULL;
1941 unsigned int count = host->health_log.max;
1942 for(ae = host->health_log.alarms; ae && count ; count--, last = ae, ae = ae->next) ;
1944 if(ae && last && last->next == ae)
1950 ALARM_ENTRY *t = ae->next;
1956 freez(ae->recipient);
1965 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1968 static inline int rrdcalc_isrunnable(RRDCALC *rc, time_t now, time_t *next_run) {
1969 if (unlikely(!rc->rrdset)) {
1970 debug(D_HEALTH, "Health not running alarm '%s.%s'. It is not linked to a chart.", rc->chart?rc->chart:"NOCHART", rc->name);
1974 if (unlikely(!rc->rrdset->last_collected_time.tv_sec)) {
1975 debug(D_HEALTH, "Health not running alarm '%s.%s'. Chart is not yet collected.", rc->chart?rc->chart:"NOCHART", rc->name);
1979 if (unlikely(!rc->update_every)) {
1980 debug(D_HEALTH, "Health not running alarm '%s.%s'. It does not have an update frequency", rc->chart?rc->chart:"NOCHART", rc->name);
1984 if (unlikely(rc->next_update > now)) {
1985 if (unlikely(*next_run > rc->next_update))
1986 *next_run = rc->next_update;
1988 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));
1993 // we should check that the DB lookup is possible
1995 // - the duration of the chart includes the required timeframe
1996 // we SHOULD NOT check the dimensions - there might be alarms that refer non-existing dimensions (e.g. cpu steal)
2001 void *health_main(void *ptr) {
2004 info("HEALTH thread created with task id %d", gettid());
2006 if(pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, NULL) != 0)
2007 error("Cannot set pthread cancel type to DEFERRED.");
2009 if(pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL) != 0)
2010 error("Cannot set pthread cancel state to ENABLE.");
2012 int min_run_every = (int)config_get_number("health", "run at least every seconds", 10);
2013 if(min_run_every < 1) min_run_every = 1;
2015 BUFFER *wb = buffer_create(100);
2017 unsigned int loop = 0;
2018 while(health_enabled) {
2020 debug(D_HEALTH, "Health monitoring iteration no %u started", loop);
2022 int oldstate, runnable = 0;
2023 time_t now = time(NULL);
2024 time_t next_run = now + min_run_every;
2027 if (unlikely(pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate) != 0))
2028 error("Cannot set pthread cancel state to DISABLE.");
2030 rrdhost_rdlock(&localhost);
2032 // the first loop is to lookup values from the db
2033 for (rc = localhost.alarms; rc; rc = rc->next) {
2034 if (unlikely(!rrdcalc_isrunnable(rc, now, &next_run)))
2038 rc->old_value = rc->value;
2040 // 1. if there is database lookup, do it
2041 // 2. if there is calculation expression, run it
2043 if (unlikely(RRDCALC_HAS_DB_LOOKUP(rc))) {
2044 time_t old_db_timestamp = rc->db_before;
2045 int value_is_null = 0;
2047 int ret = rrd2value(rc->rrdset, wb, &rc->value,
2048 rc->dimensions, 1, rc->after, rc->before, rc->group,
2049 rc->options, &rc->db_after, &rc->db_before, &value_is_null);
2051 if (unlikely(ret != 200)) {
2052 // database lookup failed
2055 debug(D_HEALTH, "Health alarm '%s.%s': database lookup returned error %d", rc->chart?rc->chart:"NOCHART", rc->name, ret);
2057 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_ERROR))) {
2058 rc->rrdcalc_flags |= RRDCALC_FLAG_DB_ERROR;
2059 error("Health alarm '%s.%s': database lookup returned error %d", rc->chart?rc->chart:"NOCHART", rc->name, ret);
2062 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_ERROR))
2063 rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_ERROR;
2065 if (unlikely(old_db_timestamp == rc->db_before)) {
2066 // database is stale
2068 debug(D_HEALTH, "Health alarm '%s.%s': database is stale", rc->chart?rc->chart:"NOCHART", rc->name);
2070 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_STALE))) {
2071 rc->rrdcalc_flags |= RRDCALC_FLAG_DB_STALE;
2072 error("Health alarm '%s.%s': database is stale", rc->chart?rc->chart:"NOCHART", rc->name);
2075 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_STALE))
2076 rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_STALE;
2078 if (unlikely(value_is_null)) {
2079 // collected value is null
2083 debug(D_HEALTH, "Health alarm '%s.%s': database lookup returned empty value (possibly value is not collected yet)",
2084 rc->chart?rc->chart:"NOCHART", rc->name);
2086 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_NAN))) {
2087 rc->rrdcalc_flags |= RRDCALC_FLAG_DB_NAN;
2088 error("Health alarm '%s.%s': database lookup returned empty value (possibly value is not collected yet)",
2089 rc->chart?rc->chart:"NOCHART", rc->name);
2092 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_NAN))
2093 rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_NAN;
2095 debug(D_HEALTH, "Health alarm '%s.%s': database lookup gave value "
2096 CALCULATED_NUMBER_FORMAT, rc->chart?rc->chart:"NOCHART", rc->name, rc->value);
2099 if(unlikely(rc->calculation)) {
2100 if (unlikely(!expression_evaluate(rc->calculation))) {
2101 // calculation failed
2105 debug(D_HEALTH, "Health alarm '%s.%s': failed to evaluate calculation with error: %s",
2106 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->calculation->error_msg));
2108 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_CALC_ERROR))) {
2109 rc->rrdcalc_flags |= RRDCALC_FLAG_CALC_ERROR;
2110 error("Health alarm '%s.%s': failed to evaluate calculation with error: %s",
2111 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->calculation->error_msg));
2115 if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_CALC_ERROR))
2116 rc->rrdcalc_flags &= ~RRDCALC_FLAG_CALC_ERROR;
2118 debug(D_HEALTH, "Health alarm '%s.%s': calculation expression gave value "
2119 CALCULATED_NUMBER_FORMAT
2120 ": %s (source: %s)",
2121 rc->chart?rc->chart:"NOCHART", rc->name,
2122 rc->calculation->result,
2123 buffer_tostring(rc->calculation->error_msg),
2127 rc->value = rc->calculation->result;
2131 rrdhost_unlock(&localhost);
2134 rrdhost_rdlock(&localhost);
2136 for (rc = localhost.alarms; rc; rc = rc->next) {
2137 if (unlikely(!rrdcalc_isrunnable(rc, now, &next_run)))
2140 int warning_status = RRDCALC_STATUS_UNDEFINED;
2141 int critical_status = RRDCALC_STATUS_UNDEFINED;
2143 if(likely(rc->warning)) {
2144 if(unlikely(!expression_evaluate(rc->warning))) {
2145 // calculation failed
2147 debug(D_HEALTH, "Health alarm '%s.%s': warning expression failed with error: %s",
2148 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->warning->error_msg));
2150 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_WARN_ERROR))) {
2151 rc->rrdcalc_flags |= RRDCALC_FLAG_WARN_ERROR;
2152 error("Health alarm '%s.%s': warning expression failed with error: %s",
2153 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->warning->error_msg));
2157 if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_WARN_ERROR))
2158 rc->rrdcalc_flags &= ~RRDCALC_FLAG_WARN_ERROR;
2160 debug(D_HEALTH, "Health alarm '%s.%s': warning expression gave value "
2161 CALCULATED_NUMBER_FORMAT
2162 ": %s (source: %s)",
2163 rc->chart?rc->chart:"NOCHART", rc->name,
2164 rc->warning->result,
2165 buffer_tostring(rc->warning->error_msg),
2169 warning_status = rrdcalc_value2status(rc->warning->result);
2173 if(likely(rc->critical)) {
2174 if(unlikely(!expression_evaluate(rc->critical))) {
2175 // calculation failed
2177 debug(D_HEALTH, "Health alarm '%s.%s': critical expression failed with error: %s",
2178 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->critical->error_msg));
2180 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_CRIT_ERROR))) {
2181 rc->rrdcalc_flags |= RRDCALC_FLAG_CRIT_ERROR;
2182 error("Health alarm '%s.%s': critical expression failed with error: %s",
2183 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->critical->error_msg));
2187 if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_CRIT_ERROR))
2188 rc->rrdcalc_flags &= ~RRDCALC_FLAG_CRIT_ERROR;
2190 debug(D_HEALTH, "Health alarm '%s.%s': critical expression gave value "
2191 CALCULATED_NUMBER_FORMAT
2192 ": %s (source: %s)",
2193 rc->chart?rc->chart:"NOCHART", rc->name,
2194 rc->critical->result,
2195 buffer_tostring(rc->critical->error_msg),
2199 critical_status = rrdcalc_value2status(rc->critical->result);
2203 int status = RRDCALC_STATUS_UNDEFINED;
2205 switch(warning_status) {
2206 case RRDCALC_STATUS_CLEAR:
2207 status = RRDCALC_STATUS_CLEAR;
2210 case RRDCALC_STATUS_RAISED:
2211 status = RRDCALC_STATUS_WARNING;
2218 switch(critical_status) {
2219 case RRDCALC_STATUS_CLEAR:
2220 if(status == RRDCALC_STATUS_UNDEFINED)
2221 status = RRDCALC_STATUS_CLEAR;
2224 case RRDCALC_STATUS_RAISED:
2225 status = RRDCALC_STATUS_CRITICAL;
2232 if(status != rc->status) {
2233 health_alarm_log(&localhost, rc->id, rc->next_event_id++, time(NULL), 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);
2234 rc->last_status_change = now;
2235 rc->status = status;
2238 rc->last_updated = now;
2239 rc->next_update = now + rc->update_every;
2241 if (next_run > rc->next_update)
2242 next_run = rc->next_update;
2245 rrdhost_unlock(&localhost);
2248 if (unlikely(pthread_setcancelstate(oldstate, NULL) != 0))
2249 error("Cannot set pthread cancel state to RESTORE (%d).", oldstate);
2251 // execute notifications
2253 health_alarm_log_process(&localhost);
2256 if(now < next_run) {
2257 debug(D_HEALTH, "Health monitoring iteration no %u done. Next iteration in %d secs",
2258 loop, (int) (next_run - now));
2259 sleep_usec(1000000 * (unsigned long long) (next_run - now));
2262 debug(D_HEALTH, "Health monitoring iteration no %u done. Next iteration now", loop);
2268 info("HEALTH thread exiting");