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 rc->rrdset_next->rrdset_prev = rc;
451 if(rc->update_every < rc->rrdset->update_every) {
452 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);
453 rc->update_every = rc->rrdset->update_every;
456 if(!isnan(rc->green) && isnan(st->green)) {
457 debug(D_HEALTH, "Health alarm '%s.%s' green threshold set from %Lf to %Lf.", rc->rrdset->id, rc->name, rc->rrdset->green, rc->green);
458 st->green = rc->green;
461 if(!isnan(rc->red) && isnan(st->red)) {
462 debug(D_HEALTH, "Health alarm '%s.%s' red threshold set from %Lf to %Lf.", rc->rrdset->id, rc->name, rc->rrdset->red, rc->red);
466 rc->local = rrdvar_create_and_index("local", &st->variables_root_index, rc->name, RRDVAR_TYPE_CALCULATED, &rc->value);
467 rc->family = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rc->name, RRDVAR_TYPE_CALCULATED, &rc->value);
469 char fullname[RRDVAR_MAX_LENGTH + 1];
470 snprintfz(fullname, RRDVAR_MAX_LENGTH, "%s.%s", st->id, rc->name);
471 rc->hostid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, fullname, RRDVAR_TYPE_CALCULATED, &rc->value);
473 snprintfz(fullname, RRDVAR_MAX_LENGTH, "%s.%s", st->name, rc->name);
474 rc->hostname = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, fullname, RRDVAR_TYPE_CALCULATED, &rc->value);
476 if(!rc->units) rc->units = strdupz(st->units);
479 static inline int rrdcalc_is_matching_this_rrdset(RRDCALC *rc, RRDSET *st) {
480 if( (rc->hash_chart == st->hash && !strcmp(rc->chart, st->id)) ||
481 (rc->hash_chart == st->hash_name && !strcmp(rc->chart, st->name)))
487 // this has to be called while the RRDHOST is locked
488 inline void rrdsetcalc_link_matching(RRDSET *st) {
489 // debug(D_HEALTH, "find matching alarms for chart '%s'", st->id);
492 for(rc = st->rrdhost->alarms; rc ; rc = rc->next) {
493 if(unlikely(rc->rrdset))
496 if(unlikely(rrdcalc_is_matching_this_rrdset(rc, st)))
497 rrdsetcalc_link(st, rc);
501 // this has to be called while the RRDHOST is locked
502 inline void rrdsetcalc_unlink(RRDCALC *rc) {
503 RRDSET *st = rc->rrdset;
506 error("Requested to unlink RRDCALC '%s.%s' which is not linked to any RRDSET", rc->chart?rc->chart:"NOCHART", rc->name);
510 RRDHOST *host = st->rrdhost;
512 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);
516 rc->rrdset_prev->rrdset_next = rc->rrdset_next;
519 rc->rrdset_next->rrdset_prev = rc->rrdset_prev;
522 st->alarms = rc->rrdset_next;
524 rc->rrdset_prev = rc->rrdset_next = NULL;
526 rrdvar_free(st->rrdhost, &st->variables_root_index, rc->local);
529 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rc->family);
532 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rc->hostid);
535 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rc->hostname);
540 // RRDCALC will remain in RRDHOST
541 // so that if the matching chart is found in the future
542 // it will be applied automatically
545 RRDCALC *rrdcalc_find(RRDSET *st, const char *name) {
547 uint32_t hash = simple_hash(name);
549 for( rc = st->alarms; rc ; rc = rc->rrdset_next ) {
550 if(unlikely(rc->hash == hash && !strcmp(rc->name, name)))
557 static inline int rrdcalc_exists(RRDHOST *host, const char *name, uint32_t hash) {
560 // make sure it does not already exist
561 for(rc = host->alarms; rc ; rc = rc->next) {
562 if (unlikely(rc->hash == hash && !strcmp(name, rc->name))) {
563 error("Health alarm '%s' already exists in host '%s'.", name, host->hostname);
571 static inline void rrdcalc_create_part2(RRDHOST *host, RRDCALC *rc) {
572 rrdhost_check_rdlock(host);
574 if(rc->calculation) {
575 rc->calculation->this = &rc->value;
576 rc->calculation->after = &rc->db_after;
577 rc->calculation->before = &rc->db_before;
578 rc->calculation->rrdcalc = rc;
582 rc->warning->this = &rc->value;
583 rc->warning->after = &rc->db_after;
584 rc->warning->before = &rc->db_before;
585 rc->warning->rrdcalc = rc;
589 rc->critical->this = &rc->value;
590 rc->critical->after = &rc->db_after;
591 rc->critical->before = &rc->db_before;
592 rc->critical->rrdcalc = rc;
595 // link it to the host
596 if(likely(host->alarms)) {
599 for(t = host->alarms; t && t->next ; t = t->next) ;
606 // link it to its chart
608 for(st = host->rrdset_root; st ; st = st->next) {
609 if(rrdcalc_is_matching_this_rrdset(rc, st)) {
610 rrdsetcalc_link(st, rc);
616 static inline uint32_t rrdcalc_fullname(char *fullname, size_t len, const char *chart, const char *name) {
617 snprintfz(fullname, len - 1, "%s%s%s", chart?chart:"", chart?".":"", name);
618 rrdvar_fix_name(fullname);
619 return simple_hash(fullname);
622 static inline RRDCALC *rrdcalc_create(RRDHOST *host, const char *name, const char *chart, const char *dimensions,
623 const char *units, const char *info,
624 int group_method, int after, int before, int update_every, uint32_t options,
625 calculated_number green, calculated_number red,
626 const char *exec, const char *recipient, const char *source,
627 const char *calc, const char *warn, const char *crit) {
629 char fullname[RRDVAR_MAX_LENGTH + 1];
630 uint32_t hash = rrdcalc_fullname(fullname, RRDVAR_MAX_LENGTH + 1, chart, name);
632 if(rrdcalc_exists(host, fullname, hash))
635 RRDCALC *rc = callocz(1, sizeof(RRDCALC));
636 rc->id = host->health_log.next_alarm_id++;
637 rc->next_event_id = 1;
638 rc->name = strdupz(name);
639 rc->hash = simple_hash(rc->name);
640 rc->chart = strdupz(chart);
641 rc->hash_chart = simple_hash(rc->chart);
643 if(dimensions) rc->dimensions = strdupz(dimensions);
650 rc->group = group_method;
653 rc->update_every = update_every;
654 rc->options = options;
656 if(exec) rc->exec = strdupz(exec);
657 if(recipient) rc->recipient = strdupz(recipient);
658 if(source) rc->source = strdupz(source);
659 if(units) rc->units = strdupz(units);
660 if(info) rc->info = strdupz(info);
663 rc->calculation = expression_parse(calc, NULL, NULL);
665 error("Health alarm '%s.%s': failed to parse calculation expression '%s'", chart, name, calc);
668 rc->warning = expression_parse(warn, NULL, NULL);
670 error("Health alarm '%s.%s': failed to re-parse warning expression '%s'", chart, name, warn);
673 rc->critical = expression_parse(crit, NULL, NULL);
675 error("Health alarm '%s.%s': failed to re-parse critical expression '%s'", chart, name, crit);
678 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",
679 (rc->chart)?rc->chart:"NOCHART",
681 (rc->exec)?rc->exec:"DEFAULT",
682 (rc->recipient)?rc->recipient:"DEFAULT",
689 (rc->dimensions)?rc->dimensions:"NONE",
691 (rc->calculation)?rc->calculation->parsed_as:"NONE",
692 (rc->warning)?rc->warning->parsed_as:"NONE",
693 (rc->critical)?rc->critical->parsed_as:"NONE",
697 rrdcalc_create_part2(host, rc);
701 void rrdcalc_free(RRDHOST *host, RRDCALC *rc) {
704 debug(D_HEALTH, "Health removing alarm '%s.%s' of host '%s'", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);
706 // unlink it from RRDSET
707 if(rc->rrdset) rrdsetcalc_unlink(rc);
709 // unlink it from RRDHOST
710 if(unlikely(rc == host->alarms))
711 host->alarms = rc->next;
713 else if(likely(host->alarms)) {
714 RRDCALC *t, *last = host->alarms;
715 for(t = last->next; t && t != rc; last = t, t = t->next) ;
716 if(last && last->next == rc)
717 last->next = rc->next;
719 error("Cannot unlink alarm '%s.%s' from host '%s': not found", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);
722 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);
724 expression_free(rc->calculation);
725 expression_free(rc->warning);
726 expression_free(rc->critical);
731 freez(rc->dimensions);
733 freez(rc->recipient);
740 // ----------------------------------------------------------------------------
741 // RRDCALCTEMPLATE management
743 void rrdcalctemplate_link_matching(RRDSET *st) {
746 for(rt = st->rrdhost->templates; rt ; rt = rt->next) {
747 if(rt->hash_context == st->hash_context && !strcmp(rt->context, st->context)) {
748 RRDCALC *rc = rrdcalc_create(st->rrdhost, rt->name, st->id,
749 rt->dimensions, rt->units, rt->info, rt->group, rt->after, rt->before, rt->update_every, rt->options,
751 rt->exec, rt->recipient, rt->source,
752 (rt->calculation)?rt->calculation->source:NULL,
753 (rt->warning)?rt->warning->source:NULL,
754 (rt->critical)?rt->critical->source:NULL);
757 error("Health tried to create alarm from template '%s', but it failed", rt->name);
759 #ifdef NETDATA_INTERNAL_CHECKS
760 else if(rc->rrdset != st)
761 error("Health alarm '%s.%s' should be linked to chart '%s', but it is not", rc->chart?rc->chart:"NOCHART", rc->name, st->id);
767 static inline void rrdcalctemplate_free(RRDHOST *host, RRDCALCTEMPLATE *rt) {
768 debug(D_HEALTH, "Health removing template '%s' of host '%s'", rt->name, host->hostname);
770 if(host->templates) {
771 if(host->templates == rt) {
772 host->templates = rt->next;
775 RRDCALCTEMPLATE *t, *last = host->templates;
776 for (t = last->next; t && t != rt; last = t, t = t->next ) ;
777 if(last && last->next == rt) {
778 last->next = rt->next;
782 error("Cannot find RRDCALCTEMPLATE '%s' linked in host '%s'", rt->name, host->hostname);
786 expression_free(rt->calculation);
787 expression_free(rt->warning);
788 expression_free(rt->critical);
792 freez(rt->recipient);
797 freez(rt->dimensions);
801 // ----------------------------------------------------------------------------
802 // load health configuration
804 #define HEALTH_CONF_MAX_LINE 4096
806 #define HEALTH_ALARM_KEY "alarm"
807 #define HEALTH_TEMPLATE_KEY "template"
808 #define HEALTH_ON_KEY "on"
809 #define HEALTH_LOOKUP_KEY "lookup"
810 #define HEALTH_CALC_KEY "calc"
811 #define HEALTH_EVERY_KEY "every"
812 #define HEALTH_GREEN_KEY "green"
813 #define HEALTH_RED_KEY "red"
814 #define HEALTH_WARN_KEY "warn"
815 #define HEALTH_CRIT_KEY "crit"
816 #define HEALTH_EXEC_KEY "exec"
817 #define HEALTH_RECIPIENT_KEY "to"
818 #define HEALTH_UNITS_KEY "units"
819 #define HEALTH_INFO_KEY "info"
821 static inline int rrdcalc_add_alarm_from_config(RRDHOST *host, RRDCALC *rc) {
823 char fullname[RRDVAR_MAX_LENGTH + 1];
824 uint32_t hash = rrdcalc_fullname(fullname, RRDVAR_MAX_LENGTH + 1, rc->chart, rc->name);
826 if (rrdcalc_exists(host, fullname, hash))
831 error("Health configuration for alarm '%s' does not have a chart", rc->name);
835 if(!rc->update_every) {
836 error("Health configuration for alarm '%s.%s' has no frequency (parameter 'every'). Ignoring it.", rc->chart?rc->chart:"NOCHART", rc->name);
840 if(!RRDCALC_HAS_DB_LOOKUP(rc) && !rc->warning && !rc->critical) {
841 error("Health configuration for alarm '%s.%s' is useless (no calculation, no warning and no critical evaluation)", rc->chart?rc->chart:"NOCHART", rc->name);
845 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",
846 rc->chart?rc->chart:"NOCHART",
848 (rc->exec)?rc->exec:"DEFAULT",
849 (rc->recipient)?rc->recipient:"DEFAULT",
856 (rc->dimensions)?rc->dimensions:"NONE",
858 (rc->calculation)?rc->calculation->parsed_as:"NONE",
859 (rc->warning)?rc->warning->parsed_as:"NONE",
860 (rc->critical)?rc->critical->parsed_as:"NONE",
864 rrdcalc_create_part2(host, rc);
868 static inline int rrdcalctemplate_add_template_from_config(RRDHOST *host, RRDCALCTEMPLATE *rt) {
869 if(unlikely(!rt->context)) {
870 error("Health configuration for template '%s' does not have a context", rt->name);
874 if(unlikely(!rt->update_every)) {
875 error("Health configuration for template '%s' has no frequency (parameter 'every'). Ignoring it.", rt->name);
879 if(unlikely(!RRDCALCTEMPLATE_HAS_CALCULATION(rt) && !rt->warning && !rt->critical)) {
880 error("Health configuration for template '%s' is useless (no calculation, no warning and no critical evaluation)", rt->name);
884 RRDCALCTEMPLATE *t, *last = NULL;
885 for (t = host->templates; t ; last = t, t = t->next) {
886 if(unlikely(t->hash_name == rt->hash_name && !strcmp(t->name, rt->name))) {
887 error("Health configuration template '%s' already exists for host '%s'.", rt->name, host->hostname);
892 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'",
894 (rt->context)?rt->context:"NONE",
895 (rt->exec)?rt->exec:"DEFAULT",
896 (rt->recipient)?rt->recipient:"DEFAULT",
903 (rt->dimensions)?rt->dimensions:"NONE",
905 (rt->calculation)?rt->calculation->parsed_as:"NONE",
906 (rt->warning)?rt->warning->parsed_as:"NONE",
907 (rt->critical)?rt->critical->parsed_as:"NONE",
915 rt->next = host->templates;
916 host->templates = rt;
922 static inline int health_parse_duration(char *string, int *result) {
923 // make sure it is a number
924 if(!*string || !(isdigit(*string) || *string == '+' || *string == '-')) {
930 calculated_number n = strtold(string, &e);
934 *result = (int) (n * 86400 * 365);
937 *result = (int) (n * 86400 * 30);
940 *result = (int) (n * 86400 * 7);
943 *result = (int) (n * 86400);
946 *result = (int) (n * 3600);
949 *result = (int) (n * 60);
964 static inline int health_parse_db_lookup(
965 size_t line, const char *path, const char *file, char *string,
966 int *group_method, int *after, int *before, int *every,
967 uint32_t *options, char **dimensions
969 debug(D_HEALTH, "Health configuration parsing database lookup %zu@%s/%s: %s", line, path, file, string);
971 if(*dimensions) freez(*dimensions);
978 char *s = string, *key;
980 // first is the group method
982 while(*s && !isspace(*s)) s++;
983 while(*s && isspace(*s)) *s++ = '\0';
985 error("Health configuration invalid chart calculation at line %zu of file '%s/%s': expected group method followed by the 'after' time, but got '%s'",
986 line, path, file, key);
990 if((*group_method = web_client_api_request_v1_data_group(key, -1)) == -1) {
991 error("Health configuration at line %zu of file '%s/%s': invalid group method '%s'",
992 line, path, file, key);
996 // then is the 'after' time
998 while(*s && !isspace(*s)) s++;
999 while(*s && isspace(*s)) *s++ = '\0';
1001 if(!health_parse_duration(key, after)) {
1002 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' after group method",
1003 line, path, file, key);
1008 *every = abs(*after);
1010 // now we may have optional parameters
1013 while(*s && !isspace(*s)) s++;
1014 while(*s && isspace(*s)) *s++ = '\0';
1017 if(!strcasecmp(key, "at")) {
1019 while(*s && !isspace(*s)) s++;
1020 while(*s && isspace(*s)) *s++ = '\0';
1022 if (!health_parse_duration(value, before)) {
1023 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' for '%s' keyword",
1024 line, path, file, value, key);
1027 else if(!strcasecmp(key, HEALTH_EVERY_KEY)) {
1029 while(*s && !isspace(*s)) s++;
1030 while(*s && isspace(*s)) *s++ = '\0';
1032 if (!health_parse_duration(value, every)) {
1033 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' for '%s' keyword",
1034 line, path, file, value, key);
1037 else if(!strcasecmp(key, "absolute") || !strcasecmp(key, "abs") || !strcasecmp(key, "absolute_sum")) {
1038 *options |= RRDR_OPTION_ABSOLUTE;
1040 else if(!strcasecmp(key, "min2max")) {
1041 *options |= RRDR_OPTION_MIN2MAX;
1043 else if(!strcasecmp(key, "null2zero")) {
1044 *options |= RRDR_OPTION_NULL2ZERO;
1046 else if(!strcasecmp(key, "percentage")) {
1047 *options |= RRDR_OPTION_PERCENTAGE;
1049 else if(!strcasecmp(key, "unaligned")) {
1050 *options |= RRDR_OPTION_NOT_ALIGNED;
1052 else if(!strcasecmp(key, "of")) {
1053 if(*s && strcasecmp(s, "all"))
1054 *dimensions = strdupz(s);
1058 error("Health configuration at line %zu of file '%s/%s': unknown keyword '%s'",
1059 line, path, file, key);
1066 static inline char *health_source_file(size_t line, const char *path, const char *filename) {
1067 char buffer[FILENAME_MAX + 1];
1068 snprintfz(buffer, FILENAME_MAX, "%zu@%s/%s", line, path, filename);
1069 return strdupz(buffer);
1072 static inline void strip_quotes(char *s) {
1074 if(*s == '\'' || *s == '"') *s = ' ';
1079 int health_readfile(const char *path, const char *filename) {
1080 debug(D_HEALTH, "Health configuration reading file '%s/%s'", path, filename);
1082 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;
1083 char buffer[HEALTH_CONF_MAX_LINE + 1];
1085 if(unlikely(!hash_alarm)) {
1086 hash_alarm = simple_uhash(HEALTH_ALARM_KEY);
1087 hash_template = simple_uhash(HEALTH_TEMPLATE_KEY);
1088 hash_on = simple_uhash(HEALTH_ON_KEY);
1089 hash_calc = simple_uhash(HEALTH_CALC_KEY);
1090 hash_lookup = simple_uhash(HEALTH_LOOKUP_KEY);
1091 hash_green = simple_uhash(HEALTH_GREEN_KEY);
1092 hash_red = simple_uhash(HEALTH_RED_KEY);
1093 hash_warn = simple_uhash(HEALTH_WARN_KEY);
1094 hash_crit = simple_uhash(HEALTH_CRIT_KEY);
1095 hash_exec = simple_uhash(HEALTH_EXEC_KEY);
1096 hash_every = simple_uhash(HEALTH_EVERY_KEY);
1097 hash_units = simple_hash(HEALTH_UNITS_KEY);
1098 hash_info = simple_hash(HEALTH_INFO_KEY);
1099 hash_recipient = simple_hash(HEALTH_RECIPIENT_KEY);
1102 snprintfz(buffer, HEALTH_CONF_MAX_LINE, "%s/%s", path, filename);
1103 FILE *fp = fopen(buffer, "r");
1105 error("Health configuration cannot read file '%s'.", buffer);
1110 RRDCALCTEMPLATE *rt = NULL;
1112 size_t line = 0, append = 0;
1114 while((s = fgets(&buffer[append], (int)(HEALTH_CONF_MAX_LINE - append), fp)) || append) {
1115 int stop_appending = !s;
1117 // info("Line %zu of file '%s/%s': '%s'", line, path, filename, s);
1120 // info("Trimmed line %zu of file '%s/%s': '%s'", line, path, filename, s);
1123 if(!stop_appending && s[append - 1] == '\\') {
1124 s[append - 1] = ' ';
1125 append = &s[append] - buffer;
1126 if(append < HEALTH_CONF_MAX_LINE)
1129 error("Health configuration has too long muli-line at line %zu of file '%s/%s'.", line, path, filename);
1135 while(*s && *s != ':') s++;
1137 error("Health configuration has invalid line %zu of file '%s/%s'. It does not contain a ':'. Ignoring it.", line, path, filename);
1145 value = trim(value);
1148 error("Health configuration has invalid line %zu of file '%s/%s'. Keyword is empty. Ignoring it.", line, path, filename);
1153 error("Health configuration has invalid line %zu of file '%s/%s'. value is empty. Ignoring it.", line, path, filename);
1157 // info("Health file '%s/%s', key '%s', value '%s'", path, filename, key, value);
1158 uint32_t hash = simple_uhash(key);
1160 if(hash == hash_alarm && !strcasecmp(key, HEALTH_ALARM_KEY)) {
1161 if(rc && !rrdcalc_add_alarm_from_config(&localhost, rc))
1162 rrdcalc_free(&localhost, rc);
1165 if (!rrdcalctemplate_add_template_from_config(&localhost, rt))
1166 rrdcalctemplate_free(&localhost, rt);
1170 rc = callocz(1, sizeof(RRDCALC));
1171 rc->id = localhost.health_log.next_alarm_id++;
1172 rc->next_event_id = 1;
1173 rc->name = strdupz(value);
1174 rc->hash = simple_hash(rc->name);
1175 rc->source = health_source_file(line, path, filename);
1179 rc->old_value = NAN;
1181 if(rrdvar_fix_name(rc->name))
1182 error("Health configuration renamed alarm '%s' to '%s'", value, rc->name);
1184 else if(hash == hash_template && !strcasecmp(key, HEALTH_TEMPLATE_KEY)) {
1186 if(!rrdcalc_add_alarm_from_config(&localhost, rc))
1187 rrdcalc_free(&localhost, rc);
1191 if(rt && !rrdcalctemplate_add_template_from_config(&localhost, rt))
1192 rrdcalctemplate_free(&localhost, rt);
1194 rt = callocz(1, sizeof(RRDCALCTEMPLATE));
1195 rt->name = strdupz(value);
1196 rt->hash_name = simple_hash(rt->name);
1197 rt->source = health_source_file(line, path, filename);
1201 if(rrdvar_fix_name(rt->name))
1202 error("Health configuration renamed template '%s' to '%s'", value, rt->name);
1205 if(hash == hash_on && !strcasecmp(key, HEALTH_ON_KEY)) {
1207 if(strcmp(rc->chart, value))
1208 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').",
1209 line, path, filename, rc->name, key, rc->chart, value, value);
1213 rc->chart = strdupz(value);
1214 rc->hash_chart = simple_hash(rc->chart);
1216 else if(hash == hash_lookup && !strcasecmp(key, HEALTH_LOOKUP_KEY)) {
1217 health_parse_db_lookup(line, path, filename, value, &rc->group, &rc->after, &rc->before,
1219 &rc->options, &rc->dimensions);
1221 else if(hash == hash_every && !strcasecmp(key, HEALTH_EVERY_KEY)) {
1222 if(!health_parse_duration(value, &rc->update_every))
1223 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' cannot parse duration: '%s'.",
1224 line, path, filename, rc->name, key, value);
1226 else if(hash == hash_green && !strcasecmp(key, HEALTH_GREEN_KEY)) {
1228 rc->green = strtold(value, &e);
1230 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' leaves this string unmatched: '%s'.",
1231 line, path, filename, rc->name, key, e);
1234 else if(hash == hash_red && !strcasecmp(key, HEALTH_RED_KEY)) {
1236 rc->red = strtold(value, &e);
1238 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' leaves this string unmatched: '%s'.",
1239 line, path, filename, rc->name, key, e);
1242 else if(hash == hash_calc && !strcasecmp(key, HEALTH_CALC_KEY)) {
1243 const char *failed_at = NULL;
1245 rc->calculation = expression_parse(value, &failed_at, &error);
1246 if(!rc->calculation) {
1247 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1248 line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
1251 else if(hash == hash_warn && !strcasecmp(key, HEALTH_WARN_KEY)) {
1252 const char *failed_at = NULL;
1254 rc->warning = expression_parse(value, &failed_at, &error);
1256 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1257 line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
1260 else if(hash == hash_crit && !strcasecmp(key, HEALTH_CRIT_KEY)) {
1261 const char *failed_at = NULL;
1263 rc->critical = expression_parse(value, &failed_at, &error);
1265 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1266 line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
1269 else if(hash == hash_exec && !strcasecmp(key, HEALTH_EXEC_KEY)) {
1271 if(strcmp(rc->exec, value))
1272 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').",
1273 line, path, filename, rc->name, key, rc->exec, value, value);
1277 rc->exec = strdupz(value);
1279 else if(hash == hash_recipient && !strcasecmp(key, HEALTH_RECIPIENT_KEY)) {
1281 if(strcmp(rc->recipient, value))
1282 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').",
1283 line, path, filename, rc->name, key, rc->recipient, value, value);
1285 freez(rc->recipient);
1287 rc->recipient = strdupz(value);
1289 else if(hash == hash_units && !strcasecmp(key, HEALTH_UNITS_KEY)) {
1291 if(strcmp(rc->units, value))
1292 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').",
1293 line, path, filename, rc->name, key, rc->units, value, value);
1297 rc->units = strdupz(value);
1298 strip_quotes(rc->units);
1300 else if(hash == hash_info && !strcasecmp(key, HEALTH_INFO_KEY)) {
1302 if(strcmp(rc->info, value))
1303 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').",
1304 line, path, filename, rc->name, key, rc->info, value, value);
1308 rc->info = strdupz(value);
1309 strip_quotes(rc->info);
1312 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' has unknown key '%s'.",
1313 line, path, filename, rc->name, key);
1317 if(hash == hash_on && !strcasecmp(key, HEALTH_ON_KEY)) {
1319 if(strcmp(rt->context, value))
1320 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').",
1321 line, path, filename, rt->name, key, rt->context, value, value);
1325 rt->context = strdupz(value);
1326 rt->hash_context = simple_hash(rt->context);
1328 else if(hash == hash_lookup && !strcasecmp(key, HEALTH_LOOKUP_KEY)) {
1329 health_parse_db_lookup(line, path, filename, value, &rt->group, &rt->after, &rt->before,
1331 &rt->options, &rt->dimensions);
1333 else if(hash == hash_every && !strcasecmp(key, HEALTH_EVERY_KEY)) {
1334 if(!health_parse_duration(value, &rt->update_every))
1335 info("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' cannot parse duration: '%s'.",
1336 line, path, filename, rt->name, key, value);
1338 else if(hash == hash_green && !strcasecmp(key, HEALTH_GREEN_KEY)) {
1340 rt->green = strtold(value, &e);
1342 info("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' leaves this string unmatched: '%s'.",
1343 line, path, filename, rt->name, key, e);
1346 else if(hash == hash_red && !strcasecmp(key, HEALTH_RED_KEY)) {
1348 rt->red = strtold(value, &e);
1350 info("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' leaves this string unmatched: '%s'.",
1351 line, path, filename, rt->name, key, e);
1354 else if(hash == hash_calc && !strcasecmp(key, HEALTH_CALC_KEY)) {
1355 const char *failed_at = NULL;
1357 rt->calculation = expression_parse(value, &failed_at, &error);
1358 if(!rt->calculation) {
1359 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1360 line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
1363 else if(hash == hash_warn && !strcasecmp(key, HEALTH_WARN_KEY)) {
1364 const char *failed_at = NULL;
1366 rt->warning = expression_parse(value, &failed_at, &error);
1368 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1369 line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
1372 else if(hash == hash_crit && !strcasecmp(key, HEALTH_CRIT_KEY)) {
1373 const char *failed_at = NULL;
1375 rt->critical = expression_parse(value, &failed_at, &error);
1377 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1378 line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
1381 else if(hash == hash_exec && !strcasecmp(key, HEALTH_EXEC_KEY)) {
1383 if(strcmp(rt->exec, value))
1384 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').",
1385 line, path, filename, rt->name, key, rt->exec, value, value);
1389 rt->exec = strdupz(value);
1391 else if(hash == hash_recipient && !strcasecmp(key, HEALTH_RECIPIENT_KEY)) {
1393 if(strcmp(rt->recipient, value))
1394 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').",
1395 line, path, filename, rt->name, key, rt->recipient, value, value);
1397 freez(rt->recipient);
1399 rt->recipient = strdupz(value);
1401 else if(hash == hash_units && !strcasecmp(key, HEALTH_UNITS_KEY)) {
1403 if(strcmp(rt->units, value))
1404 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').",
1405 line, path, filename, rt->name, key, rt->units, value, value);
1409 rt->units = strdupz(value);
1410 strip_quotes(rt->units);
1412 else if(hash == hash_info && !strcasecmp(key, HEALTH_INFO_KEY)) {
1414 if(strcmp(rt->info, value))
1415 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').",
1416 line, path, filename, rt->name, key, rt->info, value, value);
1420 rt->info = strdupz(value);
1421 strip_quotes(rt->info);
1424 error("Health configuration at line %zu of file '%s/%s' for template '%s' has unknown key '%s'.",
1425 line, path, filename, rt->name, key);
1429 error("Health configuration at line %zu of file '%s/%s' has unknown key '%s'. Expected either '" HEALTH_ALARM_KEY "' or '" HEALTH_TEMPLATE_KEY "'.",
1430 line, path, filename, key);
1434 if(rc && !rrdcalc_add_alarm_from_config(&localhost, rc))
1435 rrdcalc_free(&localhost, rc);
1437 if(rt && !rrdcalctemplate_add_template_from_config(&localhost, rt))
1438 rrdcalctemplate_free(&localhost, rt);
1444 void health_readdir(const char *path) {
1445 size_t pathlen = strlen(path);
1447 debug(D_HEALTH, "Health configuration reading directory '%s'", path);
1449 DIR *dir = opendir(path);
1451 error("Health configuration cannot open directory '%s'.", path);
1455 struct dirent *de = NULL;
1456 while ((de = readdir(dir))) {
1457 size_t len = strlen(de->d_name);
1459 if(de->d_type == DT_DIR
1461 (de->d_name[0] == '.' && de->d_name[1] == '\0')
1462 || (de->d_name[0] == '.' && de->d_name[1] == '.' && de->d_name[2] == '\0')
1466 else if(de->d_type == DT_DIR) {
1467 char *s = mallocz(pathlen + strlen(de->d_name) + 2);
1470 strcat(s, de->d_name);
1476 else if((de->d_type == DT_LNK || de->d_type == DT_REG) &&
1477 len > 5 && !strcmp(&de->d_name[len - 5], ".conf")) {
1478 health_readfile(path, de->d_name);
1485 static inline char *health_config_dir(void) {
1486 char buffer[FILENAME_MAX + 1];
1487 snprintfz(buffer, FILENAME_MAX, "%s/health.d", config_get("global", "config directory", CONFIG_DIR));
1488 return config_get("health", "health configuration directory", buffer);
1491 void health_init(void) {
1492 debug(D_HEALTH, "Health configuration initializing");
1494 if(!(health_enabled = config_get_boolean("health", "enabled", 1))) {
1495 debug(D_HEALTH, "Health is disabled.");
1499 char *path = health_config_dir();
1502 char buffer[FILENAME_MAX + 1];
1503 snprintfz(buffer, FILENAME_MAX, "%s/alarm-email.sh", config_get("global", "plugins directory", PLUGINS_DIR));
1504 health_default_exec = config_get("health", "script to execute on alarm", buffer);
1507 long n = config_get_number("health", "in memory max health log entries", (long)localhost.health_log.max);
1509 error("Health configuration has invalid max log entries %ld. Using default %u", n, localhost.health_log.max);
1510 config_set_number("health", "in memory max health log entries", (long)localhost.health_log.max);
1512 else localhost.health_log.max = (unsigned int)n;
1514 rrdhost_rwlock(&localhost);
1515 health_readdir(path);
1516 rrdhost_unlock(&localhost);
1519 // ----------------------------------------------------------------------------
1522 static inline void health_string2json(BUFFER *wb, const char *prefix, const char *label, const char *value, const char *suffix) {
1524 buffer_sprintf(wb, "%s\"%s\":\"%s\"%s", prefix, label, value, suffix);
1526 buffer_sprintf(wb, "%s\"%s\":null%s", prefix, label, suffix);
1529 static inline void health_alarm_entry2json_nolock(BUFFER *wb, ALARM_ENTRY *ae) {
1530 buffer_sprintf(wb, "\n\t{\n"
1531 "\t\t\"unique_id\":%u,\n"
1532 "\t\t\"alarm_id\":%u,\n"
1533 "\t\t\"alarm_event_id\":%u,\n"
1534 "\t\t\"name\":\"%s\",\n"
1535 "\t\t\"chart\":\"%s\",\n"
1536 "\t\t\"family\":\"%s\",\n"
1537 "\t\t\"processed\":%s,\n"
1538 "\t\t\"updated\":%s,\n"
1539 "\t\t\"exec_run\":%s,\n"
1540 "\t\t\"exec_failed\":%s,\n"
1541 "\t\t\"exec\":\"%s\",\n"
1542 "\t\t\"recipient\":\"%s\",\n"
1543 "\t\t\"exec_code\":%d,\n"
1544 "\t\t\"source\":\"%s\",\n"
1545 "\t\t\"units\":\"%s\",\n"
1546 "\t\t\"info\":\"%s\",\n"
1547 "\t\t\"when\":%lu,\n"
1548 "\t\t\"duration\":%lu,\n"
1549 "\t\t\"non_clear_duration\":%lu,\n"
1550 "\t\t\"status\":\"%s\",\n"
1551 "\t\t\"old_status\":\"%s\",\n",
1558 (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_PROCESSED)?"true":"false",
1559 (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_UPDATED)?"true":"false",
1560 (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_EXEC_RUN)?"true":"false",
1561 (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_EXEC_FAILED)?"true":"false",
1562 ae->exec?ae->exec:health_default_exec,
1563 ae->recipient?ae->recipient:health_default_recipient,
1566 ae->units?ae->units:"",
1567 ae->info?ae->info:"",
1568 (unsigned long)ae->when,
1569 (unsigned long)ae->duration,
1570 (unsigned long)ae->non_clear_duration,
1571 rrdcalc_status2string(ae->new_status),
1572 rrdcalc_status2string(ae->old_status)
1575 buffer_strcat(wb, "\t\t\"value\":");
1576 buffer_rrd_value(wb, ae->new_value);
1577 buffer_strcat(wb, ",\n");
1579 buffer_strcat(wb, "\t\t\"old_value\":");
1580 buffer_rrd_value(wb, ae->old_value);
1581 buffer_strcat(wb, "\n");
1583 buffer_strcat(wb, "\t}");
1586 void health_alarm_log2json(RRDHOST *host, BUFFER *wb, uint32_t after) {
1587 pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
1589 buffer_strcat(wb, "[");
1591 unsigned int max = host->health_log.max;
1592 unsigned int count = 0;
1594 for(ae = host->health_log.alarms; ae && count < max ; count++, ae = ae->next) {
1595 if(ae->unique_id > after) {
1596 if(likely(count)) buffer_strcat(wb, ",");
1597 health_alarm_entry2json_nolock(wb, ae);
1601 buffer_strcat(wb, "\n]\n");
1603 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1606 static inline void health_rrdcalc2json_nolock(BUFFER *wb, RRDCALC *rc) {
1608 "\t\t\"%s.%s\": {\n"
1609 "\t\t\t\"name\": \"%s\",\n"
1610 "\t\t\t\"chart\": \"%s\",\n"
1611 "\t\t\t\"family\": \"%s\",\n"
1612 "\t\t\t\"active\": %s,\n"
1613 "\t\t\t\"exec\": \"%s\",\n"
1614 "\t\t\t\"recipient\": \"%s\",\n"
1615 "\t\t\t\"source\": \"%s\",\n"
1616 "\t\t\t\"units\": \"%s\",\n"
1617 "\t\t\t\"info\": \"%s\",\n"
1618 "\t\t\t\"status\": \"%s\",\n"
1619 "\t\t\t\"last_status_change\": %lu,\n"
1620 "\t\t\t\"last_updated\": %lu,\n"
1621 "\t\t\t\"next_update\": %lu,\n"
1622 "\t\t\t\"update_every\": %d,\n"
1623 , rc->chart, rc->name
1626 , (rc->rrdset && rc->rrdset->family)?rc->rrdset->family:""
1627 , (rc->rrdset)?"true":"false"
1628 , rc->exec?rc->exec:health_default_exec
1629 , rc->recipient?rc->recipient:health_default_recipient
1631 , rc->units?rc->units:""
1632 , rc->info?rc->info:""
1633 , rrdcalc_status2string(rc->status)
1634 , (unsigned long)rc->last_status_change
1635 , (unsigned long)rc->last_updated
1636 , (unsigned long)rc->next_update
1640 if(RRDCALC_HAS_DB_LOOKUP(rc)) {
1641 if(rc->dimensions && *rc->dimensions)
1642 health_string2json(wb, "\t\t\t", "lookup_dimensions", rc->dimensions, ",\n");
1645 "\t\t\t\"db_after\": %lu,\n"
1646 "\t\t\t\"db_before\": %lu,\n"
1647 "\t\t\t\"lookup_method\": \"%s\",\n"
1648 "\t\t\t\"lookup_after\": %d,\n"
1649 "\t\t\t\"lookup_before\": %d,\n"
1650 "\t\t\t\"lookup_options\": \"",
1651 (unsigned long) rc->db_after,
1652 (unsigned long) rc->db_before,
1653 group_method2string(rc->group),
1657 buffer_data_options2string(wb, rc->options);
1658 buffer_strcat(wb, "\",\n");
1661 if(rc->calculation) {
1662 health_string2json(wb, "\t\t\t", "calc", rc->calculation->source, ",\n");
1663 health_string2json(wb, "\t\t\t", "calc_parsed", rc->calculation->parsed_as, ",\n");
1667 health_string2json(wb, "\t\t\t", "warn", rc->warning->source, ",\n");
1668 health_string2json(wb, "\t\t\t", "warn_parsed", rc->warning->parsed_as, ",\n");
1672 health_string2json(wb, "\t\t\t", "crit", rc->critical->source, ",\n");
1673 health_string2json(wb, "\t\t\t", "crit_parsed", rc->critical->parsed_as, ",\n");
1676 buffer_strcat(wb, "\t\t\t\"green\":");
1677 buffer_rrd_value(wb, rc->green);
1678 buffer_strcat(wb, ",\n");
1680 buffer_strcat(wb, "\t\t\t\"red\":");
1681 buffer_rrd_value(wb, rc->red);
1682 buffer_strcat(wb, ",\n");
1684 buffer_strcat(wb, "\t\t\t\"value\":");
1685 buffer_rrd_value(wb, rc->value);
1686 buffer_strcat(wb, "\n");
1688 buffer_strcat(wb, "\t\t}");
1691 //void health_rrdcalctemplate2json_nolock(BUFFER *wb, RRDCALCTEMPLATE *rt) {
1695 void health_alarms2json(RRDHOST *host, BUFFER *wb, int all) {
1697 rrdhost_rdlock(&localhost);
1699 buffer_strcat(wb, "{\n\t\"alarms\": {\n");
1701 for(i = 0, rc = host->alarms; rc ; rc = rc->next) {
1702 if(unlikely(!rc->rrdset || !rc->rrdset->last_collected_time.tv_sec))
1705 if(likely(!all && !(rc->status == RRDCALC_STATUS_WARNING || rc->status == RRDCALC_STATUS_CRITICAL)))
1708 if(likely(i)) buffer_strcat(wb, ",\n");
1709 health_rrdcalc2json_nolock(wb, rc);
1713 // buffer_strcat(wb, "\n\t},\n\t\"templates\": {");
1715 // RRDCALCTEMPLATE *rt;
1716 // for(rt = host->templates; rt ; rt = rt->next)
1717 // health_rrdcalctemplate2json_nolock(wb, rt);
1719 buffer_sprintf(wb, "\n\t},\n\t\"now\": %lu\n}\n", (unsigned long)time(NULL));
1720 rrdhost_unlock(&localhost);
1724 // ----------------------------------------------------------------------------
1725 // re-load health configuration
1727 static inline void health_free_all_nolock(RRDHOST *host) {
1728 while(host->templates)
1729 rrdcalctemplate_free(host, host->templates);
1732 rrdcalc_free(host, host->alarms);
1735 void health_reload(void) {
1736 if(!health_enabled) {
1737 error("Health reload is requested, but health is not enabled.");
1741 char *path = health_config_dir();
1743 rrdhost_rwlock(&localhost);
1744 health_free_all_nolock(&localhost);
1745 rrdhost_unlock(&localhost);
1748 for(st = localhost.rrdset_root; st ; st = st->next) {
1753 rrdhost_rwlock(&localhost);
1754 health_readdir(path);
1755 rrdhost_unlock(&localhost);
1757 for(st = localhost.rrdset_root; st ; st = st->next) {
1758 rrdhost_rwlock(&localhost);
1760 rrdsetcalc_link_matching(st);
1761 rrdcalctemplate_link_matching(st);
1763 rrdhost_unlock(&localhost);
1768 // ----------------------------------------------------------------------------
1769 // health main thread and friends
1771 static inline int rrdcalc_value2status(calculated_number n) {
1772 if(isnan(n)) return RRDCALC_STATUS_UNDEFINED;
1773 if(n) return RRDCALC_STATUS_RAISED;
1774 return RRDCALC_STATUS_CLEAR;
1777 static inline void health_alarm_execute(RRDHOST *host, ALARM_ENTRY *ae) {
1778 if(ae->old_status == RRDCALC_STATUS_UNINITIALIZED && ae->new_status == RRDCALC_STATUS_CLEAR)
1781 char buffer[FILENAME_MAX + 1];
1784 const char *exec = ae->exec;
1785 if(!exec) exec = health_default_exec;
1787 const char *recipient = ae->recipient;
1788 if(!recipient) recipient = health_default_recipient;
1790 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'",
1799 ae->chart?ae->chart:"NOCAHRT",
1800 ae->family?ae->family:"NOFAMILY",
1801 rrdcalc_status2string(ae->new_status),
1802 rrdcalc_status2string(ae->old_status),
1805 ae->source?ae->source:"UNKNOWN",
1806 (uint32_t)ae->duration,
1807 (uint32_t)ae->non_clear_duration,
1808 ae->units?ae->units:"",
1809 ae->info?ae->info:""
1812 ae->notifications |= HEALTH_ENTRY_NOTIFICATIONS_EXEC_RUN;
1814 debug(D_HEALTH, "executing command '%s'", buffer);
1815 FILE *fp = mypopen(buffer, &command_pid);
1817 error("HEALTH: Cannot popen(\"%s\", \"r\").", buffer);
1820 debug(D_HEALTH, "HEALTH reading from command");
1821 char *s = fgets(buffer, FILENAME_MAX, fp);
1823 debug(D_HEALTH, "HEALTH closing command");
1824 ae->exec_code = mypclose(fp, command_pid);
1825 debug(D_HEALTH, "done executing command - returned with code %d", ae->exec_code);
1827 if(ae->exec_code != 0)
1828 ae->notifications |= HEALTH_ENTRY_NOTIFICATIONS_EXEC_FAILED;
1831 static inline void health_process_notifications(RRDHOST *host, ALARM_ENTRY *ae) {
1832 info("Health alarm '%s.%s' = %0.2Lf - changed status from %s to %s",
1833 ae->chart?ae->chart:"NOCHART", ae->name,
1835 rrdcalc_status2string(ae->old_status),
1836 rrdcalc_status2string(ae->new_status)
1839 health_alarm_execute(host, ae);
1842 static inline void health_alarm_log(RRDHOST *host,
1843 uint32_t alarm_id, uint32_t alarm_event_id,
1845 const char *name, const char *chart, const char *family,
1846 const char *exec, const char *recipient, time_t duration,
1847 calculated_number old_value, calculated_number new_value,
1848 int old_status, int new_status,
1853 ALARM_ENTRY *ae = callocz(1, sizeof(ALARM_ENTRY));
1854 ae->name = strdupz(name);
1855 ae->hash_name = simple_hash(ae->name);
1858 ae->chart = strdupz(chart);
1859 ae->hash_chart = simple_hash(ae->chart);
1863 ae->family = strdupz(family);
1865 if(exec) ae->exec = strdupz(exec);
1866 if(recipient) ae->recipient = strdupz(recipient);
1867 if(source) ae->source = strdupz(source);
1868 if(units) ae->units = strdupz(units);
1869 if(info) ae->info = strdupz(info);
1871 ae->unique_id = host->health_log.next_log_id++;
1872 ae->alarm_id = alarm_id;
1873 ae->alarm_event_id = alarm_event_id;
1875 ae->old_value = old_value;
1876 ae->new_value = new_value;
1877 ae->old_status = old_status;
1878 ae->new_status = new_status;
1879 ae->duration = duration;
1881 if(ae->old_status == RRDCALC_STATUS_WARNING || ae->old_status == RRDCALC_STATUS_CRITICAL)
1882 ae->non_clear_duration += ae->duration;
1885 pthread_rwlock_wrlock(&host->health_log.alarm_log_rwlock);
1886 ae->next = host->health_log.alarms;
1887 host->health_log.alarms = ae;
1888 host->health_log.count++;
1889 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1891 // match previous alarms
1892 pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
1894 for(t = host->health_log.alarms ; t ; t = t->next) {
1896 t->hash_name == ae->hash_name &&
1897 t->hash_chart == ae->hash_chart &&
1898 !strcmp(t->name, ae->name) &&
1899 t->chart && ae->chart && !strcmp(t->chart, ae->chart)) {
1901 if(!(t->notifications & HEALTH_ENTRY_NOTIFICATIONS_UPDATED) && !t->updated_by) {
1902 t->notifications |= HEALTH_ENTRY_NOTIFICATIONS_UPDATED;
1905 if((t->new_status == RRDCALC_STATUS_WARNING || t->new_status == RRDCALC_STATUS_CRITICAL) &&
1906 (t->old_status == RRDCALC_STATUS_WARNING || t->old_status == RRDCALC_STATUS_CRITICAL))
1907 ae->non_clear_duration += t->non_clear_duration;
1910 // no need to continue
1915 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1918 static inline void health_alarm_log_process(RRDHOST *host) {
1919 static uint32_t last_processed = 0;
1922 pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
1924 for(ae = host->health_log.alarms; ae ;ae = ae->next) {
1925 if(last_processed >= ae->unique_id) break;
1927 if(!(ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_PROCESSED) &&
1928 !(ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_UPDATED)) {
1929 ae->notifications |= HEALTH_ENTRY_NOTIFICATIONS_PROCESSED;
1930 health_process_notifications(host, ae);
1934 if(host->health_log.alarms)
1935 last_processed = host->health_log.alarms->unique_id;
1937 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1939 if(host->health_log.count <= host->health_log.max)
1942 // cleanup excess entries in the log
1943 pthread_rwlock_wrlock(&host->health_log.alarm_log_rwlock);
1945 ALARM_ENTRY *last = NULL;
1946 unsigned int count = host->health_log.max;
1947 for(ae = host->health_log.alarms; ae && count ; count--, last = ae, ae = ae->next) ;
1949 if(ae && last && last->next == ae)
1955 ALARM_ENTRY *t = ae->next;
1961 freez(ae->recipient);
1970 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1973 static inline int rrdcalc_isrunnable(RRDCALC *rc, time_t now, time_t *next_run) {
1974 if (unlikely(!rc->rrdset)) {
1975 debug(D_HEALTH, "Health not running alarm '%s.%s'. It is not linked to a chart.", rc->chart?rc->chart:"NOCHART", rc->name);
1979 if (unlikely(!rc->rrdset->last_collected_time.tv_sec)) {
1980 debug(D_HEALTH, "Health not running alarm '%s.%s'. Chart is not yet collected.", rc->chart?rc->chart:"NOCHART", rc->name);
1984 if (unlikely(!rc->update_every)) {
1985 debug(D_HEALTH, "Health not running alarm '%s.%s'. It does not have an update frequency", rc->chart?rc->chart:"NOCHART", rc->name);
1989 if (unlikely(rc->next_update > now)) {
1990 if (unlikely(*next_run > rc->next_update))
1991 *next_run = rc->next_update;
1993 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));
1998 // we should check that the DB lookup is possible
2000 // - the duration of the chart includes the required timeframe
2001 // we SHOULD NOT check the dimensions - there might be alarms that refer non-existing dimensions (e.g. cpu steal)
2006 void *health_main(void *ptr) {
2009 info("HEALTH thread created with task id %d", gettid());
2011 if(pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, NULL) != 0)
2012 error("Cannot set pthread cancel type to DEFERRED.");
2014 if(pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL) != 0)
2015 error("Cannot set pthread cancel state to ENABLE.");
2017 int min_run_every = (int)config_get_number("health", "run at least every seconds", 10);
2018 if(min_run_every < 1) min_run_every = 1;
2020 BUFFER *wb = buffer_create(100);
2022 unsigned int loop = 0;
2023 while(health_enabled) {
2025 debug(D_HEALTH, "Health monitoring iteration no %u started", loop);
2027 int oldstate, runnable = 0;
2028 time_t now = time(NULL);
2029 time_t next_run = now + min_run_every;
2032 if (unlikely(pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate) != 0))
2033 error("Cannot set pthread cancel state to DISABLE.");
2035 rrdhost_rdlock(&localhost);
2037 // the first loop is to lookup values from the db
2038 for (rc = localhost.alarms; rc; rc = rc->next) {
2039 if (unlikely(!rrdcalc_isrunnable(rc, now, &next_run)))
2043 rc->old_value = rc->value;
2045 // 1. if there is database lookup, do it
2046 // 2. if there is calculation expression, run it
2048 if (unlikely(RRDCALC_HAS_DB_LOOKUP(rc))) {
2049 time_t old_db_timestamp = rc->db_before;
2050 int value_is_null = 0;
2052 int ret = rrd2value(rc->rrdset, wb, &rc->value,
2053 rc->dimensions, 1, rc->after, rc->before, rc->group,
2054 rc->options, &rc->db_after, &rc->db_before, &value_is_null);
2056 if (unlikely(ret != 200)) {
2057 // database lookup failed
2060 debug(D_HEALTH, "Health alarm '%s.%s': database lookup returned error %d", rc->chart?rc->chart:"NOCHART", rc->name, ret);
2062 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_ERROR))) {
2063 rc->rrdcalc_flags |= RRDCALC_FLAG_DB_ERROR;
2064 error("Health alarm '%s.%s': database lookup returned error %d", rc->chart?rc->chart:"NOCHART", rc->name, ret);
2067 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_ERROR))
2068 rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_ERROR;
2070 if (unlikely(old_db_timestamp == rc->db_before)) {
2071 // database is stale
2073 debug(D_HEALTH, "Health alarm '%s.%s': database is stale", rc->chart?rc->chart:"NOCHART", rc->name);
2075 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_STALE))) {
2076 rc->rrdcalc_flags |= RRDCALC_FLAG_DB_STALE;
2077 error("Health alarm '%s.%s': database is stale", rc->chart?rc->chart:"NOCHART", rc->name);
2080 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_STALE))
2081 rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_STALE;
2083 if (unlikely(value_is_null)) {
2084 // collected value is null
2088 debug(D_HEALTH, "Health alarm '%s.%s': database lookup returned empty value (possibly value is not collected yet)",
2089 rc->chart?rc->chart:"NOCHART", rc->name);
2091 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_NAN))) {
2092 rc->rrdcalc_flags |= RRDCALC_FLAG_DB_NAN;
2093 error("Health alarm '%s.%s': database lookup returned empty value (possibly value is not collected yet)",
2094 rc->chart?rc->chart:"NOCHART", rc->name);
2097 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_NAN))
2098 rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_NAN;
2100 debug(D_HEALTH, "Health alarm '%s.%s': database lookup gave value "
2101 CALCULATED_NUMBER_FORMAT, rc->chart?rc->chart:"NOCHART", rc->name, rc->value);
2104 if(unlikely(rc->calculation)) {
2105 if (unlikely(!expression_evaluate(rc->calculation))) {
2106 // calculation failed
2110 debug(D_HEALTH, "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));
2113 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_CALC_ERROR))) {
2114 rc->rrdcalc_flags |= RRDCALC_FLAG_CALC_ERROR;
2115 error("Health alarm '%s.%s': failed to evaluate calculation with error: %s",
2116 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->calculation->error_msg));
2120 if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_CALC_ERROR))
2121 rc->rrdcalc_flags &= ~RRDCALC_FLAG_CALC_ERROR;
2123 debug(D_HEALTH, "Health alarm '%s.%s': calculation expression gave value "
2124 CALCULATED_NUMBER_FORMAT
2125 ": %s (source: %s)",
2126 rc->chart?rc->chart:"NOCHART", rc->name,
2127 rc->calculation->result,
2128 buffer_tostring(rc->calculation->error_msg),
2132 rc->value = rc->calculation->result;
2136 rrdhost_unlock(&localhost);
2139 rrdhost_rdlock(&localhost);
2141 for (rc = localhost.alarms; rc; rc = rc->next) {
2142 if (unlikely(!rrdcalc_isrunnable(rc, now, &next_run)))
2145 int warning_status = RRDCALC_STATUS_UNDEFINED;
2146 int critical_status = RRDCALC_STATUS_UNDEFINED;
2148 if(likely(rc->warning)) {
2149 if(unlikely(!expression_evaluate(rc->warning))) {
2150 // calculation failed
2152 debug(D_HEALTH, "Health alarm '%s.%s': warning expression failed with error: %s",
2153 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->warning->error_msg));
2155 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_WARN_ERROR))) {
2156 rc->rrdcalc_flags |= RRDCALC_FLAG_WARN_ERROR;
2157 error("Health alarm '%s.%s': warning expression failed with error: %s",
2158 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->warning->error_msg));
2162 if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_WARN_ERROR))
2163 rc->rrdcalc_flags &= ~RRDCALC_FLAG_WARN_ERROR;
2165 debug(D_HEALTH, "Health alarm '%s.%s': warning expression gave value "
2166 CALCULATED_NUMBER_FORMAT
2167 ": %s (source: %s)",
2168 rc->chart?rc->chart:"NOCHART", rc->name,
2169 rc->warning->result,
2170 buffer_tostring(rc->warning->error_msg),
2174 warning_status = rrdcalc_value2status(rc->warning->result);
2178 if(likely(rc->critical)) {
2179 if(unlikely(!expression_evaluate(rc->critical))) {
2180 // calculation failed
2182 debug(D_HEALTH, "Health alarm '%s.%s': critical expression failed with error: %s",
2183 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->critical->error_msg));
2185 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_CRIT_ERROR))) {
2186 rc->rrdcalc_flags |= RRDCALC_FLAG_CRIT_ERROR;
2187 error("Health alarm '%s.%s': critical expression failed with error: %s",
2188 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->critical->error_msg));
2192 if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_CRIT_ERROR))
2193 rc->rrdcalc_flags &= ~RRDCALC_FLAG_CRIT_ERROR;
2195 debug(D_HEALTH, "Health alarm '%s.%s': critical expression gave value "
2196 CALCULATED_NUMBER_FORMAT
2197 ": %s (source: %s)",
2198 rc->chart?rc->chart:"NOCHART", rc->name,
2199 rc->critical->result,
2200 buffer_tostring(rc->critical->error_msg),
2204 critical_status = rrdcalc_value2status(rc->critical->result);
2208 int status = RRDCALC_STATUS_UNDEFINED;
2210 switch(warning_status) {
2211 case RRDCALC_STATUS_CLEAR:
2212 status = RRDCALC_STATUS_CLEAR;
2215 case RRDCALC_STATUS_RAISED:
2216 status = RRDCALC_STATUS_WARNING;
2223 switch(critical_status) {
2224 case RRDCALC_STATUS_CLEAR:
2225 if(status == RRDCALC_STATUS_UNDEFINED)
2226 status = RRDCALC_STATUS_CLEAR;
2229 case RRDCALC_STATUS_RAISED:
2230 status = RRDCALC_STATUS_CRITICAL;
2237 if(status != rc->status) {
2238 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);
2239 rc->last_status_change = now;
2240 rc->status = status;
2243 rc->last_updated = now;
2244 rc->next_update = now + rc->update_every;
2246 if (next_run > rc->next_update)
2247 next_run = rc->next_update;
2250 rrdhost_unlock(&localhost);
2253 if (unlikely(pthread_setcancelstate(oldstate, NULL) != 0))
2254 error("Cannot set pthread cancel state to RESTORE (%d).", oldstate);
2256 // execute notifications
2258 health_alarm_log_process(&localhost);
2261 if(now < next_run) {
2262 debug(D_HEALTH, "Health monitoring iteration no %u done. Next iteration in %d secs",
2263 loop, (int) (next_run - now));
2264 sleep_usec(1000000 * (unsigned long long) (next_run - now));
2267 debug(D_HEALTH, "Health monitoring iteration no %u done. Next iteration now", loop);
2273 info("HEALTH thread exiting");