3 #define RRDVAR_MAX_LENGTH 1024
5 static const char *health_default_exec = PLUGINS_DIR "/alarm-email.sh";
6 int health_enabled = 1;
8 // ----------------------------------------------------------------------------
11 static inline int rrdvar_fix_name(char *variable) {
14 if (!isalnum(*variable) && *variable != '.' && *variable != '_') {
25 int rrdvar_compare(void* a, void* b) {
26 if(((RRDVAR *)a)->hash < ((RRDVAR *)b)->hash) return -1;
27 else if(((RRDVAR *)a)->hash > ((RRDVAR *)b)->hash) return 1;
28 else return strcmp(((RRDVAR *)a)->name, ((RRDVAR *)b)->name);
31 static inline RRDVAR *rrdvar_index_add(avl_tree_lock *tree, RRDVAR *rv) {
32 RRDVAR *ret = (RRDVAR *)avl_insert_lock(tree, (avl *)(rv));
34 debug(D_VARIABLES, "Request to insert RRDVAR '%s' into index failed. Already exists.", rv->name);
39 static inline RRDVAR *rrdvar_index_del(avl_tree_lock *tree, RRDVAR *rv) {
40 RRDVAR *ret = (RRDVAR *)avl_remove_lock(tree, (avl *)(rv));
42 error("Request to remove RRDVAR '%s' from index failed. Not Found.", rv->name);
47 static inline RRDVAR *rrdvar_index_find(avl_tree_lock *tree, const char *name, uint32_t hash) {
49 tmp.name = (char *)name;
50 tmp.hash = (hash)?hash:simple_hash(tmp.name);
52 return (RRDVAR *)avl_search_lock(tree, (avl *)&tmp);
55 static inline void rrdvar_free(RRDHOST *host, avl_tree_lock *tree, RRDVAR *rv) {
61 rrdvar_index_del(tree, rv);
67 static inline RRDVAR *rrdvar_create_and_index(const char *scope, avl_tree_lock *tree, const char *name, int type, calculated_number *value) {
68 char *variable = strdupz(name);
69 rrdvar_fix_name(variable);
70 uint32_t hash = simple_hash(variable);
72 RRDVAR *rv = rrdvar_index_find(tree, variable, hash);
74 debug(D_VARIABLES, "Variable '%s' not found in scope '%s'. Creating a new one.", variable, scope);
76 rv = callocz(1, sizeof(RRDVAR));
82 RRDVAR *ret = rrdvar_index_add(tree, rv);
83 if(unlikely(ret != rv)) {
84 debug(D_VARIABLES, "Variable '%s' in scope '%s' already exists", variable, scope);
85 rrdvar_free(NULL, NULL, rv);
89 debug(D_VARIABLES, "Variable '%s' created in scope '%s'", variable, scope);
100 // ----------------------------------------------------------------------------
103 calculated_number rrdvar2number(RRDVAR *rv) {
105 case RRDVAR_TYPE_CALCULATED: {
106 calculated_number *n = (calculated_number *)rv->value;
110 case RRDVAR_TYPE_TIME_T: {
111 time_t *n = (time_t *)rv->value;
115 case RRDVAR_TYPE_COLLECTED: {
116 collected_number *n = (collected_number *)rv->value;
120 case RRDVAR_TYPE_TOTAL: {
121 total_number *n = (total_number *)rv->value;
125 case RRDVAR_TYPE_INT: {
126 int *n = (int *)rv->value;
131 error("I don't know how to convert RRDVAR type %d to calculated_number", rv->type);
136 void dump_variable(void *data) {
137 RRDVAR *rv = (RRDVAR *)data;
138 debug(D_HEALTH, "%50s : %20.5Lf", rv->name, rrdvar2number(rv));
141 int health_variable_lookup(const char *variable, uint32_t hash, RRDCALC *rc, calculated_number *result) {
142 RRDSET *st = rc->rrdset;
147 rv = rrdvar_index_find(&st->variables_root_index, variable, hash);
149 *result = rrdvar2number(rv);
153 rv = rrdvar_index_find(&st->rrdfamily->variables_root_index, variable, hash);
155 *result = rrdvar2number(rv);
159 rv = rrdvar_index_find(&st->rrdhost->variables_root_index, variable, hash);
161 *result = rrdvar2number(rv);
165 debug(D_HEALTH, "Available local chart '%s' variables:", st->id);
166 avl_traverse_lock(&st->variables_root_index, dump_variable);
168 debug(D_HEALTH, "Available family '%s' variables:", st->rrdfamily->family);
169 avl_traverse_lock(&st->rrdfamily->variables_root_index, dump_variable);
171 debug(D_HEALTH, "Available host '%s' variables:", st->rrdhost->hostname);
172 avl_traverse_lock(&st->rrdhost->variables_root_index, dump_variable);
177 // ----------------------------------------------------------------------------
178 // RRDSETVAR management
180 RRDSETVAR *rrdsetvar_create(RRDSET *st, const char *variable, int type, void *value, uint32_t options) {
181 debug(D_VARIABLES, "RRDVARSET create for chart id '%s' name '%s' with variable name '%s'", st->id, st->name, variable);
182 RRDSETVAR *rs = (RRDSETVAR *)callocz(1, sizeof(RRDSETVAR));
184 char buffer[RRDVAR_MAX_LENGTH + 1];
185 snprintfz(buffer, RRDVAR_MAX_LENGTH, "%s.%s", st->id, variable);
186 rs->fullid = strdupz(buffer);
188 snprintfz(buffer, RRDVAR_MAX_LENGTH, "%s.%s", st->name, variable);
189 rs->fullname = strdupz(buffer);
191 rs->variable = strdupz(variable);
195 rs->options = options;
198 rs->local = rrdvar_create_and_index("local", &st->variables_root_index, rs->variable, rs->type, rs->value);
199 rs->family = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->fullid, rs->type, rs->value);
200 rs->host = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullid, rs->type, rs->value);
201 rs->family_name = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->fullname, rs->type, rs->value);
202 rs->host_name = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullname, rs->type, rs->value);
204 rs->next = st->variables;
210 void rrdsetvar_rename_all(RRDSET *st) {
211 debug(D_VARIABLES, "RRDSETVAR rename for chart id '%s' name '%s'", st->id, st->name);
213 // only these 2 can change name
217 char buffer[RRDVAR_MAX_LENGTH + 1];
218 RRDSETVAR *rs, *next = st->variables;
222 snprintfz(buffer, RRDVAR_MAX_LENGTH, "%s.%s", st->name, rs->variable);
224 if (strcmp(buffer, rs->fullname)) {
226 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family_name);
227 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_name);
230 rs->fullname = strdupz(st->name);
231 rs->family_name = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->fullname, rs->type, rs->value);
232 rs->host_name = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullname, rs->type, rs->value);
236 rrdsetcalc_link_matching(st);
239 void rrdsetvar_free(RRDSETVAR *rs) {
240 RRDSET *st = rs->rrdset;
241 debug(D_VARIABLES, "RRDSETVAR free for chart id '%s' name '%s', variable '%s'", st->id, st->name, rs->variable);
243 rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local);
244 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family);
245 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host);
246 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family_name);
247 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_name);
249 if(st->variables == rs) {
250 st->variables = rs->next;
254 for (t = st->variables; t && t->next != rs; t = t->next);
255 if(!t) error("RRDSETVAR '%s' not found in chart '%s' variables linked list", rs->fullname, st->id);
256 else t->next = rs->next;
265 // ----------------------------------------------------------------------------
266 // RRDDIMVAR management
268 #define RRDDIMVAR_ID_MAX 1024
270 RRDDIMVAR *rrddimvar_create(RRDDIM *rd, int type, const char *prefix, const char *suffix, void *value, uint32_t options) {
271 RRDSET *st = rd->rrdset;
273 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:"");
275 if(!prefix) prefix = "";
276 if(!suffix) suffix = "";
278 char buffer[RRDDIMVAR_ID_MAX + 1];
279 RRDDIMVAR *rs = (RRDDIMVAR *)callocz(1, sizeof(RRDDIMVAR));
281 rs->prefix = strdupz(prefix);
282 rs->suffix = strdupz(suffix);
284 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s%s%s", rs->prefix, rd->id, rs->suffix);
285 rs->id = strdupz(buffer);
287 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s%s%s", rs->prefix, rd->name, rs->suffix);
288 rs->name = strdupz(buffer);
290 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->id, rs->id);
291 rs->fullidid = strdupz(buffer);
293 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->id, rs->name);
294 rs->fullidname = strdupz(buffer);
296 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->name, rs->id);
297 rs->fullnameid = strdupz(buffer);
299 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->name, rs->name);
300 rs->fullnamename = strdupz(buffer);
304 rs->options = options;
307 rs->local_id = rrdvar_create_and_index("local", &st->variables_root_index, rs->id, rs->type, rs->value);
308 rs->local_name = rrdvar_create_and_index("local", &st->variables_root_index, rs->name, rs->type, rs->value);
310 rs->family_id = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->id, rs->type, rs->value);
311 rs->family_name = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->name, rs->type, rs->value);
313 rs->host_fullidid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullidid, rs->type, rs->value);
314 rs->host_fullidname = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullidname, rs->type, rs->value);
315 rs->host_fullnameid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullnameid, rs->type, rs->value);
316 rs->host_fullnamename = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullnamename, rs->type, rs->value);
318 rs->next = rd->variables;
324 void rrddimvar_rename_all(RRDDIM *rd) {
325 RRDSET *st = rd->rrdset;
326 debug(D_VARIABLES, "RRDDIMSET rename for chart id '%s' name '%s', dimension id '%s', name '%s'", st->id, st->name, rd->id, rd->name);
328 RRDDIMVAR *rs, *next = rd->variables;
332 if (strcmp(rd->name, rs->name)) {
333 char buffer[RRDDIMVAR_ID_MAX + 1];
337 rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local_name);
339 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s%s%s", rs->prefix, rd->name, rs->suffix);
340 rs->name = strdupz(buffer);
341 rs->local_name = rrdvar_create_and_index("local", &st->variables_root_index, rs->name, rs->type, rs->value);
343 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullidname);
344 freez(rs->fullidname);
345 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->id, rs->name);
346 rs->fullidname = strdupz(buffer);
347 rs->host_fullidname = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index,
348 rs->fullidname, rs->type, rs->value);
351 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnameid);
352 freez(rs->fullnameid);
353 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->name, rs->id);
354 rs->fullnameid = strdupz(buffer);
355 rs->host_fullnameid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index,
356 rs->fullnameid, rs->type, rs->value);
359 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnamename);
360 freez(rs->fullnamename);
361 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->name, rs->name);
362 rs->fullnamename = strdupz(buffer);
363 rs->host_fullnamename = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index,
364 rs->fullnamename, rs->type, rs->value);
369 void rrddimvar_free(RRDDIMVAR *rs) {
370 RRDDIM *rd = rs->rrddim;
371 RRDSET *st = rd->rrdset;
372 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);
374 rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local_id);
375 rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local_name);
377 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family_id);
378 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family_name);
380 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullidid);
381 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullidname);
382 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnameid);
383 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnamename);
385 if(rd->variables == rs) {
386 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);
387 rd->variables = rs->next;
390 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);
392 for (t = rd->variables; t && t->next != rs; t = t->next) ;
393 if(!t) error("RRDDIMVAR '%s' not found in dimension '%s/%s' variables linked list", rs->name, st->id, rd->id);
394 else t->next = rs->next;
402 freez(rs->fullidname);
403 freez(rs->fullnameid);
404 freez(rs->fullnamename);
408 // ----------------------------------------------------------------------------
409 // RRDCALC management
411 static inline const char *rrdcalc_status2string(int status) {
413 case RRDCALC_STATUS_UNINITIALIZED:
414 return "UNINITIALIZED";
416 case RRDCALC_STATUS_UNDEFINED:
419 case RRDCALC_STATUS_CLEAR:
422 case RRDCALC_STATUS_RAISED:
425 case RRDCALC_STATUS_WARNING:
428 case RRDCALC_STATUS_CRITICAL:
436 static void rrdsetcalc_link(RRDSET *st, RRDCALC *rc) {
437 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);
439 rc->last_status_change = time(NULL);
442 rc->rrdset_next = st->alarms;
443 rc->rrdset_prev = NULL;
446 if(rc->update_every < rc->rrdset->update_every) {
447 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);
448 rc->update_every = rc->rrdset->update_every;
451 if(!isnan(rc->green) && isnan(st->green)) {
452 debug(D_HEALTH, "Health alarm '%s.%s' green threshold set from %Lf to %Lf.", rc->rrdset->id, rc->name, rc->rrdset->green, rc->green);
453 st->green = rc->green;
456 if(!isnan(rc->red) && isnan(st->red)) {
457 debug(D_HEALTH, "Health alarm '%s.%s' red threshold set from %Lf to %Lf.", rc->rrdset->id, rc->name, rc->rrdset->red, rc->red);
461 rc->local = rrdvar_create_and_index("local", &st->variables_root_index, rc->name, RRDVAR_TYPE_CALCULATED, &rc->value);
462 rc->family = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rc->name, RRDVAR_TYPE_CALCULATED, &rc->value);
464 char fullname[RRDVAR_MAX_LENGTH + 1];
465 snprintfz(fullname, RRDVAR_MAX_LENGTH, "%s.%s", st->id, rc->name);
466 rc->hostid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, fullname, RRDVAR_TYPE_CALCULATED, &rc->value);
468 snprintfz(fullname, RRDVAR_MAX_LENGTH, "%s.%s", st->name, rc->name);
469 rc->hostname = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, fullname, RRDVAR_TYPE_CALCULATED, &rc->value);
471 if(!rc->units) rc->units = strdupz(st->units);
474 static inline int rrdcalc_is_matching_this_rrdset(RRDCALC *rc, RRDSET *st) {
475 if( (rc->hash_chart == st->hash && !strcmp(rc->chart, st->id)) ||
476 (rc->hash_chart == st->hash_name && !strcmp(rc->chart, st->name)))
482 // this has to be called while the RRDHOST is locked
483 inline void rrdsetcalc_link_matching(RRDSET *st) {
484 // debug(D_HEALTH, "find matching alarms for chart '%s'", st->id);
487 for(rc = st->rrdhost->alarms; rc ; rc = rc->next) {
488 if(rc->rrdset) continue;
490 if(rrdcalc_is_matching_this_rrdset(rc, st))
491 rrdsetcalc_link(st, rc);
495 // this has to be called while the RRDHOST is locked
496 inline void rrdsetcalc_unlink(RRDCALC *rc) {
497 RRDSET *st = rc->rrdset;
500 error("Requested to unlink RRDCALC '%s.%s' which is not linked to any RRDSET", rc->chart?rc->chart:"NOCHART", rc->name);
504 RRDHOST *host = st->rrdhost;
506 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);
510 rc->rrdset_prev->rrdset_next = rc->rrdset_next;
513 rc->rrdset_next->rrdset_prev = rc->rrdset_prev;
516 st->alarms = rc->rrdset_next;
518 rc->rrdset_prev = rc->rrdset_next = NULL;
520 rrdvar_free(st->rrdhost, &st->variables_root_index, rc->local);
523 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rc->family);
526 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rc->hostid);
529 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rc->hostname);
534 // RRDCALC will remain in RRDHOST
535 // so that if the matching chart is found in the future
536 // it will be applied automatically
539 RRDCALC *rrdcalc_find(RRDSET *st, const char *name) {
541 uint32_t hash = simple_hash(name);
543 for( rc = st->alarms; rc ; rc = rc->rrdset_next ) {
544 if(rc->hash == hash && !strcmp(rc->name, name))
551 static inline int rrdcalc_exists(RRDHOST *host, const char *name, uint32_t hash) {
554 // make sure it does not already exist
555 for(rc = host->alarms; rc ; rc = rc->next) {
556 if (rc->hash == hash && !strcmp(name, rc->name)) {
557 error("Health alarm '%s' already exists in host '%s'.", name, host->hostname);
565 static inline void rrdcalc_create_part2(RRDHOST *host, RRDCALC *rc) {
566 rrdhost_check_rdlock(host);
568 if(rc->calculation) {
569 rc->calculation->this = &rc->value;
570 rc->calculation->after = &rc->db_after;
571 rc->calculation->before = &rc->db_before;
572 rc->calculation->rrdcalc = rc;
576 rc->warning->this = &rc->value;
577 rc->warning->after = &rc->db_after;
578 rc->warning->before = &rc->db_before;
579 rc->warning->rrdcalc = rc;
583 rc->critical->this = &rc->value;
584 rc->critical->after = &rc->db_after;
585 rc->critical->before = &rc->db_before;
586 rc->critical->rrdcalc = rc;
589 // link it to the host
590 rc->next = host->alarms;
593 // link it to its chart
595 for(st = host->rrdset_root; st ; st = st->next) {
596 if(rrdcalc_is_matching_this_rrdset(rc, st)) {
597 rrdsetcalc_link(st, rc);
603 static inline uint32_t rrdcalc_fullname(char *fullname, size_t len, const char *chart, const char *name) {
604 snprintfz(fullname, len - 1, "%s%s%s", chart?chart:"", chart?".":"", name);
605 rrdvar_fix_name(fullname);
606 return simple_hash(fullname);
609 static inline RRDCALC *rrdcalc_create(RRDHOST *host, const char *name, const char *chart, const char *dimensions,
610 const char *units, const char *info,
611 int group_method, int after, int before, int update_every, uint32_t options,
612 calculated_number green, calculated_number red,
613 const char *exec, const char *source,
614 const char *calc, const char *warn, const char *crit) {
616 char fullname[RRDVAR_MAX_LENGTH + 1];
617 uint32_t hash = rrdcalc_fullname(fullname, RRDVAR_MAX_LENGTH + 1, chart, name);
619 if(rrdcalc_exists(host, fullname, hash))
622 RRDCALC *rc = callocz(1, sizeof(RRDCALC));
624 rc->name = strdupz(name);
625 rc->hash = simple_hash(rc->name);
627 rc->chart = strdupz(chart);
628 rc->hash_chart = simple_hash(rc->chart);
630 if(dimensions) rc->dimensions = strdupz(dimensions);
637 rc->group = group_method;
640 rc->update_every = update_every;
641 rc->options = options;
643 if(exec) rc->exec = strdupz(exec);
644 if(source) rc->source = strdupz(source);
645 if(units) rc->units = strdupz(units);
646 if(info) rc->info = strdupz(info);
649 rc->calculation = expression_parse(calc, NULL, NULL);
651 error("Health alarm '%s.%s': failed to parse calculation expression '%s'", chart, name, calc);
654 rc->warning = expression_parse(warn, NULL, NULL);
656 error("Health alarm '%s.%s': failed to re-parse warning expression '%s'", chart, name, warn);
659 rc->critical = expression_parse(crit, NULL, NULL);
661 error("Health alarm '%s.%s': failed to re-parse critical expression '%s'", chart, name, crit);
664 debug(D_HEALTH, "Health runtime added alarm '%s.%s': exec '%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",
665 (rc->chart)?rc->chart:"NOCHART",
667 (rc->exec)?rc->exec:"DEFAULT",
674 (rc->dimensions)?rc->dimensions:"NONE",
676 (rc->calculation)?rc->calculation->parsed_as:"NONE",
677 (rc->warning)?rc->warning->parsed_as:"NONE",
678 (rc->critical)?rc->critical->parsed_as:"NONE",
682 rrdcalc_create_part2(host, rc);
686 void rrdcalc_free(RRDHOST *host, RRDCALC *rc) {
689 debug(D_HEALTH, "Health removing alarm '%s.%s' of host '%s'", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);
691 // unlink it from RRDSET
692 if(rc->rrdset) rrdsetcalc_unlink(rc);
694 // unlink it from RRDHOST
695 if(rc == host->alarms)
696 host->alarms = rc->next;
698 else if(host->alarms) {
699 RRDCALC *t, *last = host->alarms;
701 for(t = last->next; t && t != rc; last = t, t = t->next) ;
702 if(last && last->next == rc)
703 last->next = rc->next;
705 error("Cannot unlink alarm '%s.%s' from host '%s': not found", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);
708 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);
710 expression_free(rc->calculation);
711 expression_free(rc->warning);
712 expression_free(rc->critical);
717 freez(rc->dimensions);
725 // ----------------------------------------------------------------------------
726 // RRDCALCTEMPLATE management
728 void rrdcalctemplate_link_matching(RRDSET *st) {
731 for(rt = st->rrdhost->templates; rt ; rt = rt->next) {
732 if(rt->hash_context == st->hash_context && !strcmp(rt->context, st->context)) {
734 RRDCALC *rc = rrdcalc_create(st->rrdhost, rt->name, st->id,
735 rt->dimensions, rt->units, rt->info, rt->group, rt->after, rt->before, rt->update_every, rt->options,
736 rt->green, rt->red, rt->exec, rt->source,
737 (rt->calculation)?rt->calculation->source:NULL,
738 (rt->warning)?rt->warning->source:NULL,
739 (rt->critical)?rt->critical->source:NULL);
742 error("Health tried to create alarm from template '%s', but it failed", rt->name);
744 #ifdef NETDATA_INTERNAL_CHECKS
745 else if(rc->rrdset != st)
746 error("Health alarm '%s.%s' should be linked to chart '%s', but it is not", rc->chart?rc->chart:"NOCHART", rc->name, st->id);
754 static inline void rrdcalctemplate_free(RRDHOST *host, RRDCALCTEMPLATE *rt) {
755 debug(D_HEALTH, "Health removing template '%s' of host '%s'", rt->name, host->hostname);
757 if(host->templates) {
758 if(host->templates == rt) {
759 host->templates = rt->next;
762 RRDCALCTEMPLATE *t, *last = host->templates;
763 for (t = last->next; t && t != rt; last = t, t = t->next ) ;
764 if(last && last->next == rt) {
765 last->next = rt->next;
769 error("Cannot find RRDCALCTEMPLATE '%s' linked in host '%s'", rt->name, host->hostname);
773 expression_free(rt->calculation);
774 expression_free(rt->warning);
775 expression_free(rt->critical);
783 freez(rt->dimensions);
787 // ----------------------------------------------------------------------------
788 // load health configuration
790 #define HEALTH_CONF_MAX_LINE 4096
792 #define HEALTH_ALARM_KEY "alarm"
793 #define HEALTH_TEMPLATE_KEY "template"
794 #define HEALTH_ON_KEY "on"
795 #define HEALTH_LOOKUP_KEY "lookup"
796 #define HEALTH_CALC_KEY "calc"
797 #define HEALTH_EVERY_KEY "every"
798 #define HEALTH_GREEN_KEY "green"
799 #define HEALTH_RED_KEY "red"
800 #define HEALTH_WARN_KEY "warn"
801 #define HEALTH_CRIT_KEY "crit"
802 #define HEALTH_EXEC_KEY "exec"
803 #define HEALTH_UNITS_KEY "units"
804 #define HEALTH_INFO_KEY "info"
806 static inline int rrdcalc_add_alarm_from_config(RRDHOST *host, RRDCALC *rc) {
808 char fullname[RRDVAR_MAX_LENGTH + 1];
809 uint32_t hash = rrdcalc_fullname(fullname, RRDVAR_MAX_LENGTH + 1, rc->chart, rc->name);
811 if (rrdcalc_exists(host, fullname, hash))
816 error("Health configuration for alarm '%s' does not have a chart", rc->name);
820 if(!rc->update_every) {
821 error("Health configuration for alarm '%s.%s' has no frequency (parameter 'every'). Ignoring it.", rc->chart?rc->chart:"NOCHART", rc->name);
825 if(!RRDCALC_HAS_DB_LOOKUP(rc) && !rc->warning && !rc->critical) {
826 error("Health configuration for alarm '%s.%s' is useless (no calculation, no warning and no critical evaluation)", rc->chart?rc->chart:"NOCHART", rc->name);
830 debug(D_HEALTH, "Health configuration adding alarm '%s.%s': exec '%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",
831 rc->chart?rc->chart:"NOCHART",
833 (rc->exec)?rc->exec:"DEFAULT",
840 (rc->dimensions)?rc->dimensions:"NONE",
842 (rc->calculation)?rc->calculation->parsed_as:"NONE",
843 (rc->warning)?rc->warning->parsed_as:"NONE",
844 (rc->critical)?rc->critical->parsed_as:"NONE",
848 rrdcalc_create_part2(host, rc);
852 static inline int rrdcalctemplate_add_template_from_config(RRDHOST *host, RRDCALCTEMPLATE *rt) {
854 error("Health configuration for template '%s' does not have a context", rt->name);
858 if(!rt->update_every) {
859 error("Health configuration for template '%s' has no frequency (parameter 'every'). Ignoring it.", rt->name);
863 if(!RRDCALCTEMPLATE_HAS_CALCULATION(rt) && !rt->warning && !rt->critical) {
864 error("Health configuration for template '%s' is useless (no calculation, no warning and no critical evaluation)", rt->name);
869 for (t = host->templates; t ; t = t->next) {
870 if(t->hash_name == rt->hash_name && !strcmp(t->name, rt->name)) {
871 error("Health configuration template '%s' already exists for host '%s'.", rt->name, host->hostname);
876 debug(D_HEALTH, "Health configuration adding template '%s': context '%s', exec '%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'",
878 (rt->context)?rt->context:"NONE",
879 (rt->exec)?rt->exec:"DEFAULT",
886 (rt->dimensions)?rt->dimensions:"NONE",
888 (rt->calculation)?rt->calculation->parsed_as:"NONE",
889 (rt->warning)?rt->warning->parsed_as:"NONE",
890 (rt->critical)?rt->critical->parsed_as:"NONE",
894 rt->next = host->templates;
895 host->templates = rt;
899 static inline int health_parse_duration(char *string, int *result) {
900 // make sure it is a number
901 if(!*string || !(isdigit(*string) || *string == '+' || *string == '-')) {
907 calculated_number n = strtold(string, &e);
911 *result = (int) (n * 86400 * 365);
914 *result = (int) (n * 86400 * 30);
917 *result = (int) (n * 86400 * 7);
920 *result = (int) (n * 86400);
923 *result = (int) (n * 3600);
926 *result = (int) (n * 60);
941 static inline int health_parse_db_lookup(
942 size_t line, const char *path, const char *file, char *string,
943 int *group_method, int *after, int *before, int *every,
944 uint32_t *options, char **dimensions
946 debug(D_HEALTH, "Health configuration parsing database lookup %zu@%s/%s: %s", line, path, file, string);
948 if(*dimensions) freez(*dimensions);
955 char *s = string, *key;
957 // first is the group method
959 while(*s && !isspace(*s)) s++;
960 while(*s && isspace(*s)) *s++ = '\0';
962 error("Health configuration invalid chart calculation at line %zu of file '%s/%s': expected group method followed by the 'after' time, but got '%s'",
963 line, path, file, key);
967 if((*group_method = web_client_api_request_v1_data_group(key, -1)) == -1) {
968 error("Health configuration at line %zu of file '%s/%s': invalid group method '%s'",
969 line, path, file, key);
973 // then is the 'after' time
975 while(*s && !isspace(*s)) s++;
976 while(*s && isspace(*s)) *s++ = '\0';
978 if(!health_parse_duration(key, after)) {
979 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' after group method",
980 line, path, file, key);
985 *every = abs(*after);
987 // now we may have optional parameters
990 while(*s && !isspace(*s)) s++;
991 while(*s && isspace(*s)) *s++ = '\0';
994 if(!strcasecmp(key, "at")) {
996 while(*s && !isspace(*s)) s++;
997 while(*s && isspace(*s)) *s++ = '\0';
999 if (!health_parse_duration(value, before)) {
1000 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' for '%s' keyword",
1001 line, path, file, value, key);
1004 else if(!strcasecmp(key, HEALTH_EVERY_KEY)) {
1006 while(*s && !isspace(*s)) s++;
1007 while(*s && isspace(*s)) *s++ = '\0';
1009 if (!health_parse_duration(value, every)) {
1010 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' for '%s' keyword",
1011 line, path, file, value, key);
1014 else if(!strcasecmp(key, "absolute") || !strcasecmp(key, "abs") || !strcasecmp(key, "absolute_sum")) {
1015 *options |= RRDR_OPTION_ABSOLUTE;
1017 else if(!strcasecmp(key, "min2max")) {
1018 *options |= RRDR_OPTION_MIN2MAX;
1020 else if(!strcasecmp(key, "null2zero")) {
1021 *options |= RRDR_OPTION_NULL2ZERO;
1023 else if(!strcasecmp(key, "percentage")) {
1024 *options |= RRDR_OPTION_PERCENTAGE;
1026 else if(!strcasecmp(key, "unaligned")) {
1027 *options |= RRDR_OPTION_NOT_ALIGNED;
1029 else if(!strcasecmp(key, "of")) {
1030 if(*s && strcasecmp(s, "all"))
1031 *dimensions = strdupz(s);
1035 error("Health configuration at line %zu of file '%s/%s': unknown keyword '%s'",
1036 line, path, file, key);
1043 static inline char *health_source_file(size_t line, const char *path, const char *filename) {
1044 char buffer[FILENAME_MAX + 1];
1045 snprintfz(buffer, FILENAME_MAX, "%zu@%s/%s", line, path, filename);
1046 return strdupz(buffer);
1049 static inline void strip_quotes(char *s) {
1051 if(*s == '\'' || *s == '"') *s = ' ';
1056 int health_readfile(const char *path, const char *filename) {
1057 debug(D_HEALTH, "Health configuration reading file '%s/%s'", path, filename);
1059 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;
1060 char buffer[HEALTH_CONF_MAX_LINE + 1];
1062 if(unlikely(!hash_alarm)) {
1063 hash_alarm = simple_uhash(HEALTH_ALARM_KEY);
1064 hash_template = simple_uhash(HEALTH_TEMPLATE_KEY);
1065 hash_on = simple_uhash(HEALTH_ON_KEY);
1066 hash_calc = simple_uhash(HEALTH_CALC_KEY);
1067 hash_lookup = simple_uhash(HEALTH_LOOKUP_KEY);
1068 hash_green = simple_uhash(HEALTH_GREEN_KEY);
1069 hash_red = simple_uhash(HEALTH_RED_KEY);
1070 hash_warn = simple_uhash(HEALTH_WARN_KEY);
1071 hash_crit = simple_uhash(HEALTH_CRIT_KEY);
1072 hash_exec = simple_uhash(HEALTH_EXEC_KEY);
1073 hash_every = simple_uhash(HEALTH_EVERY_KEY);
1074 hash_units = simple_hash(HEALTH_UNITS_KEY);
1075 hash_info = simple_hash(HEALTH_INFO_KEY);
1078 snprintfz(buffer, HEALTH_CONF_MAX_LINE, "%s/%s", path, filename);
1079 FILE *fp = fopen(buffer, "r");
1081 error("Health configuration cannot read file '%s'.", buffer);
1086 RRDCALCTEMPLATE *rt = NULL;
1088 size_t line = 0, append = 0;
1090 while((s = fgets(&buffer[append], (int)(HEALTH_CONF_MAX_LINE - append), fp)) || append) {
1091 int stop_appending = !s;
1093 // info("Line %zu of file '%s/%s': '%s'", line, path, filename, s);
1096 // info("Trimmed line %zu of file '%s/%s': '%s'", line, path, filename, s);
1099 if(!stop_appending && s[append - 1] == '\\') {
1100 s[append - 1] = ' ';
1101 append = &s[append] - buffer;
1102 if(append < HEALTH_CONF_MAX_LINE)
1105 error("Health configuration has too long muli-line at line %zu of file '%s/%s'.", line, path, filename);
1111 while(*s && *s != ':') s++;
1113 error("Health configuration has invalid line %zu of file '%s/%s'. It does not contain a ':'. Ignoring it.", line, path, filename);
1121 value = trim(value);
1124 error("Health configuration has invalid line %zu of file '%s/%s'. Keyword is empty. Ignoring it.", line, path, filename);
1129 error("Health configuration has invalid line %zu of file '%s/%s'. value is empty. Ignoring it.", line, path, filename);
1133 // info("Health file '%s/%s', key '%s', value '%s'", path, filename, key, value);
1134 uint32_t hash = simple_uhash(key);
1136 if(hash == hash_alarm && !strcasecmp(key, HEALTH_ALARM_KEY)) {
1137 if(rc && !rrdcalc_add_alarm_from_config(&localhost, rc))
1138 rrdcalc_free(&localhost, rc);
1141 if (!rrdcalctemplate_add_template_from_config(&localhost, rt))
1142 rrdcalctemplate_free(&localhost, rt);
1146 rc = callocz(1, sizeof(RRDCALC));
1147 rc->name = strdupz(value);
1148 rc->hash = simple_hash(rc->name);
1149 rc->source = health_source_file(line, path, filename);
1153 rc->old_value = NAN;
1155 if(rrdvar_fix_name(rc->name))
1156 error("Health configuration renamed alarm '%s' to '%s'", value, rc->name);
1158 else if(hash == hash_template && !strcasecmp(key, HEALTH_TEMPLATE_KEY)) {
1160 if(!rrdcalc_add_alarm_from_config(&localhost, rc))
1161 rrdcalc_free(&localhost, rc);
1165 if(rt && !rrdcalctemplate_add_template_from_config(&localhost, rt))
1166 rrdcalctemplate_free(&localhost, rt);
1168 rt = callocz(1, sizeof(RRDCALCTEMPLATE));
1169 rt->name = strdupz(value);
1170 rt->hash_name = simple_hash(rt->name);
1171 rt->source = health_source_file(line, path, filename);
1175 if(rrdvar_fix_name(rt->name))
1176 error("Health configuration renamed template '%s' to '%s'", value, rt->name);
1179 if(hash == hash_on && !strcasecmp(key, HEALTH_ON_KEY)) {
1181 if(strcmp(rc->chart, value))
1182 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').",
1183 line, path, filename, rc->name, key, rc->chart, value, value);
1187 rc->chart = strdupz(value);
1188 rc->hash_chart = simple_hash(rc->chart);
1190 else if(hash == hash_lookup && !strcasecmp(key, HEALTH_LOOKUP_KEY)) {
1191 health_parse_db_lookup(line, path, filename, value, &rc->group, &rc->after, &rc->before,
1193 &rc->options, &rc->dimensions);
1195 else if(hash == hash_every && !strcasecmp(key, HEALTH_EVERY_KEY)) {
1196 if(!health_parse_duration(value, &rc->update_every))
1197 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' cannot parse duration: '%s'.",
1198 line, path, filename, rc->name, key, value);
1200 else if(hash == hash_green && !strcasecmp(key, HEALTH_GREEN_KEY)) {
1202 rc->green = strtold(value, &e);
1204 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' leaves this string unmatched: '%s'.",
1205 line, path, filename, rc->name, key, e);
1208 else if(hash == hash_red && !strcasecmp(key, HEALTH_RED_KEY)) {
1210 rc->red = strtold(value, &e);
1212 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' leaves this string unmatched: '%s'.",
1213 line, path, filename, rc->name, key, e);
1216 else if(hash == hash_calc && !strcasecmp(key, HEALTH_CALC_KEY)) {
1217 const char *failed_at = NULL;
1219 rc->calculation = expression_parse(value, &failed_at, &error);
1220 if(!rc->calculation) {
1221 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1222 line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
1225 else if(hash == hash_warn && !strcasecmp(key, HEALTH_WARN_KEY)) {
1226 const char *failed_at = NULL;
1228 rc->warning = expression_parse(value, &failed_at, &error);
1230 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1231 line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
1234 else if(hash == hash_crit && !strcasecmp(key, HEALTH_CRIT_KEY)) {
1235 const char *failed_at = NULL;
1237 rc->critical = expression_parse(value, &failed_at, &error);
1239 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1240 line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
1243 else if(hash == hash_exec && !strcasecmp(key, HEALTH_EXEC_KEY)) {
1245 if(strcmp(rc->exec, value))
1246 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').",
1247 line, path, filename, rc->name, key, rc->exec, value, value);
1251 rc->exec = strdupz(value);
1253 else if(hash == hash_units && !strcasecmp(key, HEALTH_UNITS_KEY)) {
1255 if(strcmp(rc->units, value))
1256 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').",
1257 line, path, filename, rc->name, key, rc->units, value, value);
1261 rc->units = strdupz(value);
1262 strip_quotes(rc->units);
1264 else if(hash == hash_info && !strcasecmp(key, HEALTH_INFO_KEY)) {
1266 if(strcmp(rc->info, value))
1267 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').",
1268 line, path, filename, rc->name, key, rc->info, value, value);
1272 rc->info = strdupz(value);
1273 strip_quotes(rc->info);
1276 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' has unknown key '%s'.",
1277 line, path, filename, rc->name, key);
1281 if(hash == hash_on && !strcasecmp(key, HEALTH_ON_KEY)) {
1283 if(strcmp(rt->context, value))
1284 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').",
1285 line, path, filename, rt->name, key, rt->context, value, value);
1289 rt->context = strdupz(value);
1290 rt->hash_context = simple_hash(rt->context);
1292 else if(hash == hash_lookup && !strcasecmp(key, HEALTH_LOOKUP_KEY)) {
1293 health_parse_db_lookup(line, path, filename, value, &rt->group, &rt->after, &rt->before,
1295 &rt->options, &rt->dimensions);
1297 else if(hash == hash_every && !strcasecmp(key, HEALTH_EVERY_KEY)) {
1298 if(!health_parse_duration(value, &rt->update_every))
1299 info("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' cannot parse duration: '%s'.",
1300 line, path, filename, rt->name, key, value);
1302 else if(hash == hash_green && !strcasecmp(key, HEALTH_GREEN_KEY)) {
1304 rt->green = strtold(value, &e);
1306 info("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' leaves this string unmatched: '%s'.",
1307 line, path, filename, rt->name, key, e);
1310 else if(hash == hash_red && !strcasecmp(key, HEALTH_RED_KEY)) {
1312 rt->red = strtold(value, &e);
1314 info("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' leaves this string unmatched: '%s'.",
1315 line, path, filename, rt->name, key, e);
1318 else if(hash == hash_calc && !strcasecmp(key, HEALTH_CALC_KEY)) {
1319 const char *failed_at = NULL;
1321 rt->calculation = expression_parse(value, &failed_at, &error);
1322 if(!rt->calculation) {
1323 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1324 line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
1327 else if(hash == hash_warn && !strcasecmp(key, HEALTH_WARN_KEY)) {
1328 const char *failed_at = NULL;
1330 rt->warning = expression_parse(value, &failed_at, &error);
1332 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1333 line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
1336 else if(hash == hash_crit && !strcasecmp(key, HEALTH_CRIT_KEY)) {
1337 const char *failed_at = NULL;
1339 rt->critical = expression_parse(value, &failed_at, &error);
1341 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1342 line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
1345 else if(hash == hash_exec && !strcasecmp(key, HEALTH_EXEC_KEY)) {
1347 if(strcmp(rt->exec, value))
1348 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').",
1349 line, path, filename, rt->name, key, rt->exec, value, value);
1353 rt->exec = strdupz(value);
1355 else if(hash == hash_units && !strcasecmp(key, HEALTH_UNITS_KEY)) {
1357 if(strcmp(rt->units, value))
1358 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').",
1359 line, path, filename, rt->name, key, rt->units, value, value);
1363 rt->units = strdupz(value);
1364 strip_quotes(rt->units);
1366 else if(hash == hash_info && !strcasecmp(key, HEALTH_INFO_KEY)) {
1368 if(strcmp(rt->info, value))
1369 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').",
1370 line, path, filename, rt->name, key, rt->info, value, value);
1374 rt->info = strdupz(value);
1375 strip_quotes(rt->info);
1378 error("Health configuration at line %zu of file '%s/%s' for template '%s' has unknown key '%s'.",
1379 line, path, filename, rt->name, key);
1383 error("Health configuration at line %zu of file '%s/%s' has unknown key '%s'. Expected either '" HEALTH_ALARM_KEY "' or '" HEALTH_TEMPLATE_KEY "'.",
1384 line, path, filename, key);
1388 if(rc && !rrdcalc_add_alarm_from_config(&localhost, rc))
1389 rrdcalc_free(&localhost, rc);
1391 if(rt && !rrdcalctemplate_add_template_from_config(&localhost, rt))
1392 rrdcalctemplate_free(&localhost, rt);
1398 void health_readdir(const char *path) {
1399 size_t pathlen = strlen(path);
1401 debug(D_HEALTH, "Health configuration reading directory '%s'", path);
1403 DIR *dir = opendir(path);
1405 error("Health configuration cannot open directory '%s'.", path);
1409 struct dirent *de = NULL;
1410 while ((de = readdir(dir))) {
1411 size_t len = strlen(de->d_name);
1413 if(de->d_type == DT_DIR
1415 (de->d_name[0] == '.' && de->d_name[1] == '\0')
1416 || (de->d_name[0] == '.' && de->d_name[1] == '.' && de->d_name[2] == '\0')
1420 else if(de->d_type == DT_DIR) {
1421 char *s = mallocz(pathlen + strlen(de->d_name) + 2);
1424 strcat(s, de->d_name);
1430 else if((de->d_type == DT_LNK || de->d_type == DT_REG) &&
1431 len > 5 && !strcmp(&de->d_name[len - 5], ".conf")) {
1432 health_readfile(path, de->d_name);
1439 static inline char *health_config_dir(void) {
1440 char buffer[FILENAME_MAX + 1];
1441 snprintfz(buffer, FILENAME_MAX, "%s/health.d", config_get("global", "config directory", CONFIG_DIR));
1442 return config_get("health", "health configuration directory", buffer);
1445 void health_init(void) {
1446 debug(D_HEALTH, "Health configuration initializing");
1448 if(!(health_enabled = config_get_boolean("health", "enabled", 1))) {
1449 debug(D_HEALTH, "Health is disabled.");
1453 char *path = health_config_dir();
1456 char buffer[FILENAME_MAX + 1];
1457 snprintfz(buffer, FILENAME_MAX, "%s/alarm-email.sh", config_get("global", "plugins directory", PLUGINS_DIR));
1458 health_default_exec = config_get("health", "script to execute on alarm", buffer);
1461 long n = config_get_number("health", "in memory max health log entries", (long)localhost.health_log.max);
1463 error("Health configuration has invalid max log entries %ld. Using default %u", n, localhost.health_log.max);
1464 config_set_number("health", "in memory max health log entries", (long)localhost.health_log.max);
1466 else localhost.health_log.max = (unsigned int)n;
1468 rrdhost_rwlock(&localhost);
1469 health_readdir(path);
1470 rrdhost_unlock(&localhost);
1473 // ----------------------------------------------------------------------------
1476 static inline void health_string2json(BUFFER *wb, const char *prefix, const char *label, const char *value, const char *suffix) {
1478 buffer_sprintf(wb, "%s\"%s\":\"%s\"%s", prefix, label, value, suffix);
1480 buffer_sprintf(wb, "%s\"%s\":null%s", prefix, label, suffix);
1483 static inline void health_alarm_entry2json_nolock(BUFFER *wb, ALARM_ENTRY *ae) {
1484 buffer_sprintf(wb, "\n\t{\n"
1486 "\t\t\"name\":\"%s\",\n"
1487 "\t\t\"chart\":\"%s\",\n"
1488 "\t\t\"family\":\"%s\",\n"
1489 "\t\t\"processed\":%s,\n"
1490 "\t\t\"updated\":%s,\n"
1491 "\t\t\"exec_run\":%s,\n"
1492 "\t\t\"exec_failed\":%s,\n"
1493 "\t\t\"exec\":\"%s\",\n"
1494 "\t\t\"exec_code\":%d,\n"
1495 "\t\t\"source\":\"%s\",\n"
1496 "\t\t\"units\":\"%s\",\n"
1497 "\t\t\"info\":\"%s\",\n"
1498 "\t\t\"when\":%lu,\n"
1499 "\t\t\"duration\":%lu,\n"
1500 "\t\t\"non_clear_duration\":%lu,\n"
1501 "\t\t\"status\":\"%s\",\n"
1502 "\t\t\"old_status\":\"%s\",\n",
1507 (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_PROCESSED)?"true":"false",
1508 (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_UPDATED)?"true":"false",
1509 (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_EXEC_RUN)?"true":"false",
1510 (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_EXEC_FAILED)?"true":"false",
1511 ae->exec?ae->exec:health_default_exec,
1514 ae->units?ae->units:"",
1515 ae->info?ae->info:"",
1516 (unsigned long)ae->when,
1517 (unsigned long)ae->duration,
1518 (unsigned long)ae->non_clear_duration,
1519 rrdcalc_status2string(ae->new_status),
1520 rrdcalc_status2string(ae->old_status)
1523 buffer_strcat(wb, "\t\t\"value\":");
1524 buffer_rrd_value(wb, ae->new_value);
1525 buffer_strcat(wb, ",\n");
1527 buffer_strcat(wb, "\t\t\"old_value\":");
1528 buffer_rrd_value(wb, ae->old_value);
1529 buffer_strcat(wb, "\n");
1531 buffer_strcat(wb, "\t}");
1534 void health_alarm_log2json(RRDHOST *host, BUFFER *wb) {
1535 pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
1537 buffer_strcat(wb, "[");
1539 unsigned int max = host->health_log.max;
1540 unsigned int count = 0;
1542 for(ae = host->health_log.alarms; ae && count < max ; count++, ae = ae->next) {
1543 if(likely(count)) buffer_strcat(wb, ",");
1544 health_alarm_entry2json_nolock(wb, ae);
1547 buffer_strcat(wb, "\n]\n");
1549 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1552 static inline void health_rrdcalc2json_nolock(BUFFER *wb, RRDCALC *rc) {
1554 "\t\t\"%s.%s\": {\n"
1555 "\t\t\t\"name\": \"%s\",\n"
1556 "\t\t\t\"chart\": \"%s\",\n"
1557 "\t\t\t\"family\": \"%s\",\n"
1558 "\t\t\t\"active\": %s,\n"
1559 "\t\t\t\"exec\": \"%s\",\n"
1560 "\t\t\t\"source\": \"%s\",\n"
1561 "\t\t\t\"units\": \"%s\",\n"
1562 "\t\t\t\"info\": \"%s\",\n"
1563 "\t\t\t\"status\": \"%s\",\n"
1564 "\t\t\t\"last_status_change\": %lu,\n"
1565 "\t\t\t\"last_updated\": %lu,\n"
1566 "\t\t\t\"next_update\": %lu,\n"
1567 "\t\t\t\"update_every\": %d,\n"
1568 , rc->chart, rc->name
1571 , (rc->rrdset && rc->rrdset->family)?rc->rrdset->family:""
1572 , (rc->rrdset)?"true":"false"
1573 , rc->exec?rc->exec:health_default_exec
1575 , rc->units?rc->units:""
1576 , rc->info?rc->info:""
1577 , rrdcalc_status2string(rc->status)
1578 , (unsigned long)rc->last_status_change
1579 , (unsigned long)rc->last_updated
1580 , (unsigned long)rc->next_update
1584 if(RRDCALC_HAS_DB_LOOKUP(rc)) {
1585 if(rc->dimensions && *rc->dimensions)
1586 health_string2json(wb, "\t\t\t", "lookup_dimensions", rc->dimensions, ",\n");
1589 "\t\t\t\"db_after\": %lu,\n"
1590 "\t\t\t\"db_before\": %lu,\n"
1591 "\t\t\t\"lookup_method\": \"%s\",\n"
1592 "\t\t\t\"lookup_after\": %d,\n"
1593 "\t\t\t\"lookup_before\": %d,\n"
1594 "\t\t\t\"lookup_options\": \"",
1595 (unsigned long) rc->db_after,
1596 (unsigned long) rc->db_before,
1597 group_method2string(rc->group),
1601 buffer_data_options2string(wb, rc->options);
1602 buffer_strcat(wb, "\",\n");
1605 if(rc->calculation) {
1606 health_string2json(wb, "\t\t\t", "calc", rc->calculation->source, ",\n");
1607 health_string2json(wb, "\t\t\t", "calc_parsed", rc->calculation->parsed_as, ",\n");
1611 health_string2json(wb, "\t\t\t", "warn", rc->warning->source, ",\n");
1612 health_string2json(wb, "\t\t\t", "warn_parsed", rc->warning->parsed_as, ",\n");
1616 health_string2json(wb, "\t\t\t", "crit", rc->critical->source, ",\n");
1617 health_string2json(wb, "\t\t\t", "crit_parsed", rc->critical->parsed_as, ",\n");
1620 buffer_strcat(wb, "\t\t\t\"green\":");
1621 buffer_rrd_value(wb, rc->green);
1622 buffer_strcat(wb, ",\n");
1624 buffer_strcat(wb, "\t\t\t\"red\":");
1625 buffer_rrd_value(wb, rc->red);
1626 buffer_strcat(wb, ",\n");
1628 buffer_strcat(wb, "\t\t\t\"value\":");
1629 buffer_rrd_value(wb, rc->value);
1630 buffer_strcat(wb, "\n");
1632 buffer_strcat(wb, "\t\t}");
1635 //void health_rrdcalctemplate2json_nolock(BUFFER *wb, RRDCALCTEMPLATE *rt) {
1639 void health_alarms2json(RRDHOST *host, BUFFER *wb, int all) {
1641 rrdhost_rdlock(&localhost);
1643 buffer_strcat(wb, "{\n\t\"alarms\": {\n");
1645 for(i = 0, rc = host->alarms; rc ; rc = rc->next) {
1649 if(!all && !(rc->status == RRDCALC_STATUS_WARNING || rc->status == RRDCALC_STATUS_CRITICAL))
1652 if(likely(i)) buffer_strcat(wb, ",\n");
1653 health_rrdcalc2json_nolock(wb, rc);
1657 // buffer_strcat(wb, "\n\t},\n\t\"templates\": {");
1659 // RRDCALCTEMPLATE *rt;
1660 // for(rt = host->templates; rt ; rt = rt->next)
1661 // health_rrdcalctemplate2json_nolock(wb, rt);
1663 buffer_sprintf(wb, "\n\t},\n\t\"now\": %lu\n}\n", (unsigned long)time(NULL));
1664 rrdhost_unlock(&localhost);
1668 // ----------------------------------------------------------------------------
1669 // re-load health configuration
1671 static inline void health_free_all_nolock(RRDHOST *host) {
1672 while(host->templates)
1673 rrdcalctemplate_free(host, host->templates);
1676 rrdcalc_free(host, host->alarms);
1679 void health_reload(void) {
1680 if(!health_enabled) {
1681 error("Health reload is requested, but health is not enabled.");
1685 char *path = health_config_dir();
1687 rrdhost_rwlock(&localhost);
1688 health_free_all_nolock(&localhost);
1689 rrdhost_unlock(&localhost);
1692 for(st = localhost.rrdset_root; st ; st = st->next) {
1697 rrdhost_rwlock(&localhost);
1698 health_readdir(path);
1699 rrdhost_unlock(&localhost);
1701 for(st = localhost.rrdset_root; st ; st = st->next) {
1702 rrdhost_rwlock(&localhost);
1704 rrdsetcalc_link_matching(st);
1705 rrdcalctemplate_link_matching(st);
1707 rrdhost_unlock(&localhost);
1712 // ----------------------------------------------------------------------------
1713 // health main thread and friends
1715 static inline int rrdcalc_isrunnable(RRDCALC *rc, time_t now, time_t *next_run) {
1716 if (unlikely(!rc->rrdset)) {
1717 debug(D_HEALTH, "Health not running alarm '%s.%s'. It is not linked to a chart.", rc->chart?rc->chart:"NOCHART", rc->name);
1721 if (unlikely(!rc->update_every)) {
1722 debug(D_HEALTH, "Health not running alarm '%s.%s'. It does not have an update frequency", rc->chart?rc->chart:"NOCHART", rc->name);
1726 if (unlikely(rc->next_update > now)) {
1727 if (*next_run > rc->next_update)
1728 *next_run = rc->next_update;
1730 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));
1737 static inline int rrdcalc_value2status(calculated_number n) {
1738 if(isnan(n)) return RRDCALC_STATUS_UNDEFINED;
1739 if(n) return RRDCALC_STATUS_RAISED;
1740 return RRDCALC_STATUS_CLEAR;
1743 static inline void health_alarm_execute(ALARM_ENTRY *ae) {
1744 if(ae->old_status == RRDCALC_STATUS_UNINITIALIZED && ae->new_status == RRDCALC_STATUS_CLEAR)
1747 char buffer[FILENAME_MAX + 1];
1750 const char *exec = ae->exec;
1751 if(!exec) exec = health_default_exec;
1753 snprintfz(buffer, FILENAME_MAX, "exec %s '%s' '%s' '%s' '%s' '%s' '%0.0Lf' '%0.0Lf' '%s' '%u' '%u' '%s' '%s'",
1756 ae->chart?ae->chart:"NOCAHRT",
1757 ae->family?ae->family:"NOFAMILY",
1758 rrdcalc_status2string(ae->new_status),
1759 rrdcalc_status2string(ae->old_status),
1762 ae->source?ae->source:"UNKNOWN",
1763 (uint32_t)ae->duration,
1764 (uint32_t)ae->non_clear_duration,
1765 ae->units?ae->units:"",
1766 ae->info?ae->info:""
1769 ae->notifications |= HEALTH_ENTRY_NOTIFICATIONS_EXEC_RUN;
1771 debug(D_HEALTH, "executing command '%s'", buffer);
1772 FILE *fp = mypopen(buffer, &command_pid);
1774 error("HEALTH: Cannot popen(\"%s\", \"r\").", buffer);
1777 debug(D_HEALTH, "HEALTH reading from command");
1778 char *s = fgets(buffer, FILENAME_MAX, fp);
1780 debug(D_HEALTH, "HEALTH closing command");
1781 ae->exec_code = mypclose(fp, command_pid);
1782 debug(D_HEALTH, "done executing command - returned with code %d", ae->exec_code);
1784 if(ae->exec_code != 0)
1785 ae->notifications |= HEALTH_ENTRY_NOTIFICATIONS_EXEC_FAILED;
1788 static inline void health_process_notifications(ALARM_ENTRY *ae) {
1789 info("Health alarm '%s.%s' = %0.2Lf - changed status from %s to %s",
1790 ae->chart?ae->chart:"NOCHART", ae->name,
1792 rrdcalc_status2string(ae->old_status),
1793 rrdcalc_status2string(ae->new_status)
1796 health_alarm_execute(ae);
1799 static inline void health_alarm_log(RRDHOST *host, time_t when,
1800 const char *name, const char *chart, const char *family,
1801 const char *exec, time_t duration,
1802 calculated_number old_value, calculated_number new_value,
1803 int old_status, int new_status,
1808 ALARM_ENTRY *ae = callocz(1, sizeof(ALARM_ENTRY));
1809 ae->name = strdupz(name);
1810 ae->hash_name = simple_hash(ae->name);
1813 ae->chart = strdupz(chart);
1814 ae->hash_chart = simple_hash(ae->chart);
1818 ae->family = strdupz(family);
1820 if(exec) ae->exec = strdupz(exec);
1821 if(source) ae->source = strdupz(source);
1822 if(units) ae->units = strdupz(units);
1823 if(info) ae->info = strdupz(info);
1825 ae->id = host->health_log.nextid++;
1827 ae->old_value = old_value;
1828 ae->new_value = new_value;
1829 ae->old_status = old_status;
1830 ae->new_status = new_status;
1831 ae->duration = duration;
1833 if(ae->old_status == RRDCALC_STATUS_WARNING || ae->old_status == RRDCALC_STATUS_CRITICAL)
1834 ae->non_clear_duration += ae->duration;
1837 pthread_rwlock_wrlock(&host->health_log.alarm_log_rwlock);
1838 ae->next = host->health_log.alarms;
1839 host->health_log.alarms = ae;
1840 host->health_log.count++;
1841 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1843 // match previous alarms
1844 pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
1846 for(t = host->health_log.alarms ; t ; t = t->next) {
1848 t->hash_name == ae->hash_name &&
1849 t->hash_chart == ae->hash_chart &&
1850 !strcmp(t->name, ae->name) &&
1851 t->chart && ae->chart && !strcmp(t->chart, ae->chart)) {
1853 if(!(t->notifications & HEALTH_ENTRY_NOTIFICATIONS_UPDATED) && !t->updated_by) {
1854 t->notifications |= HEALTH_ENTRY_NOTIFICATIONS_UPDATED;
1857 if((t->new_status == RRDCALC_STATUS_WARNING || t->new_status == RRDCALC_STATUS_CRITICAL) &&
1858 (t->old_status == RRDCALC_STATUS_WARNING || t->old_status == RRDCALC_STATUS_CRITICAL))
1859 ae->non_clear_duration += t->non_clear_duration;
1862 // no need to continue
1867 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1870 static inline void health_alarm_log_process(RRDHOST *host) {
1871 static uint32_t last_processed = 0;
1874 pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
1876 for(ae = host->health_log.alarms; ae ;ae = ae->next) {
1877 if(last_processed >= ae->id) break;
1879 if(!(ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_PROCESSED) &&
1880 !(ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_UPDATED)) {
1881 ae->notifications |= HEALTH_ENTRY_NOTIFICATIONS_PROCESSED;
1882 health_process_notifications(ae);
1886 if(host->health_log.alarms)
1887 last_processed = host->health_log.alarms->id;
1889 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1891 if(host->health_log.count <= host->health_log.max)
1894 // cleanup excess entries in the log
1895 pthread_rwlock_wrlock(&host->health_log.alarm_log_rwlock);
1897 ALARM_ENTRY *last = NULL;
1898 unsigned int count = host->health_log.max;
1899 for(ae = host->health_log.alarms; ae && count ; count--, last = ae, ae = ae->next) ;
1901 if(ae && last && last->next == ae)
1907 ALARM_ENTRY *t = ae->next;
1921 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1924 void *health_main(void *ptr) {
1927 info("HEALTH thread created with task id %d", gettid());
1929 if(pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, NULL) != 0)
1930 error("Cannot set pthread cancel type to DEFERRED.");
1932 if(pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL) != 0)
1933 error("Cannot set pthread cancel state to ENABLE.");
1935 int min_run_every = (int)config_get_number("health", "run at least every seconds", 10);
1936 if(min_run_every < 1) min_run_every = 1;
1938 BUFFER *wb = buffer_create(100);
1940 unsigned int loop = 0;
1941 while(health_enabled) {
1943 debug(D_HEALTH, "Health monitoring iteration no %u started", loop);
1945 int oldstate, runnable = 0;
1946 time_t now = time(NULL);
1947 time_t next_run = now + min_run_every;
1950 if (unlikely(pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate) != 0))
1951 error("Cannot set pthread cancel state to DISABLE.");
1953 rrdhost_rdlock(&localhost);
1955 // the first loop is to lookup values from the db
1956 for (rc = localhost.alarms; rc; rc = rc->next) {
1957 if (unlikely(!rrdcalc_isrunnable(rc, now, &next_run)))
1961 rc->old_value = rc->value;
1963 // 1. if there is database lookup, do it
1964 // 2. if there is calculation expression, run it
1966 if (unlikely(RRDCALC_HAS_DB_LOOKUP(rc))) {
1967 time_t old_db_timestamp = rc->db_before;
1968 int value_is_null = 0;
1970 int ret = rrd2value(rc->rrdset, wb, &rc->value,
1971 rc->dimensions, 1, rc->after, rc->before, rc->group,
1972 rc->options, &rc->db_after, &rc->db_before, &value_is_null);
1974 if (unlikely(ret != 200)) {
1975 // database lookup failed
1978 debug(D_HEALTH, "Health alarm '%s.%s': database lookup returned error %d", rc->chart?rc->chart:"NOCHART", rc->name, ret);
1980 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_ERROR))) {
1981 rc->rrdcalc_flags |= RRDCALC_FLAG_DB_ERROR;
1982 error("Health alarm '%s.%s': database lookup returned error %d", rc->chart?rc->chart:"NOCHART", rc->name, ret);
1985 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_ERROR))
1986 rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_ERROR;
1988 if (unlikely(old_db_timestamp == rc->db_before)) {
1989 // database is stale
1991 debug(D_HEALTH, "Health alarm '%s.%s': database is stale", rc->chart?rc->chart:"NOCHART", rc->name);
1993 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_STALE))) {
1994 rc->rrdcalc_flags |= RRDCALC_FLAG_DB_STALE;
1995 error("Health alarm '%s.%s': database is stale", rc->chart?rc->chart:"NOCHART", rc->name);
1998 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_STALE))
1999 rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_STALE;
2001 if (unlikely(value_is_null)) {
2002 // collected value is null
2006 debug(D_HEALTH, "Health alarm '%s.%s': database lookup returned empty value (possibly value is not collected yet)",
2007 rc->chart?rc->chart:"NOCHART", rc->name);
2009 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_NAN))) {
2010 rc->rrdcalc_flags |= RRDCALC_FLAG_DB_NAN;
2011 error("Health alarm '%s.%s': database lookup returned empty value (possibly value is not collected yet)",
2012 rc->chart?rc->chart:"NOCHART", rc->name);
2015 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_NAN))
2016 rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_NAN;
2018 debug(D_HEALTH, "Health alarm '%s.%s': database lookup gave value "
2019 CALCULATED_NUMBER_FORMAT, rc->chart?rc->chart:"NOCHART", rc->name, rc->value);
2022 if(unlikely(rc->calculation)) {
2023 if (unlikely(!expression_evaluate(rc->calculation))) {
2024 // calculation failed
2028 debug(D_HEALTH, "Health alarm '%s.%s': failed to evaluate calculation with error: %s",
2029 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->calculation->error_msg));
2031 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_CALC_ERROR))) {
2032 rc->rrdcalc_flags |= RRDCALC_FLAG_CALC_ERROR;
2033 error("Health alarm '%s.%s': failed to evaluate calculation with error: %s",
2034 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->calculation->error_msg));
2038 if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_CALC_ERROR))
2039 rc->rrdcalc_flags &= ~RRDCALC_FLAG_CALC_ERROR;
2041 debug(D_HEALTH, "Health alarm '%s.%s': calculation expression gave value "
2042 CALCULATED_NUMBER_FORMAT
2043 ": %s (source: %s)",
2044 rc->chart?rc->chart:"NOCHART", rc->name,
2045 rc->calculation->result,
2046 buffer_tostring(rc->calculation->error_msg),
2050 rc->value = rc->calculation->result;
2054 rrdhost_unlock(&localhost);
2057 rrdhost_rdlock(&localhost);
2059 for (rc = localhost.alarms; rc; rc = rc->next) {
2060 if (unlikely(!rrdcalc_isrunnable(rc, now, &next_run)))
2063 int warning_status = RRDCALC_STATUS_UNDEFINED;
2064 int critical_status = RRDCALC_STATUS_UNDEFINED;
2066 if(unlikely(rc->warning)) {
2067 if(unlikely(!expression_evaluate(rc->warning))) {
2068 // calculation failed
2070 debug(D_HEALTH, "Health alarm '%s.%s': warning expression failed with error: %s",
2071 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->warning->error_msg));
2073 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_WARN_ERROR))) {
2074 rc->rrdcalc_flags |= RRDCALC_FLAG_WARN_ERROR;
2075 error("Health alarm '%s.%s': warning expression failed with error: %s",
2076 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->warning->error_msg));
2080 if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_WARN_ERROR))
2081 rc->rrdcalc_flags &= ~RRDCALC_FLAG_WARN_ERROR;
2083 debug(D_HEALTH, "Health alarm '%s.%s': warning expression gave value "
2084 CALCULATED_NUMBER_FORMAT
2085 ": %s (source: %s)",
2086 rc->chart?rc->chart:"NOCHART", rc->name,
2087 rc->warning->result,
2088 buffer_tostring(rc->warning->error_msg),
2092 warning_status = rrdcalc_value2status(rc->warning->result);
2096 if(unlikely(rc->critical)) {
2097 if(unlikely(!expression_evaluate(rc->critical))) {
2098 // calculation failed
2100 debug(D_HEALTH, "Health alarm '%s.%s': critical expression failed with error: %s",
2101 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->critical->error_msg));
2103 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_CRIT_ERROR))) {
2104 rc->rrdcalc_flags |= RRDCALC_FLAG_CRIT_ERROR;
2105 error("Health alarm '%s.%s': critical expression failed with error: %s",
2106 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->critical->error_msg));
2110 if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_CRIT_ERROR))
2111 rc->rrdcalc_flags &= ~RRDCALC_FLAG_CRIT_ERROR;
2113 debug(D_HEALTH, "Health alarm '%s.%s': critical expression gave value "
2114 CALCULATED_NUMBER_FORMAT
2115 ": %s (source: %s)",
2116 rc->chart?rc->chart:"NOCHART", rc->name,
2117 rc->critical->result,
2118 buffer_tostring(rc->critical->error_msg),
2122 critical_status = rrdcalc_value2status(rc->critical->result);
2126 int status = RRDCALC_STATUS_UNDEFINED;
2128 switch(warning_status) {
2129 case RRDCALC_STATUS_CLEAR:
2130 status = RRDCALC_STATUS_CLEAR;
2133 case RRDCALC_STATUS_RAISED:
2134 status = RRDCALC_STATUS_WARNING;
2141 switch(critical_status) {
2142 case RRDCALC_STATUS_CLEAR:
2143 if(status == RRDCALC_STATUS_UNDEFINED)
2144 status = RRDCALC_STATUS_CLEAR;
2147 case RRDCALC_STATUS_RAISED:
2148 status = RRDCALC_STATUS_CRITICAL;
2155 if(status != rc->status) {
2156 health_alarm_log(&localhost, time(NULL), rc->name, rc->rrdset->id, rc->rrdset->family, rc->exec, now - rc->last_status_change, rc->old_value, rc->value, rc->status, status, rc->source, rc->units, rc->info);
2157 rc->last_status_change = now;
2158 rc->status = status;
2161 rc->last_updated = now;
2162 rc->next_update = now + rc->update_every;
2164 if (next_run > rc->next_update)
2165 next_run = rc->next_update;
2168 rrdhost_unlock(&localhost);
2171 if (unlikely(pthread_setcancelstate(oldstate, NULL) != 0))
2172 error("Cannot set pthread cancel state to RESTORE (%d).", oldstate);
2174 // execute notifications
2176 health_alarm_log_process(&localhost);
2179 if(now < next_run) {
2180 debug(D_HEALTH, "Health monitoring iteration no %u done. Next iteration in %d secs",
2181 loop, (int) (next_run - now));
2182 sleep_usec(1000000 * (unsigned long long) (next_run - now));
2185 debug(D_HEALTH, "Health monitoring iteration no %u done. Next iteration now", loop);
2191 info("HEALTH thread exiting");