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(unlikely(rc->rrdset))
491 if(unlikely(rrdcalc_is_matching_this_rrdset(rc, st)))
492 rrdsetcalc_link(st, rc);
496 // this has to be called while the RRDHOST is locked
497 inline void rrdsetcalc_unlink(RRDCALC *rc) {
498 RRDSET *st = rc->rrdset;
501 error("Requested to unlink RRDCALC '%s.%s' which is not linked to any RRDSET", rc->chart?rc->chart:"NOCHART", rc->name);
505 RRDHOST *host = st->rrdhost;
507 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);
511 rc->rrdset_prev->rrdset_next = rc->rrdset_next;
514 rc->rrdset_next->rrdset_prev = rc->rrdset_prev;
517 st->alarms = rc->rrdset_next;
519 rc->rrdset_prev = rc->rrdset_next = NULL;
521 rrdvar_free(st->rrdhost, &st->variables_root_index, rc->local);
524 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rc->family);
527 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rc->hostid);
530 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rc->hostname);
535 // RRDCALC will remain in RRDHOST
536 // so that if the matching chart is found in the future
537 // it will be applied automatically
540 RRDCALC *rrdcalc_find(RRDSET *st, const char *name) {
542 uint32_t hash = simple_hash(name);
544 for( rc = st->alarms; rc ; rc = rc->rrdset_next ) {
545 if(unlikely(rc->hash == hash && !strcmp(rc->name, name)))
552 static inline int rrdcalc_exists(RRDHOST *host, const char *name, uint32_t hash) {
555 // make sure it does not already exist
556 for(rc = host->alarms; rc ; rc = rc->next) {
557 if (unlikely(rc->hash == hash && !strcmp(name, rc->name))) {
558 error("Health alarm '%s' already exists in host '%s'.", name, host->hostname);
566 static inline void rrdcalc_create_part2(RRDHOST *host, RRDCALC *rc) {
567 rrdhost_check_rdlock(host);
569 if(rc->calculation) {
570 rc->calculation->this = &rc->value;
571 rc->calculation->after = &rc->db_after;
572 rc->calculation->before = &rc->db_before;
573 rc->calculation->rrdcalc = rc;
577 rc->warning->this = &rc->value;
578 rc->warning->after = &rc->db_after;
579 rc->warning->before = &rc->db_before;
580 rc->warning->rrdcalc = rc;
584 rc->critical->this = &rc->value;
585 rc->critical->after = &rc->db_after;
586 rc->critical->before = &rc->db_before;
587 rc->critical->rrdcalc = rc;
590 // link it to the host
591 if(likely(host->alarms)) {
594 for(t = host->alarms; t && t->next ; t = t->next) ;
601 // link it to its chart
603 for(st = host->rrdset_root; st ; st = st->next) {
604 if(rrdcalc_is_matching_this_rrdset(rc, st)) {
605 rrdsetcalc_link(st, rc);
611 static inline uint32_t rrdcalc_fullname(char *fullname, size_t len, const char *chart, const char *name) {
612 snprintfz(fullname, len - 1, "%s%s%s", chart?chart:"", chart?".":"", name);
613 rrdvar_fix_name(fullname);
614 return simple_hash(fullname);
617 static inline RRDCALC *rrdcalc_create(RRDHOST *host, const char *name, const char *chart, const char *dimensions,
618 const char *units, const char *info,
619 int group_method, int after, int before, int update_every, uint32_t options,
620 calculated_number green, calculated_number red,
621 const char *exec, const char *source,
622 const char *calc, const char *warn, const char *crit) {
624 char fullname[RRDVAR_MAX_LENGTH + 1];
625 uint32_t hash = rrdcalc_fullname(fullname, RRDVAR_MAX_LENGTH + 1, chart, name);
627 if(rrdcalc_exists(host, fullname, hash))
630 RRDCALC *rc = callocz(1, sizeof(RRDCALC));
632 rc->name = strdupz(name);
633 rc->hash = simple_hash(rc->name);
635 rc->chart = strdupz(chart);
636 rc->hash_chart = simple_hash(rc->chart);
638 if(dimensions) rc->dimensions = strdupz(dimensions);
645 rc->group = group_method;
648 rc->update_every = update_every;
649 rc->options = options;
651 if(exec) rc->exec = strdupz(exec);
652 if(source) rc->source = strdupz(source);
653 if(units) rc->units = strdupz(units);
654 if(info) rc->info = strdupz(info);
657 rc->calculation = expression_parse(calc, NULL, NULL);
659 error("Health alarm '%s.%s': failed to parse calculation expression '%s'", chart, name, calc);
662 rc->warning = expression_parse(warn, NULL, NULL);
664 error("Health alarm '%s.%s': failed to re-parse warning expression '%s'", chart, name, warn);
667 rc->critical = expression_parse(crit, NULL, NULL);
669 error("Health alarm '%s.%s': failed to re-parse critical expression '%s'", chart, name, crit);
672 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",
673 (rc->chart)?rc->chart:"NOCHART",
675 (rc->exec)?rc->exec:"DEFAULT",
682 (rc->dimensions)?rc->dimensions:"NONE",
684 (rc->calculation)?rc->calculation->parsed_as:"NONE",
685 (rc->warning)?rc->warning->parsed_as:"NONE",
686 (rc->critical)?rc->critical->parsed_as:"NONE",
690 rrdcalc_create_part2(host, rc);
694 void rrdcalc_free(RRDHOST *host, RRDCALC *rc) {
697 debug(D_HEALTH, "Health removing alarm '%s.%s' of host '%s'", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);
699 // unlink it from RRDSET
700 if(rc->rrdset) rrdsetcalc_unlink(rc);
702 // unlink it from RRDHOST
703 if(unlikely(rc == host->alarms))
704 host->alarms = rc->next;
706 else if(likely(host->alarms)) {
707 RRDCALC *t, *last = host->alarms;
709 for(t = last->next; t && t != rc; last = t, t = t->next) ;
710 if(last && last->next == rc)
711 last->next = rc->next;
713 error("Cannot unlink alarm '%s.%s' from host '%s': not found", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);
716 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);
718 expression_free(rc->calculation);
719 expression_free(rc->warning);
720 expression_free(rc->critical);
725 freez(rc->dimensions);
733 // ----------------------------------------------------------------------------
734 // RRDCALCTEMPLATE management
736 void rrdcalctemplate_link_matching(RRDSET *st) {
739 for(rt = st->rrdhost->templates; rt ; rt = rt->next) {
740 if(rt->hash_context == st->hash_context && !strcmp(rt->context, st->context)) {
741 RRDCALC *rc = rrdcalc_create(st->rrdhost, rt->name, st->id,
742 rt->dimensions, rt->units, rt->info, rt->group, rt->after, rt->before, rt->update_every, rt->options,
743 rt->green, rt->red, rt->exec, rt->source,
744 (rt->calculation)?rt->calculation->source:NULL,
745 (rt->warning)?rt->warning->source:NULL,
746 (rt->critical)?rt->critical->source:NULL);
749 error("Health tried to create alarm from template '%s', but it failed", rt->name);
751 #ifdef NETDATA_INTERNAL_CHECKS
752 else if(rc->rrdset != st)
753 error("Health alarm '%s.%s' should be linked to chart '%s', but it is not", rc->chart?rc->chart:"NOCHART", rc->name, st->id);
759 static inline void rrdcalctemplate_free(RRDHOST *host, RRDCALCTEMPLATE *rt) {
760 debug(D_HEALTH, "Health removing template '%s' of host '%s'", rt->name, host->hostname);
762 if(host->templates) {
763 if(host->templates == rt) {
764 host->templates = rt->next;
767 RRDCALCTEMPLATE *t, *last = host->templates;
768 for (t = last->next; t && t != rt; last = t, t = t->next ) ;
769 if(last && last->next == rt) {
770 last->next = rt->next;
774 error("Cannot find RRDCALCTEMPLATE '%s' linked in host '%s'", rt->name, host->hostname);
778 expression_free(rt->calculation);
779 expression_free(rt->warning);
780 expression_free(rt->critical);
788 freez(rt->dimensions);
792 // ----------------------------------------------------------------------------
793 // load health configuration
795 #define HEALTH_CONF_MAX_LINE 4096
797 #define HEALTH_ALARM_KEY "alarm"
798 #define HEALTH_TEMPLATE_KEY "template"
799 #define HEALTH_ON_KEY "on"
800 #define HEALTH_LOOKUP_KEY "lookup"
801 #define HEALTH_CALC_KEY "calc"
802 #define HEALTH_EVERY_KEY "every"
803 #define HEALTH_GREEN_KEY "green"
804 #define HEALTH_RED_KEY "red"
805 #define HEALTH_WARN_KEY "warn"
806 #define HEALTH_CRIT_KEY "crit"
807 #define HEALTH_EXEC_KEY "exec"
808 #define HEALTH_UNITS_KEY "units"
809 #define HEALTH_INFO_KEY "info"
811 static inline int rrdcalc_add_alarm_from_config(RRDHOST *host, RRDCALC *rc) {
813 char fullname[RRDVAR_MAX_LENGTH + 1];
814 uint32_t hash = rrdcalc_fullname(fullname, RRDVAR_MAX_LENGTH + 1, rc->chart, rc->name);
816 if (rrdcalc_exists(host, fullname, hash))
821 error("Health configuration for alarm '%s' does not have a chart", rc->name);
825 if(!rc->update_every) {
826 error("Health configuration for alarm '%s.%s' has no frequency (parameter 'every'). Ignoring it.", rc->chart?rc->chart:"NOCHART", rc->name);
830 if(!RRDCALC_HAS_DB_LOOKUP(rc) && !rc->warning && !rc->critical) {
831 error("Health configuration for alarm '%s.%s' is useless (no calculation, no warning and no critical evaluation)", rc->chart?rc->chart:"NOCHART", rc->name);
835 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",
836 rc->chart?rc->chart:"NOCHART",
838 (rc->exec)?rc->exec:"DEFAULT",
845 (rc->dimensions)?rc->dimensions:"NONE",
847 (rc->calculation)?rc->calculation->parsed_as:"NONE",
848 (rc->warning)?rc->warning->parsed_as:"NONE",
849 (rc->critical)?rc->critical->parsed_as:"NONE",
853 rrdcalc_create_part2(host, rc);
857 static inline int rrdcalctemplate_add_template_from_config(RRDHOST *host, RRDCALCTEMPLATE *rt) {
858 if(unlikely(!rt->context)) {
859 error("Health configuration for template '%s' does not have a context", rt->name);
863 if(unlikely(!rt->update_every)) {
864 error("Health configuration for template '%s' has no frequency (parameter 'every'). Ignoring it.", rt->name);
868 if(unlikely(!RRDCALCTEMPLATE_HAS_CALCULATION(rt) && !rt->warning && !rt->critical)) {
869 error("Health configuration for template '%s' is useless (no calculation, no warning and no critical evaluation)", rt->name);
873 RRDCALCTEMPLATE *t, *last = NULL;
874 for (t = host->templates; t ; last = t, t = t->next) {
875 if(unlikely(t->hash_name == rt->hash_name && !strcmp(t->name, rt->name))) {
876 error("Health configuration template '%s' already exists for host '%s'.", rt->name, host->hostname);
881 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'",
883 (rt->context)?rt->context:"NONE",
884 (rt->exec)?rt->exec:"DEFAULT",
891 (rt->dimensions)?rt->dimensions:"NONE",
893 (rt->calculation)?rt->calculation->parsed_as:"NONE",
894 (rt->warning)?rt->warning->parsed_as:"NONE",
895 (rt->critical)?rt->critical->parsed_as:"NONE",
903 rt->next = host->templates;
904 host->templates = rt;
910 static inline int health_parse_duration(char *string, int *result) {
911 // make sure it is a number
912 if(!*string || !(isdigit(*string) || *string == '+' || *string == '-')) {
918 calculated_number n = strtold(string, &e);
922 *result = (int) (n * 86400 * 365);
925 *result = (int) (n * 86400 * 30);
928 *result = (int) (n * 86400 * 7);
931 *result = (int) (n * 86400);
934 *result = (int) (n * 3600);
937 *result = (int) (n * 60);
952 static inline int health_parse_db_lookup(
953 size_t line, const char *path, const char *file, char *string,
954 int *group_method, int *after, int *before, int *every,
955 uint32_t *options, char **dimensions
957 debug(D_HEALTH, "Health configuration parsing database lookup %zu@%s/%s: %s", line, path, file, string);
959 if(*dimensions) freez(*dimensions);
966 char *s = string, *key;
968 // first is the group method
970 while(*s && !isspace(*s)) s++;
971 while(*s && isspace(*s)) *s++ = '\0';
973 error("Health configuration invalid chart calculation at line %zu of file '%s/%s': expected group method followed by the 'after' time, but got '%s'",
974 line, path, file, key);
978 if((*group_method = web_client_api_request_v1_data_group(key, -1)) == -1) {
979 error("Health configuration at line %zu of file '%s/%s': invalid group method '%s'",
980 line, path, file, key);
984 // then is the 'after' time
986 while(*s && !isspace(*s)) s++;
987 while(*s && isspace(*s)) *s++ = '\0';
989 if(!health_parse_duration(key, after)) {
990 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' after group method",
991 line, path, file, key);
996 *every = abs(*after);
998 // now we may have optional parameters
1001 while(*s && !isspace(*s)) s++;
1002 while(*s && isspace(*s)) *s++ = '\0';
1005 if(!strcasecmp(key, "at")) {
1007 while(*s && !isspace(*s)) s++;
1008 while(*s && isspace(*s)) *s++ = '\0';
1010 if (!health_parse_duration(value, before)) {
1011 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' for '%s' keyword",
1012 line, path, file, value, key);
1015 else if(!strcasecmp(key, HEALTH_EVERY_KEY)) {
1017 while(*s && !isspace(*s)) s++;
1018 while(*s && isspace(*s)) *s++ = '\0';
1020 if (!health_parse_duration(value, every)) {
1021 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' for '%s' keyword",
1022 line, path, file, value, key);
1025 else if(!strcasecmp(key, "absolute") || !strcasecmp(key, "abs") || !strcasecmp(key, "absolute_sum")) {
1026 *options |= RRDR_OPTION_ABSOLUTE;
1028 else if(!strcasecmp(key, "min2max")) {
1029 *options |= RRDR_OPTION_MIN2MAX;
1031 else if(!strcasecmp(key, "null2zero")) {
1032 *options |= RRDR_OPTION_NULL2ZERO;
1034 else if(!strcasecmp(key, "percentage")) {
1035 *options |= RRDR_OPTION_PERCENTAGE;
1037 else if(!strcasecmp(key, "unaligned")) {
1038 *options |= RRDR_OPTION_NOT_ALIGNED;
1040 else if(!strcasecmp(key, "of")) {
1041 if(*s && strcasecmp(s, "all"))
1042 *dimensions = strdupz(s);
1046 error("Health configuration at line %zu of file '%s/%s': unknown keyword '%s'",
1047 line, path, file, key);
1054 static inline char *health_source_file(size_t line, const char *path, const char *filename) {
1055 char buffer[FILENAME_MAX + 1];
1056 snprintfz(buffer, FILENAME_MAX, "%zu@%s/%s", line, path, filename);
1057 return strdupz(buffer);
1060 static inline void strip_quotes(char *s) {
1062 if(*s == '\'' || *s == '"') *s = ' ';
1067 int health_readfile(const char *path, const char *filename) {
1068 debug(D_HEALTH, "Health configuration reading file '%s/%s'", path, filename);
1070 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;
1071 char buffer[HEALTH_CONF_MAX_LINE + 1];
1073 if(unlikely(!hash_alarm)) {
1074 hash_alarm = simple_uhash(HEALTH_ALARM_KEY);
1075 hash_template = simple_uhash(HEALTH_TEMPLATE_KEY);
1076 hash_on = simple_uhash(HEALTH_ON_KEY);
1077 hash_calc = simple_uhash(HEALTH_CALC_KEY);
1078 hash_lookup = simple_uhash(HEALTH_LOOKUP_KEY);
1079 hash_green = simple_uhash(HEALTH_GREEN_KEY);
1080 hash_red = simple_uhash(HEALTH_RED_KEY);
1081 hash_warn = simple_uhash(HEALTH_WARN_KEY);
1082 hash_crit = simple_uhash(HEALTH_CRIT_KEY);
1083 hash_exec = simple_uhash(HEALTH_EXEC_KEY);
1084 hash_every = simple_uhash(HEALTH_EVERY_KEY);
1085 hash_units = simple_hash(HEALTH_UNITS_KEY);
1086 hash_info = simple_hash(HEALTH_INFO_KEY);
1089 snprintfz(buffer, HEALTH_CONF_MAX_LINE, "%s/%s", path, filename);
1090 FILE *fp = fopen(buffer, "r");
1092 error("Health configuration cannot read file '%s'.", buffer);
1097 RRDCALCTEMPLATE *rt = NULL;
1099 size_t line = 0, append = 0;
1101 while((s = fgets(&buffer[append], (int)(HEALTH_CONF_MAX_LINE - append), fp)) || append) {
1102 int stop_appending = !s;
1104 // info("Line %zu of file '%s/%s': '%s'", line, path, filename, s);
1107 // info("Trimmed line %zu of file '%s/%s': '%s'", line, path, filename, s);
1110 if(!stop_appending && s[append - 1] == '\\') {
1111 s[append - 1] = ' ';
1112 append = &s[append] - buffer;
1113 if(append < HEALTH_CONF_MAX_LINE)
1116 error("Health configuration has too long muli-line at line %zu of file '%s/%s'.", line, path, filename);
1122 while(*s && *s != ':') s++;
1124 error("Health configuration has invalid line %zu of file '%s/%s'. It does not contain a ':'. Ignoring it.", line, path, filename);
1132 value = trim(value);
1135 error("Health configuration has invalid line %zu of file '%s/%s'. Keyword is empty. Ignoring it.", line, path, filename);
1140 error("Health configuration has invalid line %zu of file '%s/%s'. value is empty. Ignoring it.", line, path, filename);
1144 // info("Health file '%s/%s', key '%s', value '%s'", path, filename, key, value);
1145 uint32_t hash = simple_uhash(key);
1147 if(hash == hash_alarm && !strcasecmp(key, HEALTH_ALARM_KEY)) {
1148 if(rc && !rrdcalc_add_alarm_from_config(&localhost, rc))
1149 rrdcalc_free(&localhost, rc);
1152 if (!rrdcalctemplate_add_template_from_config(&localhost, rt))
1153 rrdcalctemplate_free(&localhost, rt);
1157 rc = callocz(1, sizeof(RRDCALC));
1158 rc->name = strdupz(value);
1159 rc->hash = simple_hash(rc->name);
1160 rc->source = health_source_file(line, path, filename);
1164 rc->old_value = NAN;
1166 if(rrdvar_fix_name(rc->name))
1167 error("Health configuration renamed alarm '%s' to '%s'", value, rc->name);
1169 else if(hash == hash_template && !strcasecmp(key, HEALTH_TEMPLATE_KEY)) {
1171 if(!rrdcalc_add_alarm_from_config(&localhost, rc))
1172 rrdcalc_free(&localhost, rc);
1176 if(rt && !rrdcalctemplate_add_template_from_config(&localhost, rt))
1177 rrdcalctemplate_free(&localhost, rt);
1179 rt = callocz(1, sizeof(RRDCALCTEMPLATE));
1180 rt->name = strdupz(value);
1181 rt->hash_name = simple_hash(rt->name);
1182 rt->source = health_source_file(line, path, filename);
1186 if(rrdvar_fix_name(rt->name))
1187 error("Health configuration renamed template '%s' to '%s'", value, rt->name);
1190 if(hash == hash_on && !strcasecmp(key, HEALTH_ON_KEY)) {
1192 if(strcmp(rc->chart, value))
1193 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').",
1194 line, path, filename, rc->name, key, rc->chart, value, value);
1198 rc->chart = strdupz(value);
1199 rc->hash_chart = simple_hash(rc->chart);
1201 else if(hash == hash_lookup && !strcasecmp(key, HEALTH_LOOKUP_KEY)) {
1202 health_parse_db_lookup(line, path, filename, value, &rc->group, &rc->after, &rc->before,
1204 &rc->options, &rc->dimensions);
1206 else if(hash == hash_every && !strcasecmp(key, HEALTH_EVERY_KEY)) {
1207 if(!health_parse_duration(value, &rc->update_every))
1208 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' cannot parse duration: '%s'.",
1209 line, path, filename, rc->name, key, value);
1211 else if(hash == hash_green && !strcasecmp(key, HEALTH_GREEN_KEY)) {
1213 rc->green = strtold(value, &e);
1215 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' leaves this string unmatched: '%s'.",
1216 line, path, filename, rc->name, key, e);
1219 else if(hash == hash_red && !strcasecmp(key, HEALTH_RED_KEY)) {
1221 rc->red = strtold(value, &e);
1223 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' leaves this string unmatched: '%s'.",
1224 line, path, filename, rc->name, key, e);
1227 else if(hash == hash_calc && !strcasecmp(key, HEALTH_CALC_KEY)) {
1228 const char *failed_at = NULL;
1230 rc->calculation = expression_parse(value, &failed_at, &error);
1231 if(!rc->calculation) {
1232 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1233 line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
1236 else if(hash == hash_warn && !strcasecmp(key, HEALTH_WARN_KEY)) {
1237 const char *failed_at = NULL;
1239 rc->warning = expression_parse(value, &failed_at, &error);
1241 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1242 line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
1245 else if(hash == hash_crit && !strcasecmp(key, HEALTH_CRIT_KEY)) {
1246 const char *failed_at = NULL;
1248 rc->critical = expression_parse(value, &failed_at, &error);
1250 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1251 line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
1254 else if(hash == hash_exec && !strcasecmp(key, HEALTH_EXEC_KEY)) {
1256 if(strcmp(rc->exec, value))
1257 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').",
1258 line, path, filename, rc->name, key, rc->exec, value, value);
1262 rc->exec = strdupz(value);
1264 else if(hash == hash_units && !strcasecmp(key, HEALTH_UNITS_KEY)) {
1266 if(strcmp(rc->units, 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->units, value, value);
1272 rc->units = strdupz(value);
1273 strip_quotes(rc->units);
1275 else if(hash == hash_info && !strcasecmp(key, HEALTH_INFO_KEY)) {
1277 if(strcmp(rc->info, value))
1278 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').",
1279 line, path, filename, rc->name, key, rc->info, value, value);
1283 rc->info = strdupz(value);
1284 strip_quotes(rc->info);
1287 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' has unknown key '%s'.",
1288 line, path, filename, rc->name, key);
1292 if(hash == hash_on && !strcasecmp(key, HEALTH_ON_KEY)) {
1294 if(strcmp(rt->context, value))
1295 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').",
1296 line, path, filename, rt->name, key, rt->context, value, value);
1300 rt->context = strdupz(value);
1301 rt->hash_context = simple_hash(rt->context);
1303 else if(hash == hash_lookup && !strcasecmp(key, HEALTH_LOOKUP_KEY)) {
1304 health_parse_db_lookup(line, path, filename, value, &rt->group, &rt->after, &rt->before,
1306 &rt->options, &rt->dimensions);
1308 else if(hash == hash_every && !strcasecmp(key, HEALTH_EVERY_KEY)) {
1309 if(!health_parse_duration(value, &rt->update_every))
1310 info("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' cannot parse duration: '%s'.",
1311 line, path, filename, rt->name, key, value);
1313 else if(hash == hash_green && !strcasecmp(key, HEALTH_GREEN_KEY)) {
1315 rt->green = strtold(value, &e);
1317 info("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' leaves this string unmatched: '%s'.",
1318 line, path, filename, rt->name, key, e);
1321 else if(hash == hash_red && !strcasecmp(key, HEALTH_RED_KEY)) {
1323 rt->red = strtold(value, &e);
1325 info("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' leaves this string unmatched: '%s'.",
1326 line, path, filename, rt->name, key, e);
1329 else if(hash == hash_calc && !strcasecmp(key, HEALTH_CALC_KEY)) {
1330 const char *failed_at = NULL;
1332 rt->calculation = expression_parse(value, &failed_at, &error);
1333 if(!rt->calculation) {
1334 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1335 line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
1338 else if(hash == hash_warn && !strcasecmp(key, HEALTH_WARN_KEY)) {
1339 const char *failed_at = NULL;
1341 rt->warning = expression_parse(value, &failed_at, &error);
1343 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1344 line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
1347 else if(hash == hash_crit && !strcasecmp(key, HEALTH_CRIT_KEY)) {
1348 const char *failed_at = NULL;
1350 rt->critical = expression_parse(value, &failed_at, &error);
1352 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1353 line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
1356 else if(hash == hash_exec && !strcasecmp(key, HEALTH_EXEC_KEY)) {
1358 if(strcmp(rt->exec, value))
1359 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').",
1360 line, path, filename, rt->name, key, rt->exec, value, value);
1364 rt->exec = strdupz(value);
1366 else if(hash == hash_units && !strcasecmp(key, HEALTH_UNITS_KEY)) {
1368 if(strcmp(rt->units, 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->units, value, value);
1374 rt->units = strdupz(value);
1375 strip_quotes(rt->units);
1377 else if(hash == hash_info && !strcasecmp(key, HEALTH_INFO_KEY)) {
1379 if(strcmp(rt->info, value))
1380 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').",
1381 line, path, filename, rt->name, key, rt->info, value, value);
1385 rt->info = strdupz(value);
1386 strip_quotes(rt->info);
1389 error("Health configuration at line %zu of file '%s/%s' for template '%s' has unknown key '%s'.",
1390 line, path, filename, rt->name, key);
1394 error("Health configuration at line %zu of file '%s/%s' has unknown key '%s'. Expected either '" HEALTH_ALARM_KEY "' or '" HEALTH_TEMPLATE_KEY "'.",
1395 line, path, filename, key);
1399 if(rc && !rrdcalc_add_alarm_from_config(&localhost, rc))
1400 rrdcalc_free(&localhost, rc);
1402 if(rt && !rrdcalctemplate_add_template_from_config(&localhost, rt))
1403 rrdcalctemplate_free(&localhost, rt);
1409 void health_readdir(const char *path) {
1410 size_t pathlen = strlen(path);
1412 debug(D_HEALTH, "Health configuration reading directory '%s'", path);
1414 DIR *dir = opendir(path);
1416 error("Health configuration cannot open directory '%s'.", path);
1420 struct dirent *de = NULL;
1421 while ((de = readdir(dir))) {
1422 size_t len = strlen(de->d_name);
1424 if(de->d_type == DT_DIR
1426 (de->d_name[0] == '.' && de->d_name[1] == '\0')
1427 || (de->d_name[0] == '.' && de->d_name[1] == '.' && de->d_name[2] == '\0')
1431 else if(de->d_type == DT_DIR) {
1432 char *s = mallocz(pathlen + strlen(de->d_name) + 2);
1435 strcat(s, de->d_name);
1441 else if((de->d_type == DT_LNK || de->d_type == DT_REG) &&
1442 len > 5 && !strcmp(&de->d_name[len - 5], ".conf")) {
1443 health_readfile(path, de->d_name);
1450 static inline char *health_config_dir(void) {
1451 char buffer[FILENAME_MAX + 1];
1452 snprintfz(buffer, FILENAME_MAX, "%s/health.d", config_get("global", "config directory", CONFIG_DIR));
1453 return config_get("health", "health configuration directory", buffer);
1456 void health_init(void) {
1457 debug(D_HEALTH, "Health configuration initializing");
1459 if(!(health_enabled = config_get_boolean("health", "enabled", 1))) {
1460 debug(D_HEALTH, "Health is disabled.");
1464 char *path = health_config_dir();
1467 char buffer[FILENAME_MAX + 1];
1468 snprintfz(buffer, FILENAME_MAX, "%s/alarm-email.sh", config_get("global", "plugins directory", PLUGINS_DIR));
1469 health_default_exec = config_get("health", "script to execute on alarm", buffer);
1472 long n = config_get_number("health", "in memory max health log entries", (long)localhost.health_log.max);
1474 error("Health configuration has invalid max log entries %ld. Using default %u", n, localhost.health_log.max);
1475 config_set_number("health", "in memory max health log entries", (long)localhost.health_log.max);
1477 else localhost.health_log.max = (unsigned int)n;
1479 rrdhost_rwlock(&localhost);
1480 health_readdir(path);
1481 rrdhost_unlock(&localhost);
1484 // ----------------------------------------------------------------------------
1487 static inline void health_string2json(BUFFER *wb, const char *prefix, const char *label, const char *value, const char *suffix) {
1489 buffer_sprintf(wb, "%s\"%s\":\"%s\"%s", prefix, label, value, suffix);
1491 buffer_sprintf(wb, "%s\"%s\":null%s", prefix, label, suffix);
1494 static inline void health_alarm_entry2json_nolock(BUFFER *wb, ALARM_ENTRY *ae) {
1495 buffer_sprintf(wb, "\n\t{\n"
1497 "\t\t\"name\":\"%s\",\n"
1498 "\t\t\"chart\":\"%s\",\n"
1499 "\t\t\"family\":\"%s\",\n"
1500 "\t\t\"processed\":%s,\n"
1501 "\t\t\"updated\":%s,\n"
1502 "\t\t\"exec_run\":%s,\n"
1503 "\t\t\"exec_failed\":%s,\n"
1504 "\t\t\"exec\":\"%s\",\n"
1505 "\t\t\"exec_code\":%d,\n"
1506 "\t\t\"source\":\"%s\",\n"
1507 "\t\t\"units\":\"%s\",\n"
1508 "\t\t\"info\":\"%s\",\n"
1509 "\t\t\"when\":%lu,\n"
1510 "\t\t\"duration\":%lu,\n"
1511 "\t\t\"non_clear_duration\":%lu,\n"
1512 "\t\t\"status\":\"%s\",\n"
1513 "\t\t\"old_status\":\"%s\",\n",
1518 (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_PROCESSED)?"true":"false",
1519 (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_UPDATED)?"true":"false",
1520 (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_EXEC_RUN)?"true":"false",
1521 (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_EXEC_FAILED)?"true":"false",
1522 ae->exec?ae->exec:health_default_exec,
1525 ae->units?ae->units:"",
1526 ae->info?ae->info:"",
1527 (unsigned long)ae->when,
1528 (unsigned long)ae->duration,
1529 (unsigned long)ae->non_clear_duration,
1530 rrdcalc_status2string(ae->new_status),
1531 rrdcalc_status2string(ae->old_status)
1534 buffer_strcat(wb, "\t\t\"value\":");
1535 buffer_rrd_value(wb, ae->new_value);
1536 buffer_strcat(wb, ",\n");
1538 buffer_strcat(wb, "\t\t\"old_value\":");
1539 buffer_rrd_value(wb, ae->old_value);
1540 buffer_strcat(wb, "\n");
1542 buffer_strcat(wb, "\t}");
1545 void health_alarm_log2json(RRDHOST *host, BUFFER *wb) {
1546 pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
1548 buffer_strcat(wb, "[");
1550 unsigned int max = host->health_log.max;
1551 unsigned int count = 0;
1553 for(ae = host->health_log.alarms; ae && count < max ; count++, ae = ae->next) {
1554 if(likely(count)) buffer_strcat(wb, ",");
1555 health_alarm_entry2json_nolock(wb, ae);
1558 buffer_strcat(wb, "\n]\n");
1560 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1563 static inline void health_rrdcalc2json_nolock(BUFFER *wb, RRDCALC *rc) {
1565 "\t\t\"%s.%s\": {\n"
1566 "\t\t\t\"name\": \"%s\",\n"
1567 "\t\t\t\"chart\": \"%s\",\n"
1568 "\t\t\t\"family\": \"%s\",\n"
1569 "\t\t\t\"active\": %s,\n"
1570 "\t\t\t\"exec\": \"%s\",\n"
1571 "\t\t\t\"source\": \"%s\",\n"
1572 "\t\t\t\"units\": \"%s\",\n"
1573 "\t\t\t\"info\": \"%s\",\n"
1574 "\t\t\t\"status\": \"%s\",\n"
1575 "\t\t\t\"last_status_change\": %lu,\n"
1576 "\t\t\t\"last_updated\": %lu,\n"
1577 "\t\t\t\"next_update\": %lu,\n"
1578 "\t\t\t\"update_every\": %d,\n"
1579 , rc->chart, rc->name
1582 , (rc->rrdset && rc->rrdset->family)?rc->rrdset->family:""
1583 , (rc->rrdset)?"true":"false"
1584 , rc->exec?rc->exec:health_default_exec
1586 , rc->units?rc->units:""
1587 , rc->info?rc->info:""
1588 , rrdcalc_status2string(rc->status)
1589 , (unsigned long)rc->last_status_change
1590 , (unsigned long)rc->last_updated
1591 , (unsigned long)rc->next_update
1595 if(RRDCALC_HAS_DB_LOOKUP(rc)) {
1596 if(rc->dimensions && *rc->dimensions)
1597 health_string2json(wb, "\t\t\t", "lookup_dimensions", rc->dimensions, ",\n");
1600 "\t\t\t\"db_after\": %lu,\n"
1601 "\t\t\t\"db_before\": %lu,\n"
1602 "\t\t\t\"lookup_method\": \"%s\",\n"
1603 "\t\t\t\"lookup_after\": %d,\n"
1604 "\t\t\t\"lookup_before\": %d,\n"
1605 "\t\t\t\"lookup_options\": \"",
1606 (unsigned long) rc->db_after,
1607 (unsigned long) rc->db_before,
1608 group_method2string(rc->group),
1612 buffer_data_options2string(wb, rc->options);
1613 buffer_strcat(wb, "\",\n");
1616 if(rc->calculation) {
1617 health_string2json(wb, "\t\t\t", "calc", rc->calculation->source, ",\n");
1618 health_string2json(wb, "\t\t\t", "calc_parsed", rc->calculation->parsed_as, ",\n");
1622 health_string2json(wb, "\t\t\t", "warn", rc->warning->source, ",\n");
1623 health_string2json(wb, "\t\t\t", "warn_parsed", rc->warning->parsed_as, ",\n");
1627 health_string2json(wb, "\t\t\t", "crit", rc->critical->source, ",\n");
1628 health_string2json(wb, "\t\t\t", "crit_parsed", rc->critical->parsed_as, ",\n");
1631 buffer_strcat(wb, "\t\t\t\"green\":");
1632 buffer_rrd_value(wb, rc->green);
1633 buffer_strcat(wb, ",\n");
1635 buffer_strcat(wb, "\t\t\t\"red\":");
1636 buffer_rrd_value(wb, rc->red);
1637 buffer_strcat(wb, ",\n");
1639 buffer_strcat(wb, "\t\t\t\"value\":");
1640 buffer_rrd_value(wb, rc->value);
1641 buffer_strcat(wb, "\n");
1643 buffer_strcat(wb, "\t\t}");
1646 //void health_rrdcalctemplate2json_nolock(BUFFER *wb, RRDCALCTEMPLATE *rt) {
1650 void health_alarms2json(RRDHOST *host, BUFFER *wb, int all) {
1652 rrdhost_rdlock(&localhost);
1654 buffer_strcat(wb, "{\n\t\"alarms\": {\n");
1656 for(i = 0, rc = host->alarms; rc ; rc = rc->next) {
1657 if(unlikely(!rc->rrdset || !rc->rrdset->last_collected_time.tv_sec))
1660 if(likely(!all && !(rc->status == RRDCALC_STATUS_WARNING || rc->status == RRDCALC_STATUS_CRITICAL)))
1663 if(likely(i)) buffer_strcat(wb, ",\n");
1664 health_rrdcalc2json_nolock(wb, rc);
1668 // buffer_strcat(wb, "\n\t},\n\t\"templates\": {");
1670 // RRDCALCTEMPLATE *rt;
1671 // for(rt = host->templates; rt ; rt = rt->next)
1672 // health_rrdcalctemplate2json_nolock(wb, rt);
1674 buffer_sprintf(wb, "\n\t},\n\t\"now\": %lu\n}\n", (unsigned long)time(NULL));
1675 rrdhost_unlock(&localhost);
1679 // ----------------------------------------------------------------------------
1680 // re-load health configuration
1682 static inline void health_free_all_nolock(RRDHOST *host) {
1683 while(host->templates)
1684 rrdcalctemplate_free(host, host->templates);
1687 rrdcalc_free(host, host->alarms);
1690 void health_reload(void) {
1691 if(!health_enabled) {
1692 error("Health reload is requested, but health is not enabled.");
1696 char *path = health_config_dir();
1698 rrdhost_rwlock(&localhost);
1699 health_free_all_nolock(&localhost);
1700 rrdhost_unlock(&localhost);
1703 for(st = localhost.rrdset_root; st ; st = st->next) {
1708 rrdhost_rwlock(&localhost);
1709 health_readdir(path);
1710 rrdhost_unlock(&localhost);
1712 for(st = localhost.rrdset_root; st ; st = st->next) {
1713 rrdhost_rwlock(&localhost);
1715 rrdsetcalc_link_matching(st);
1716 rrdcalctemplate_link_matching(st);
1718 rrdhost_unlock(&localhost);
1723 // ----------------------------------------------------------------------------
1724 // health main thread and friends
1726 static inline int rrdcalc_value2status(calculated_number n) {
1727 if(isnan(n)) return RRDCALC_STATUS_UNDEFINED;
1728 if(n) return RRDCALC_STATUS_RAISED;
1729 return RRDCALC_STATUS_CLEAR;
1732 static inline void health_alarm_execute(ALARM_ENTRY *ae) {
1733 if(ae->old_status == RRDCALC_STATUS_UNINITIALIZED && ae->new_status == RRDCALC_STATUS_CLEAR)
1736 char buffer[FILENAME_MAX + 1];
1739 const char *exec = ae->exec;
1740 if(!exec) exec = health_default_exec;
1742 snprintfz(buffer, FILENAME_MAX, "exec %s '%s' '%s' '%s' '%s' '%s' '%0.0Lf' '%0.0Lf' '%s' '%u' '%u' '%s' '%s'",
1745 ae->chart?ae->chart:"NOCAHRT",
1746 ae->family?ae->family:"NOFAMILY",
1747 rrdcalc_status2string(ae->new_status),
1748 rrdcalc_status2string(ae->old_status),
1751 ae->source?ae->source:"UNKNOWN",
1752 (uint32_t)ae->duration,
1753 (uint32_t)ae->non_clear_duration,
1754 ae->units?ae->units:"",
1755 ae->info?ae->info:""
1758 ae->notifications |= HEALTH_ENTRY_NOTIFICATIONS_EXEC_RUN;
1760 debug(D_HEALTH, "executing command '%s'", buffer);
1761 FILE *fp = mypopen(buffer, &command_pid);
1763 error("HEALTH: Cannot popen(\"%s\", \"r\").", buffer);
1766 debug(D_HEALTH, "HEALTH reading from command");
1767 char *s = fgets(buffer, FILENAME_MAX, fp);
1769 debug(D_HEALTH, "HEALTH closing command");
1770 ae->exec_code = mypclose(fp, command_pid);
1771 debug(D_HEALTH, "done executing command - returned with code %d", ae->exec_code);
1773 if(ae->exec_code != 0)
1774 ae->notifications |= HEALTH_ENTRY_NOTIFICATIONS_EXEC_FAILED;
1777 static inline void health_process_notifications(ALARM_ENTRY *ae) {
1778 info("Health alarm '%s.%s' = %0.2Lf - changed status from %s to %s",
1779 ae->chart?ae->chart:"NOCHART", ae->name,
1781 rrdcalc_status2string(ae->old_status),
1782 rrdcalc_status2string(ae->new_status)
1785 health_alarm_execute(ae);
1788 static inline void health_alarm_log(RRDHOST *host, time_t when,
1789 const char *name, const char *chart, const char *family,
1790 const char *exec, time_t duration,
1791 calculated_number old_value, calculated_number new_value,
1792 int old_status, int new_status,
1797 ALARM_ENTRY *ae = callocz(1, sizeof(ALARM_ENTRY));
1798 ae->name = strdupz(name);
1799 ae->hash_name = simple_hash(ae->name);
1802 ae->chart = strdupz(chart);
1803 ae->hash_chart = simple_hash(ae->chart);
1807 ae->family = strdupz(family);
1809 if(exec) ae->exec = strdupz(exec);
1810 if(source) ae->source = strdupz(source);
1811 if(units) ae->units = strdupz(units);
1812 if(info) ae->info = strdupz(info);
1814 ae->id = host->health_log.nextid++;
1816 ae->old_value = old_value;
1817 ae->new_value = new_value;
1818 ae->old_status = old_status;
1819 ae->new_status = new_status;
1820 ae->duration = duration;
1822 if(ae->old_status == RRDCALC_STATUS_WARNING || ae->old_status == RRDCALC_STATUS_CRITICAL)
1823 ae->non_clear_duration += ae->duration;
1826 pthread_rwlock_wrlock(&host->health_log.alarm_log_rwlock);
1827 ae->next = host->health_log.alarms;
1828 host->health_log.alarms = ae;
1829 host->health_log.count++;
1830 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1832 // match previous alarms
1833 pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
1835 for(t = host->health_log.alarms ; t ; t = t->next) {
1837 t->hash_name == ae->hash_name &&
1838 t->hash_chart == ae->hash_chart &&
1839 !strcmp(t->name, ae->name) &&
1840 t->chart && ae->chart && !strcmp(t->chart, ae->chart)) {
1842 if(!(t->notifications & HEALTH_ENTRY_NOTIFICATIONS_UPDATED) && !t->updated_by) {
1843 t->notifications |= HEALTH_ENTRY_NOTIFICATIONS_UPDATED;
1846 if((t->new_status == RRDCALC_STATUS_WARNING || t->new_status == RRDCALC_STATUS_CRITICAL) &&
1847 (t->old_status == RRDCALC_STATUS_WARNING || t->old_status == RRDCALC_STATUS_CRITICAL))
1848 ae->non_clear_duration += t->non_clear_duration;
1851 // no need to continue
1856 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1859 static inline void health_alarm_log_process(RRDHOST *host) {
1860 static uint32_t last_processed = 0;
1863 pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
1865 for(ae = host->health_log.alarms; ae ;ae = ae->next) {
1866 if(last_processed >= ae->id) break;
1868 if(!(ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_PROCESSED) &&
1869 !(ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_UPDATED)) {
1870 ae->notifications |= HEALTH_ENTRY_NOTIFICATIONS_PROCESSED;
1871 health_process_notifications(ae);
1875 if(host->health_log.alarms)
1876 last_processed = host->health_log.alarms->id;
1878 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1880 if(host->health_log.count <= host->health_log.max)
1883 // cleanup excess entries in the log
1884 pthread_rwlock_wrlock(&host->health_log.alarm_log_rwlock);
1886 ALARM_ENTRY *last = NULL;
1887 unsigned int count = host->health_log.max;
1888 for(ae = host->health_log.alarms; ae && count ; count--, last = ae, ae = ae->next) ;
1890 if(ae && last && last->next == ae)
1896 ALARM_ENTRY *t = ae->next;
1910 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1913 static inline int rrdcalc_isrunnable(RRDCALC *rc, time_t now, time_t *next_run) {
1914 if (unlikely(!rc->rrdset)) {
1915 debug(D_HEALTH, "Health not running alarm '%s.%s'. It is not linked to a chart.", rc->chart?rc->chart:"NOCHART", rc->name);
1919 if (unlikely(!rc->rrdset->last_collected_time.tv_sec)) {
1920 debug(D_HEALTH, "Health not running alarm '%s.%s'. Chart is not yet collected.", rc->chart?rc->chart:"NOCHART", rc->name);
1924 if (unlikely(!rc->update_every)) {
1925 debug(D_HEALTH, "Health not running alarm '%s.%s'. It does not have an update frequency", rc->chart?rc->chart:"NOCHART", rc->name);
1929 if (unlikely(rc->next_update > now)) {
1930 if (unlikely(*next_run > rc->next_update))
1931 *next_run = rc->next_update;
1933 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));
1938 // we should check that the DB lookup is possible
1940 // - the duration of the chart includes the required timeframe
1941 // we SHOULD NOT check the dimensions - there might be alarms that refer non-existing dimensions (e.g. cpu steal)
1946 void *health_main(void *ptr) {
1949 info("HEALTH thread created with task id %d", gettid());
1951 if(pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, NULL) != 0)
1952 error("Cannot set pthread cancel type to DEFERRED.");
1954 if(pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL) != 0)
1955 error("Cannot set pthread cancel state to ENABLE.");
1957 int min_run_every = (int)config_get_number("health", "run at least every seconds", 10);
1958 if(min_run_every < 1) min_run_every = 1;
1960 BUFFER *wb = buffer_create(100);
1962 unsigned int loop = 0;
1963 while(health_enabled) {
1965 debug(D_HEALTH, "Health monitoring iteration no %u started", loop);
1967 int oldstate, runnable = 0;
1968 time_t now = time(NULL);
1969 time_t next_run = now + min_run_every;
1972 if (unlikely(pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate) != 0))
1973 error("Cannot set pthread cancel state to DISABLE.");
1975 rrdhost_rdlock(&localhost);
1977 // the first loop is to lookup values from the db
1978 for (rc = localhost.alarms; rc; rc = rc->next) {
1979 if (unlikely(!rrdcalc_isrunnable(rc, now, &next_run)))
1983 rc->old_value = rc->value;
1985 // 1. if there is database lookup, do it
1986 // 2. if there is calculation expression, run it
1988 if (unlikely(RRDCALC_HAS_DB_LOOKUP(rc))) {
1989 time_t old_db_timestamp = rc->db_before;
1990 int value_is_null = 0;
1992 int ret = rrd2value(rc->rrdset, wb, &rc->value,
1993 rc->dimensions, 1, rc->after, rc->before, rc->group,
1994 rc->options, &rc->db_after, &rc->db_before, &value_is_null);
1996 if (unlikely(ret != 200)) {
1997 // database lookup failed
2000 debug(D_HEALTH, "Health alarm '%s.%s': database lookup returned error %d", rc->chart?rc->chart:"NOCHART", rc->name, ret);
2002 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_ERROR))) {
2003 rc->rrdcalc_flags |= RRDCALC_FLAG_DB_ERROR;
2004 error("Health alarm '%s.%s': database lookup returned error %d", rc->chart?rc->chart:"NOCHART", rc->name, ret);
2007 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_ERROR))
2008 rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_ERROR;
2010 if (unlikely(old_db_timestamp == rc->db_before)) {
2011 // database is stale
2013 debug(D_HEALTH, "Health alarm '%s.%s': database is stale", rc->chart?rc->chart:"NOCHART", rc->name);
2015 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_STALE))) {
2016 rc->rrdcalc_flags |= RRDCALC_FLAG_DB_STALE;
2017 error("Health alarm '%s.%s': database is stale", rc->chart?rc->chart:"NOCHART", rc->name);
2020 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_STALE))
2021 rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_STALE;
2023 if (unlikely(value_is_null)) {
2024 // collected value is null
2028 debug(D_HEALTH, "Health alarm '%s.%s': database lookup returned empty value (possibly value is not collected yet)",
2029 rc->chart?rc->chart:"NOCHART", rc->name);
2031 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_NAN))) {
2032 rc->rrdcalc_flags |= RRDCALC_FLAG_DB_NAN;
2033 error("Health alarm '%s.%s': database lookup returned empty value (possibly value is not collected yet)",
2034 rc->chart?rc->chart:"NOCHART", rc->name);
2037 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_NAN))
2038 rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_NAN;
2040 debug(D_HEALTH, "Health alarm '%s.%s': database lookup gave value "
2041 CALCULATED_NUMBER_FORMAT, rc->chart?rc->chart:"NOCHART", rc->name, rc->value);
2044 if(unlikely(rc->calculation)) {
2045 if (unlikely(!expression_evaluate(rc->calculation))) {
2046 // calculation failed
2050 debug(D_HEALTH, "Health alarm '%s.%s': failed to evaluate calculation with error: %s",
2051 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->calculation->error_msg));
2053 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_CALC_ERROR))) {
2054 rc->rrdcalc_flags |= RRDCALC_FLAG_CALC_ERROR;
2055 error("Health alarm '%s.%s': failed to evaluate calculation with error: %s",
2056 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->calculation->error_msg));
2060 if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_CALC_ERROR))
2061 rc->rrdcalc_flags &= ~RRDCALC_FLAG_CALC_ERROR;
2063 debug(D_HEALTH, "Health alarm '%s.%s': calculation expression gave value "
2064 CALCULATED_NUMBER_FORMAT
2065 ": %s (source: %s)",
2066 rc->chart?rc->chart:"NOCHART", rc->name,
2067 rc->calculation->result,
2068 buffer_tostring(rc->calculation->error_msg),
2072 rc->value = rc->calculation->result;
2076 rrdhost_unlock(&localhost);
2079 rrdhost_rdlock(&localhost);
2081 for (rc = localhost.alarms; rc; rc = rc->next) {
2082 if (unlikely(!rrdcalc_isrunnable(rc, now, &next_run)))
2085 int warning_status = RRDCALC_STATUS_UNDEFINED;
2086 int critical_status = RRDCALC_STATUS_UNDEFINED;
2088 if(likely(rc->warning)) {
2089 if(unlikely(!expression_evaluate(rc->warning))) {
2090 // calculation failed
2092 debug(D_HEALTH, "Health alarm '%s.%s': warning expression failed with error: %s",
2093 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->warning->error_msg));
2095 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_WARN_ERROR))) {
2096 rc->rrdcalc_flags |= RRDCALC_FLAG_WARN_ERROR;
2097 error("Health alarm '%s.%s': warning expression failed with error: %s",
2098 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->warning->error_msg));
2102 if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_WARN_ERROR))
2103 rc->rrdcalc_flags &= ~RRDCALC_FLAG_WARN_ERROR;
2105 debug(D_HEALTH, "Health alarm '%s.%s': warning expression gave value "
2106 CALCULATED_NUMBER_FORMAT
2107 ": %s (source: %s)",
2108 rc->chart?rc->chart:"NOCHART", rc->name,
2109 rc->warning->result,
2110 buffer_tostring(rc->warning->error_msg),
2114 warning_status = rrdcalc_value2status(rc->warning->result);
2118 if(likely(rc->critical)) {
2119 if(unlikely(!expression_evaluate(rc->critical))) {
2120 // calculation failed
2122 debug(D_HEALTH, "Health alarm '%s.%s': critical expression failed with error: %s",
2123 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->critical->error_msg));
2125 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_CRIT_ERROR))) {
2126 rc->rrdcalc_flags |= RRDCALC_FLAG_CRIT_ERROR;
2127 error("Health alarm '%s.%s': critical expression failed with error: %s",
2128 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->critical->error_msg));
2132 if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_CRIT_ERROR))
2133 rc->rrdcalc_flags &= ~RRDCALC_FLAG_CRIT_ERROR;
2135 debug(D_HEALTH, "Health alarm '%s.%s': critical expression gave value "
2136 CALCULATED_NUMBER_FORMAT
2137 ": %s (source: %s)",
2138 rc->chart?rc->chart:"NOCHART", rc->name,
2139 rc->critical->result,
2140 buffer_tostring(rc->critical->error_msg),
2144 critical_status = rrdcalc_value2status(rc->critical->result);
2148 int status = RRDCALC_STATUS_UNDEFINED;
2150 switch(warning_status) {
2151 case RRDCALC_STATUS_CLEAR:
2152 status = RRDCALC_STATUS_CLEAR;
2155 case RRDCALC_STATUS_RAISED:
2156 status = RRDCALC_STATUS_WARNING;
2163 switch(critical_status) {
2164 case RRDCALC_STATUS_CLEAR:
2165 if(status == RRDCALC_STATUS_UNDEFINED)
2166 status = RRDCALC_STATUS_CLEAR;
2169 case RRDCALC_STATUS_RAISED:
2170 status = RRDCALC_STATUS_CRITICAL;
2177 if(status != rc->status) {
2178 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);
2179 rc->last_status_change = now;
2180 rc->status = status;
2183 rc->last_updated = now;
2184 rc->next_update = now + rc->update_every;
2186 if (next_run > rc->next_update)
2187 next_run = rc->next_update;
2190 rrdhost_unlock(&localhost);
2193 if (unlikely(pthread_setcancelstate(oldstate, NULL) != 0))
2194 error("Cannot set pthread cancel state to RESTORE (%d).", oldstate);
2196 // execute notifications
2198 health_alarm_log_process(&localhost);
2201 if(now < next_run) {
2202 debug(D_HEALTH, "Health monitoring iteration no %u done. Next iteration in %d secs",
2203 loop, (int) (next_run - now));
2204 sleep_usec(1000000 * (unsigned long long) (next_run - now));
2207 debug(D_HEALTH, "Health monitoring iteration no %u done. Next iteration now", loop);
2213 info("HEALTH thread exiting");