3 #define RRDVAR_MAX_LENGTH 1024
5 static const char *health_default_exec = PLUGINS_DIR "/alarm-notify.sh";
6 static const char *health_default_recipient = "root";
7 int health_enabled = 1;
9 // ----------------------------------------------------------------------------
10 // Health Alarms Log Management
12 static inline void health_alarm_log(RRDHOST *host,
13 uint32_t alarm_id, uint32_t alarm_event_id,
15 const char *name, const char *chart, const char *family,
16 const char *exec, const char *recipient, time_t duration,
17 calculated_number old_value, calculated_number new_value,
18 int old_status, int new_status,
24 debug(D_HEALTH, "Health adding alarm log entry with id: %u", host->health_log.next_log_id);
26 ALARM_ENTRY *ae = callocz(1, sizeof(ALARM_ENTRY));
27 ae->name = strdupz(name);
28 ae->hash_name = simple_hash(ae->name);
31 ae->chart = strdupz(chart);
32 ae->hash_chart = simple_hash(ae->chart);
36 ae->family = strdupz(family);
38 if(exec) ae->exec = strdupz(exec);
39 if(recipient) ae->recipient = strdupz(recipient);
40 if(source) ae->source = strdupz(source);
41 if(units) ae->units = strdupz(units);
42 if(info) ae->info = strdupz(info);
44 ae->unique_id = host->health_log.next_log_id++;
45 ae->alarm_id = alarm_id;
46 ae->alarm_event_id = alarm_event_id;
48 ae->old_value = old_value;
49 ae->new_value = new_value;
50 ae->old_status = old_status;
51 ae->new_status = new_status;
52 ae->duration = duration;
54 ae->delay_up_to_timestamp = when + delay;
56 if(ae->old_status == RRDCALC_STATUS_WARNING || ae->old_status == RRDCALC_STATUS_CRITICAL)
57 ae->non_clear_duration += ae->duration;
60 pthread_rwlock_wrlock(&host->health_log.alarm_log_rwlock);
61 ae->next = host->health_log.alarms;
62 host->health_log.alarms = ae;
63 host->health_log.count++;
64 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
66 // match previous alarms
67 pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
69 for(t = host->health_log.alarms ; t ; t = t->next) {
70 if(t != ae && t->alarm_id == ae->alarm_id) {
71 if(!(t->notifications & HEALTH_ENTRY_NOTIFICATIONS_UPDATED) && !t->updated_by) {
72 t->notifications |= HEALTH_ENTRY_NOTIFICATIONS_UPDATED;
75 if((t->new_status == RRDCALC_STATUS_WARNING || t->new_status == RRDCALC_STATUS_CRITICAL) &&
76 (t->old_status == RRDCALC_STATUS_WARNING || t->old_status == RRDCALC_STATUS_CRITICAL))
77 ae->non_clear_duration += t->non_clear_duration;
80 // no need to continue
85 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
88 // ----------------------------------------------------------------------------
91 static inline int rrdvar_fix_name(char *variable) {
94 if (!isalnum(*variable) && *variable != '.' && *variable != '_') {
105 int rrdvar_compare(void* a, void* b) {
106 if(((RRDVAR *)a)->hash < ((RRDVAR *)b)->hash) return -1;
107 else if(((RRDVAR *)a)->hash > ((RRDVAR *)b)->hash) return 1;
108 else return strcmp(((RRDVAR *)a)->name, ((RRDVAR *)b)->name);
111 static inline RRDVAR *rrdvar_index_add(avl_tree_lock *tree, RRDVAR *rv) {
112 RRDVAR *ret = (RRDVAR *)avl_insert_lock(tree, (avl *)(rv));
114 debug(D_VARIABLES, "Request to insert RRDVAR '%s' into index failed. Already exists.", rv->name);
119 static inline RRDVAR *rrdvar_index_del(avl_tree_lock *tree, RRDVAR *rv) {
120 RRDVAR *ret = (RRDVAR *)avl_remove_lock(tree, (avl *)(rv));
122 error("Request to remove RRDVAR '%s' from index failed. Not Found.", rv->name);
127 static inline RRDVAR *rrdvar_index_find(avl_tree_lock *tree, const char *name, uint32_t hash) {
129 tmp.name = (char *)name;
130 tmp.hash = (hash)?hash:simple_hash(tmp.name);
132 return (RRDVAR *)avl_search_lock(tree, (avl *)&tmp);
135 static inline void rrdvar_free(RRDHOST *host, avl_tree_lock *tree, RRDVAR *rv) {
141 rrdvar_index_del(tree, rv);
147 static inline RRDVAR *rrdvar_create_and_index(const char *scope, avl_tree_lock *tree, const char *name, int type, calculated_number *value) {
148 char *variable = strdupz(name);
149 rrdvar_fix_name(variable);
150 uint32_t hash = simple_hash(variable);
152 RRDVAR *rv = rrdvar_index_find(tree, variable, hash);
154 debug(D_VARIABLES, "Variable '%s' not found in scope '%s'. Creating a new one.", variable, scope);
156 rv = callocz(1, sizeof(RRDVAR));
162 RRDVAR *ret = rrdvar_index_add(tree, rv);
163 if(unlikely(ret != rv)) {
164 debug(D_VARIABLES, "Variable '%s' in scope '%s' already exists", variable, scope);
165 rrdvar_free(NULL, NULL, rv);
169 debug(D_VARIABLES, "Variable '%s' created in scope '%s'", variable, scope);
180 // ----------------------------------------------------------------------------
183 calculated_number rrdvar2number(RRDVAR *rv) {
185 case RRDVAR_TYPE_CALCULATED: {
186 calculated_number *n = (calculated_number *)rv->value;
190 case RRDVAR_TYPE_TIME_T: {
191 time_t *n = (time_t *)rv->value;
195 case RRDVAR_TYPE_COLLECTED: {
196 collected_number *n = (collected_number *)rv->value;
200 case RRDVAR_TYPE_TOTAL: {
201 total_number *n = (total_number *)rv->value;
205 case RRDVAR_TYPE_INT: {
206 int *n = (int *)rv->value;
211 error("I don't know how to convert RRDVAR type %d to calculated_number", rv->type);
216 void dump_variable(void *data) {
217 RRDVAR *rv = (RRDVAR *)data;
218 debug(D_HEALTH, "%50s : %20.5Lf", rv->name, rrdvar2number(rv));
221 int health_variable_lookup(const char *variable, uint32_t hash, RRDCALC *rc, calculated_number *result) {
222 RRDSET *st = rc->rrdset;
227 rv = rrdvar_index_find(&st->variables_root_index, variable, hash);
229 *result = rrdvar2number(rv);
233 rv = rrdvar_index_find(&st->rrdfamily->variables_root_index, variable, hash);
235 *result = rrdvar2number(rv);
239 rv = rrdvar_index_find(&st->rrdhost->variables_root_index, variable, hash);
241 *result = rrdvar2number(rv);
245 debug(D_HEALTH, "Available local chart '%s' variables:", st->id);
246 avl_traverse_lock(&st->variables_root_index, dump_variable);
248 debug(D_HEALTH, "Available family '%s' variables:", st->rrdfamily->family);
249 avl_traverse_lock(&st->rrdfamily->variables_root_index, dump_variable);
251 debug(D_HEALTH, "Available host '%s' variables:", st->rrdhost->hostname);
252 avl_traverse_lock(&st->rrdhost->variables_root_index, dump_variable);
257 // ----------------------------------------------------------------------------
258 // RRDSETVAR management
260 RRDSETVAR *rrdsetvar_create(RRDSET *st, const char *variable, int type, void *value, uint32_t options) {
261 debug(D_VARIABLES, "RRDVARSET create for chart id '%s' name '%s' with variable name '%s'", st->id, st->name, variable);
262 RRDSETVAR *rs = (RRDSETVAR *)callocz(1, sizeof(RRDSETVAR));
264 char buffer[RRDVAR_MAX_LENGTH + 1];
265 snprintfz(buffer, RRDVAR_MAX_LENGTH, "%s.%s", st->id, variable);
266 rs->fullid = strdupz(buffer);
268 snprintfz(buffer, RRDVAR_MAX_LENGTH, "%s.%s", st->name, variable);
269 rs->fullname = strdupz(buffer);
271 rs->variable = strdupz(variable);
275 rs->options = options;
278 rs->local = rrdvar_create_and_index("local", &st->variables_root_index, rs->variable, rs->type, rs->value);
279 rs->family = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->fullid, rs->type, rs->value);
280 rs->host = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullid, rs->type, rs->value);
281 rs->family_name = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->fullname, rs->type, rs->value);
282 rs->host_name = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullname, rs->type, rs->value);
284 rs->next = st->variables;
290 void rrdsetvar_rename_all(RRDSET *st) {
291 debug(D_VARIABLES, "RRDSETVAR rename for chart id '%s' name '%s'", st->id, st->name);
293 // only these 2 can change name
297 char buffer[RRDVAR_MAX_LENGTH + 1];
298 RRDSETVAR *rs, *next = st->variables;
302 snprintfz(buffer, RRDVAR_MAX_LENGTH, "%s.%s", st->name, rs->variable);
304 if (strcmp(buffer, rs->fullname)) {
306 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family_name);
307 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_name);
310 rs->fullname = strdupz(st->name);
311 rs->family_name = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->fullname, rs->type, rs->value);
312 rs->host_name = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullname, rs->type, rs->value);
316 rrdsetcalc_link_matching(st);
319 void rrdsetvar_free(RRDSETVAR *rs) {
320 RRDSET *st = rs->rrdset;
321 debug(D_VARIABLES, "RRDSETVAR free for chart id '%s' name '%s', variable '%s'", st->id, st->name, rs->variable);
323 rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local);
324 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family);
325 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host);
326 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family_name);
327 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_name);
329 if(st->variables == rs) {
330 st->variables = rs->next;
334 for (t = st->variables; t && t->next != rs; t = t->next);
335 if(!t) error("RRDSETVAR '%s' not found in chart '%s' variables linked list", rs->fullname, st->id);
336 else t->next = rs->next;
345 // ----------------------------------------------------------------------------
346 // RRDDIMVAR management
348 #define RRDDIMVAR_ID_MAX 1024
350 RRDDIMVAR *rrddimvar_create(RRDDIM *rd, int type, const char *prefix, const char *suffix, void *value, uint32_t options) {
351 RRDSET *st = rd->rrdset;
353 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:"");
355 if(!prefix) prefix = "";
356 if(!suffix) suffix = "";
358 char buffer[RRDDIMVAR_ID_MAX + 1];
359 RRDDIMVAR *rs = (RRDDIMVAR *)callocz(1, sizeof(RRDDIMVAR));
361 rs->prefix = strdupz(prefix);
362 rs->suffix = strdupz(suffix);
364 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s%s%s", rs->prefix, rd->id, rs->suffix);
365 rs->id = strdupz(buffer);
367 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s%s%s", rs->prefix, rd->name, rs->suffix);
368 rs->name = strdupz(buffer);
370 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->id, rs->id);
371 rs->fullidid = strdupz(buffer);
373 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->id, rs->name);
374 rs->fullidname = strdupz(buffer);
376 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->name, rs->id);
377 rs->fullnameid = strdupz(buffer);
379 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->name, rs->name);
380 rs->fullnamename = strdupz(buffer);
384 rs->options = options;
387 rs->local_id = rrdvar_create_and_index("local", &st->variables_root_index, rs->id, rs->type, rs->value);
388 rs->local_name = rrdvar_create_and_index("local", &st->variables_root_index, rs->name, rs->type, rs->value);
390 rs->family_id = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->id, rs->type, rs->value);
391 rs->family_name = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->name, rs->type, rs->value);
393 rs->host_fullidid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullidid, rs->type, rs->value);
394 rs->host_fullidname = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullidname, rs->type, rs->value);
395 rs->host_fullnameid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullnameid, rs->type, rs->value);
396 rs->host_fullnamename = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullnamename, rs->type, rs->value);
398 rs->next = rd->variables;
404 void rrddimvar_rename_all(RRDDIM *rd) {
405 RRDSET *st = rd->rrdset;
406 debug(D_VARIABLES, "RRDDIMSET rename for chart id '%s' name '%s', dimension id '%s', name '%s'", st->id, st->name, rd->id, rd->name);
408 RRDDIMVAR *rs, *next = rd->variables;
412 if (strcmp(rd->name, rs->name)) {
413 char buffer[RRDDIMVAR_ID_MAX + 1];
417 rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local_name);
419 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s%s%s", rs->prefix, rd->name, rs->suffix);
420 rs->name = strdupz(buffer);
421 rs->local_name = rrdvar_create_and_index("local", &st->variables_root_index, rs->name, rs->type, rs->value);
423 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullidname);
424 freez(rs->fullidname);
425 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->id, rs->name);
426 rs->fullidname = strdupz(buffer);
427 rs->host_fullidname = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index,
428 rs->fullidname, rs->type, rs->value);
431 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnameid);
432 freez(rs->fullnameid);
433 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->name, rs->id);
434 rs->fullnameid = strdupz(buffer);
435 rs->host_fullnameid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index,
436 rs->fullnameid, rs->type, rs->value);
439 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnamename);
440 freez(rs->fullnamename);
441 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->name, rs->name);
442 rs->fullnamename = strdupz(buffer);
443 rs->host_fullnamename = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index,
444 rs->fullnamename, rs->type, rs->value);
449 void rrddimvar_free(RRDDIMVAR *rs) {
450 RRDDIM *rd = rs->rrddim;
451 RRDSET *st = rd->rrdset;
452 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);
454 rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local_id);
455 rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local_name);
457 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family_id);
458 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family_name);
460 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullidid);
461 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullidname);
462 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnameid);
463 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnamename);
465 if(rd->variables == rs) {
466 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);
467 rd->variables = rs->next;
470 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);
472 for (t = rd->variables; t && t->next != rs; t = t->next) ;
473 if(!t) error("RRDDIMVAR '%s' not found in dimension '%s/%s' variables linked list", rs->name, st->id, rd->id);
474 else t->next = rs->next;
482 freez(rs->fullidname);
483 freez(rs->fullnameid);
484 freez(rs->fullnamename);
488 // ----------------------------------------------------------------------------
489 // RRDCALC management
491 static inline const char *rrdcalc_status2string(int status) {
493 case RRDCALC_STATUS_REMOVED:
496 case RRDCALC_STATUS_UNDEFINED:
499 case RRDCALC_STATUS_UNINITIALIZED:
500 return "UNINITIALIZED";
502 case RRDCALC_STATUS_CLEAR:
505 case RRDCALC_STATUS_RAISED:
508 case RRDCALC_STATUS_WARNING:
511 case RRDCALC_STATUS_CRITICAL:
515 error("Unknown alarm status %d", status);
520 static void rrdsetcalc_link(RRDSET *st, RRDCALC *rc) {
521 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);
523 rc->last_status_change = time(NULL);
526 rc->rrdset_next = st->alarms;
527 rc->rrdset_prev = NULL;
530 rc->rrdset_next->rrdset_prev = rc;
534 if(rc->update_every < rc->rrdset->update_every) {
535 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);
536 rc->update_every = rc->rrdset->update_every;
539 if(!isnan(rc->green) && isnan(st->green)) {
540 debug(D_HEALTH, "Health alarm '%s.%s' green threshold set from %Lf to %Lf.", rc->rrdset->id, rc->name, rc->rrdset->green, rc->green);
541 st->green = rc->green;
544 if(!isnan(rc->red) && isnan(st->red)) {
545 debug(D_HEALTH, "Health alarm '%s.%s' red threshold set from %Lf to %Lf.", rc->rrdset->id, rc->name, rc->rrdset->red, rc->red);
549 rc->local = rrdvar_create_and_index("local", &st->variables_root_index, rc->name, RRDVAR_TYPE_CALCULATED, &rc->value);
550 rc->family = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rc->name, RRDVAR_TYPE_CALCULATED, &rc->value);
552 char fullname[RRDVAR_MAX_LENGTH + 1];
553 snprintfz(fullname, RRDVAR_MAX_LENGTH, "%s.%s", st->id, rc->name);
554 rc->hostid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, fullname, RRDVAR_TYPE_CALCULATED, &rc->value);
556 snprintfz(fullname, RRDVAR_MAX_LENGTH, "%s.%s", st->name, rc->name);
557 rc->hostname = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, fullname, RRDVAR_TYPE_CALCULATED, &rc->value);
559 if(!rc->units) rc->units = strdupz(st->units);
562 time_t now = time(NULL);
563 health_alarm_log(st->rrdhost, rc->id, rc->next_event_id++, now, rc->name, rc->rrdset->id, rc->rrdset->family, rc->exec, rc->recipient, now - rc->last_status_change, rc->old_value, rc->value, rc->status, RRDCALC_STATUS_UNINITIALIZED, rc->source, rc->units, rc->info, 0);
567 static inline int rrdcalc_is_matching_this_rrdset(RRDCALC *rc, RRDSET *st) {
568 if( (rc->hash_chart == st->hash && !strcmp(rc->chart, st->id)) ||
569 (rc->hash_chart == st->hash_name && !strcmp(rc->chart, st->name)))
575 // this has to be called while the RRDHOST is locked
576 inline void rrdsetcalc_link_matching(RRDSET *st) {
577 // debug(D_HEALTH, "find matching alarms for chart '%s'", st->id);
580 for(rc = st->rrdhost->alarms; rc ; rc = rc->next) {
581 if(unlikely(rc->rrdset))
584 if(unlikely(rrdcalc_is_matching_this_rrdset(rc, st)))
585 rrdsetcalc_link(st, rc);
589 // this has to be called while the RRDHOST is locked
590 inline void rrdsetcalc_unlink(RRDCALC *rc) {
591 RRDSET *st = rc->rrdset;
594 debug(D_HEALTH, "Requested to unlink RRDCALC '%s.%s' which is not linked to any RRDSET", rc->chart?rc->chart:"NOCHART", rc->name);
595 error("Requested to unlink RRDCALC '%s.%s' which is not linked to any RRDSET", rc->chart?rc->chart:"NOCHART", rc->name);
600 time_t now = time(NULL);
601 health_alarm_log(st->rrdhost, rc->id, rc->next_event_id++, now, rc->name, rc->rrdset->id, rc->rrdset->family, rc->exec, rc->recipient, now - rc->last_status_change, rc->old_value, rc->value, rc->status, RRDCALC_STATUS_REMOVED, rc->source, rc->units, rc->info, 0);
604 RRDHOST *host = st->rrdhost;
606 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);
610 rc->rrdset_prev->rrdset_next = rc->rrdset_next;
613 rc->rrdset_next->rrdset_prev = rc->rrdset_prev;
616 st->alarms = rc->rrdset_next;
618 rc->rrdset_prev = rc->rrdset_next = NULL;
620 rrdvar_free(st->rrdhost, &st->variables_root_index, rc->local);
623 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rc->family);
626 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rc->hostid);
629 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rc->hostname);
634 // RRDCALC will remain in RRDHOST
635 // so that if the matching chart is found in the future
636 // it will be applied automatically
639 RRDCALC *rrdcalc_find(RRDSET *st, const char *name) {
641 uint32_t hash = simple_hash(name);
643 for( rc = st->alarms; rc ; rc = rc->rrdset_next ) {
644 if(unlikely(rc->hash == hash && !strcmp(rc->name, name)))
651 static inline int rrdcalc_exists(RRDHOST *host, const char *chart, const char *name, uint32_t hash_chart, uint32_t hash_name) {
654 if(unlikely(!chart)) {
655 error("attempt to find RRDCALC '%s' without giving a chart name", name);
659 if(unlikely(!hash_chart)) hash_chart = simple_hash(chart);
660 if(unlikely(!hash_name)) hash_name = simple_hash(name);
662 // make sure it does not already exist
663 for(rc = host->alarms; rc ; rc = rc->next) {
664 if (unlikely(rc->chart && rc->hash == hash_name && rc->hash_chart == hash_chart && !strcmp(name, rc->name) && !strcmp(chart, rc->chart))) {
665 debug(D_HEALTH, "Health alarm '%s.%s' already exists in host '%s'.", chart, name, host->hostname);
666 error("Health alarm '%s.%s' already exists in host '%s'.", chart, name, host->hostname);
674 static inline uint32_t rrdcalc_get_unique_id(RRDHOST *host, const char *chart, const char *name, uint32_t *next_event_id) {
676 uint32_t hash_chart = simple_hash(chart);
677 uint32_t hash_name = simple_hash(name);
679 // re-use old IDs, by looking them up in the alarm log
681 for(ae = host->health_log.alarms; ae ;ae = ae->next) {
682 if(unlikely(ae->hash_name == hash_name && ae->hash_chart == hash_chart && !strcmp(name, ae->name) && !strcmp(chart, ae->chart))) {
683 if(next_event_id) *next_event_id = ae->alarm_event_id + 1;
689 return host->health_log.next_alarm_id++;
692 static inline void rrdcalc_create_part2(RRDHOST *host, RRDCALC *rc) {
693 rrdhost_check_rdlock(host);
695 if(rc->calculation) {
696 rc->calculation->this = &rc->value;
697 rc->calculation->after = &rc->db_after;
698 rc->calculation->before = &rc->db_before;
699 rc->calculation->rrdcalc = rc;
703 rc->warning->this = &rc->value;
704 rc->warning->after = &rc->db_after;
705 rc->warning->before = &rc->db_before;
706 rc->warning->rrdcalc = rc;
710 rc->critical->this = &rc->value;
711 rc->critical->after = &rc->db_after;
712 rc->critical->before = &rc->db_before;
713 rc->critical->rrdcalc = rc;
716 // link it to the host
717 if(likely(host->alarms)) {
720 for(t = host->alarms; t && t->next ; t = t->next) ;
727 // link it to its chart
729 for(st = host->rrdset_root; st ; st = st->next) {
730 if(rrdcalc_is_matching_this_rrdset(rc, st)) {
731 rrdsetcalc_link(st, rc);
737 static inline RRDCALC *rrdcalc_create(RRDHOST *host, RRDCALCTEMPLATE *rt, const char *chart) {
739 debug(D_HEALTH, "Health creating dynamic alarm (from template) '%s.%s'", chart, rt->name);
741 if(rrdcalc_exists(host, chart, rt->name, 0, 0))
744 RRDCALC *rc = callocz(1, sizeof(RRDCALC));
745 rc->next_event_id = 1;
746 rc->id = rrdcalc_get_unique_id(host, chart, rt->name, &rc->next_event_id);
747 rc->name = strdupz(rt->name);
748 rc->hash = simple_hash(rc->name);
749 rc->chart = strdupz(chart);
750 rc->hash_chart = simple_hash(rc->chart);
752 if(rt->dimensions) rc->dimensions = strdupz(rt->dimensions);
754 rc->green = rt->green;
759 rc->delay_up_duration = rt->delay_up_duration;
760 rc->delay_down_duration = rt->delay_down_duration;
761 rc->delay_max_duration = rt->delay_max_duration;
762 rc->delay_multiplier = rt->delay_multiplier;
764 rc->group = rt->group;
765 rc->after = rt->after;
766 rc->before = rt->before;
767 rc->update_every = rt->update_every;
768 rc->options = rt->options;
770 if(rt->exec) rc->exec = strdupz(rt->exec);
771 if(rt->recipient) rc->recipient = strdupz(rt->recipient);
772 if(rt->source) rc->source = strdupz(rt->source);
773 if(rt->units) rc->units = strdupz(rt->units);
774 if(rt->info) rc->info = strdupz(rt->info);
776 if(rt->calculation) {
777 rc->calculation = expression_parse(rt->calculation->source, NULL, NULL);
779 error("Health alarm '%s.%s': failed to parse calculation expression '%s'", chart, rt->name, rt->calculation->source);
782 rc->warning = expression_parse(rt->warning->source, NULL, NULL);
784 error("Health alarm '%s.%s': failed to re-parse warning expression '%s'", chart, rt->name, rt->warning->source);
787 rc->critical = expression_parse(rt->critical->source, NULL, NULL);
789 error("Health alarm '%s.%s': failed to re-parse critical expression '%s'", chart, rt->name, rt->critical->source);
792 debug(D_HEALTH, "Health runtime added alarm '%s.%s': exec '%s', recipient '%s', green %Lf, red %Lf, lookup: group %d, after %d, before %d, options %u, dimensions '%s', update every %d, calculation '%s', warning '%s', critical '%s', source '%s', delay up %d, delay down %d, delay max %d, delay_multiplier %f",
793 (rc->chart)?rc->chart:"NOCHART",
795 (rc->exec)?rc->exec:"DEFAULT",
796 (rc->recipient)?rc->recipient:"DEFAULT",
803 (rc->dimensions)?rc->dimensions:"NONE",
805 (rc->calculation)?rc->calculation->parsed_as:"NONE",
806 (rc->warning)?rc->warning->parsed_as:"NONE",
807 (rc->critical)?rc->critical->parsed_as:"NONE",
809 rc->delay_up_duration,
810 rc->delay_down_duration,
811 rc->delay_max_duration,
815 rrdcalc_create_part2(host, rc);
819 void rrdcalc_free(RRDHOST *host, RRDCALC *rc) {
822 debug(D_HEALTH, "Health removing alarm '%s.%s' of host '%s'", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);
824 // unlink it from RRDSET
825 if(rc->rrdset) rrdsetcalc_unlink(rc);
827 // unlink it from RRDHOST
828 if(unlikely(rc == host->alarms))
829 host->alarms = rc->next;
831 else if(likely(host->alarms)) {
832 RRDCALC *t, *last = host->alarms;
833 for(t = last->next; t && t != rc; last = t, t = t->next) ;
834 if(last && last->next == rc)
835 last->next = rc->next;
837 error("Cannot unlink alarm '%s.%s' from host '%s': not found", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);
840 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);
842 expression_free(rc->calculation);
843 expression_free(rc->warning);
844 expression_free(rc->critical);
849 freez(rc->dimensions);
851 freez(rc->recipient);
858 // ----------------------------------------------------------------------------
859 // RRDCALCTEMPLATE management
861 void rrdcalctemplate_link_matching(RRDSET *st) {
864 for(rt = st->rrdhost->templates; rt ; rt = rt->next) {
865 if(rt->hash_context == st->hash_context && !strcmp(rt->context, st->context)) {
866 RRDCALC *rc = rrdcalc_create(st->rrdhost, rt, st->id);
868 error("Health tried to create alarm from template '%s', but it failed", rt->name);
870 #ifdef NETDATA_INTERNAL_CHECKS
871 else if(rc->rrdset != st)
872 error("Health alarm '%s.%s' should be linked to chart '%s', but it is not", rc->chart?rc->chart:"NOCHART", rc->name, st->id);
878 static inline void rrdcalctemplate_free(RRDHOST *host, RRDCALCTEMPLATE *rt) {
879 debug(D_HEALTH, "Health removing template '%s' of host '%s'", rt->name, host->hostname);
881 if(host->templates) {
882 if(host->templates == rt) {
883 host->templates = rt->next;
886 RRDCALCTEMPLATE *t, *last = host->templates;
887 for (t = last->next; t && t != rt; last = t, t = t->next ) ;
888 if(last && last->next == rt) {
889 last->next = rt->next;
893 error("Cannot find RRDCALCTEMPLATE '%s' linked in host '%s'", rt->name, host->hostname);
897 expression_free(rt->calculation);
898 expression_free(rt->warning);
899 expression_free(rt->critical);
903 freez(rt->recipient);
908 freez(rt->dimensions);
912 // ----------------------------------------------------------------------------
913 // load health configuration
915 #define HEALTH_CONF_MAX_LINE 4096
917 #define HEALTH_ALARM_KEY "alarm"
918 #define HEALTH_TEMPLATE_KEY "template"
919 #define HEALTH_ON_KEY "on"
920 #define HEALTH_LOOKUP_KEY "lookup"
921 #define HEALTH_CALC_KEY "calc"
922 #define HEALTH_EVERY_KEY "every"
923 #define HEALTH_GREEN_KEY "green"
924 #define HEALTH_RED_KEY "red"
925 #define HEALTH_WARN_KEY "warn"
926 #define HEALTH_CRIT_KEY "crit"
927 #define HEALTH_EXEC_KEY "exec"
928 #define HEALTH_RECIPIENT_KEY "to"
929 #define HEALTH_UNITS_KEY "units"
930 #define HEALTH_INFO_KEY "info"
931 #define HEALTH_DELAY_KEY "delay"
933 static inline int rrdcalc_add_alarm_from_config(RRDHOST *host, RRDCALC *rc) {
935 error("Health configuration for alarm '%s' does not have a chart", rc->name);
939 if(!rc->update_every) {
940 error("Health configuration for alarm '%s.%s' has no frequency (parameter 'every'). Ignoring it.", rc->chart?rc->chart:"NOCHART", rc->name);
944 if(!RRDCALC_HAS_DB_LOOKUP(rc) && !rc->warning && !rc->critical) {
945 error("Health configuration for alarm '%s.%s' is useless (no calculation, no warning and no critical evaluation)", rc->chart?rc->chart:"NOCHART", rc->name);
949 if (rrdcalc_exists(host, rc->chart, rc->name, rc->hash_chart, rc->hash))
952 rc->id = rrdcalc_get_unique_id(&localhost, rc->chart, rc->name, &rc->next_event_id);
954 debug(D_HEALTH, "Health configuration adding alarm '%s.%s' (%u): exec '%s', recipient '%s', green %Lf, red %Lf, lookup: group %d, after %d, before %d, options %u, dimensions '%s', update every %d, calculation '%s', warning '%s', critical '%s', source '%s', delay up %d, delay down %d, delay max %d, delay_multiplier %f",
955 rc->chart?rc->chart:"NOCHART",
958 (rc->exec)?rc->exec:"DEFAULT",
959 (rc->recipient)?rc->recipient:"DEFAULT",
966 (rc->dimensions)?rc->dimensions:"NONE",
968 (rc->calculation)?rc->calculation->parsed_as:"NONE",
969 (rc->warning)?rc->warning->parsed_as:"NONE",
970 (rc->critical)?rc->critical->parsed_as:"NONE",
972 rc->delay_up_duration,
973 rc->delay_down_duration,
974 rc->delay_max_duration,
978 rrdcalc_create_part2(host, rc);
982 static inline int rrdcalctemplate_add_template_from_config(RRDHOST *host, RRDCALCTEMPLATE *rt) {
983 if(unlikely(!rt->context)) {
984 error("Health configuration for template '%s' does not have a context", rt->name);
988 if(unlikely(!rt->update_every)) {
989 error("Health configuration for template '%s' has no frequency (parameter 'every'). Ignoring it.", rt->name);
993 if(unlikely(!RRDCALCTEMPLATE_HAS_CALCULATION(rt) && !rt->warning && !rt->critical)) {
994 error("Health configuration for template '%s' is useless (no calculation, no warning and no critical evaluation)", rt->name);
998 RRDCALCTEMPLATE *t, *last = NULL;
999 for (t = host->templates; t ; last = t, t = t->next) {
1000 if(unlikely(t->hash_name == rt->hash_name && !strcmp(t->name, rt->name))) {
1001 error("Health configuration template '%s' already exists for host '%s'.", rt->name, host->hostname);
1006 debug(D_HEALTH, "Health configuration adding template '%s': context '%s', exec '%s', recipient '%s', green %Lf, red %Lf, lookup: group %d, after %d, before %d, options %u, dimensions '%s', update every %d, calculation '%s', warning '%s', critical '%s', source '%s', delay up %d, delay down %d, delay max %d, delay_multiplier %f",
1008 (rt->context)?rt->context:"NONE",
1009 (rt->exec)?rt->exec:"DEFAULT",
1010 (rt->recipient)?rt->recipient:"DEFAULT",
1017 (rt->dimensions)?rt->dimensions:"NONE",
1019 (rt->calculation)?rt->calculation->parsed_as:"NONE",
1020 (rt->warning)?rt->warning->parsed_as:"NONE",
1021 (rt->critical)?rt->critical->parsed_as:"NONE",
1023 rt->delay_up_duration,
1024 rt->delay_down_duration,
1025 rt->delay_max_duration,
1026 rt->delay_multiplier
1033 rt->next = host->templates;
1034 host->templates = rt;
1040 static inline int health_parse_duration(char *string, int *result) {
1041 // make sure it is a number
1042 if(!*string || !(isdigit(*string) || *string == '+' || *string == '-')) {
1048 calculated_number n = strtold(string, &e);
1052 *result = (int) (n * 86400 * 365);
1055 *result = (int) (n * 86400 * 30);
1058 *result = (int) (n * 86400 * 7);
1061 *result = (int) (n * 86400);
1064 *result = (int) (n * 3600);
1067 *result = (int) (n * 60);
1072 *result = (int) (n);
1082 static inline int health_parse_delay(
1083 size_t line, const char *path, const char *file, char *string,
1084 int *delay_up_duration,
1085 int *delay_down_duration,
1086 int *delay_max_duration,
1087 float *delay_multiplier) {
1093 while(*s && !isspace(*s)) s++;
1094 while(*s && isspace(*s)) *s++ = '\0';
1099 while(*s && !isspace(*s)) s++;
1100 while(*s && isspace(*s)) *s++ = '\0';
1102 if(!strcasecmp(key, "up")) {
1103 if (!health_parse_duration(value, delay_up_duration)) {
1104 error("Health configuration at line %zu of file '%s/%s': invalid value '%s' for '%s' keyword",
1105 line, path, file, value, key);
1108 else if(!strcasecmp(key, "down")) {
1109 if (!health_parse_duration(value, delay_down_duration)) {
1110 error("Health configuration at line %zu of file '%s/%s': invalid value '%s' for '%s' keyword",
1111 line, path, file, value, key);
1114 else if(!strcasecmp(key, "multiplier")) {
1115 *delay_multiplier = strtof(value, NULL);
1116 if(isnan(*delay_multiplier) || isinf(*delay_multiplier) || islessequal(*delay_multiplier, 0)) {
1117 error("Health configuration at line %zu of file '%s/%s': invalid value '%s' for '%s' keyword",
1118 line, path, file, value, key);
1121 else if(!strcasecmp(key, "max")) {
1122 if (!health_parse_duration(value, delay_max_duration)) {
1123 error("Health configuration at line %zu of file '%s/%s': invalid value '%s' for '%s' keyword",
1124 line, path, file, value, key);
1128 error("Health configuration at line %zu of file '%s/%s': unknown keyword '%s'",
1129 line, path, file, key);
1136 static inline int health_parse_db_lookup(
1137 size_t line, const char *path, const char *file, char *string,
1138 int *group_method, int *after, int *before, int *every,
1139 uint32_t *options, char **dimensions
1141 debug(D_HEALTH, "Health configuration parsing database lookup %zu@%s/%s: %s", line, path, file, string);
1143 if(*dimensions) freez(*dimensions);
1150 char *s = string, *key;
1152 // first is the group method
1154 while(*s && !isspace(*s)) s++;
1155 while(*s && isspace(*s)) *s++ = '\0';
1157 error("Health configuration invalid chart calculation at line %zu of file '%s/%s': expected group method followed by the 'after' time, but got '%s'",
1158 line, path, file, key);
1162 if((*group_method = web_client_api_request_v1_data_group(key, -1)) == -1) {
1163 error("Health configuration at line %zu of file '%s/%s': invalid group method '%s'",
1164 line, path, file, key);
1168 // then is the 'after' time
1170 while(*s && !isspace(*s)) s++;
1171 while(*s && isspace(*s)) *s++ = '\0';
1173 if(!health_parse_duration(key, after)) {
1174 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' after group method",
1175 line, path, file, key);
1180 *every = abs(*after);
1182 // now we may have optional parameters
1185 while(*s && !isspace(*s)) s++;
1186 while(*s && isspace(*s)) *s++ = '\0';
1189 if(!strcasecmp(key, "at")) {
1191 while(*s && !isspace(*s)) s++;
1192 while(*s && isspace(*s)) *s++ = '\0';
1194 if (!health_parse_duration(value, before)) {
1195 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' for '%s' keyword",
1196 line, path, file, value, key);
1199 else if(!strcasecmp(key, HEALTH_EVERY_KEY)) {
1201 while(*s && !isspace(*s)) s++;
1202 while(*s && isspace(*s)) *s++ = '\0';
1204 if (!health_parse_duration(value, every)) {
1205 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' for '%s' keyword",
1206 line, path, file, value, key);
1209 else if(!strcasecmp(key, "absolute") || !strcasecmp(key, "abs") || !strcasecmp(key, "absolute_sum")) {
1210 *options |= RRDR_OPTION_ABSOLUTE;
1212 else if(!strcasecmp(key, "min2max")) {
1213 *options |= RRDR_OPTION_MIN2MAX;
1215 else if(!strcasecmp(key, "null2zero")) {
1216 *options |= RRDR_OPTION_NULL2ZERO;
1218 else if(!strcasecmp(key, "percentage")) {
1219 *options |= RRDR_OPTION_PERCENTAGE;
1221 else if(!strcasecmp(key, "unaligned")) {
1222 *options |= RRDR_OPTION_NOT_ALIGNED;
1224 else if(!strcasecmp(key, "of")) {
1225 if(*s && strcasecmp(s, "all"))
1226 *dimensions = strdupz(s);
1230 error("Health configuration at line %zu of file '%s/%s': unknown keyword '%s'",
1231 line, path, file, key);
1238 static inline char *health_source_file(size_t line, const char *path, const char *filename) {
1239 char buffer[FILENAME_MAX + 1];
1240 snprintfz(buffer, FILENAME_MAX, "%zu@%s/%s", line, path, filename);
1241 return strdupz(buffer);
1244 static inline void strip_quotes(char *s) {
1246 if(*s == '\'' || *s == '"') *s = ' ';
1251 int health_readfile(const char *path, const char *filename) {
1252 debug(D_HEALTH, "Health configuration reading file '%s/%s'", path, filename);
1254 static uint32_t hash_alarm = 0, hash_template = 0, hash_on = 0, hash_calc = 0, hash_green = 0, hash_red = 0, hash_warn = 0, hash_crit = 0, hash_exec = 0, hash_every = 0, hash_lookup = 0, hash_units = 0, hash_info = 0, hash_recipient = 0, hash_delay = 0;
1255 char buffer[HEALTH_CONF_MAX_LINE + 1];
1257 if(unlikely(!hash_alarm)) {
1258 hash_alarm = simple_uhash(HEALTH_ALARM_KEY);
1259 hash_template = simple_uhash(HEALTH_TEMPLATE_KEY);
1260 hash_on = simple_uhash(HEALTH_ON_KEY);
1261 hash_calc = simple_uhash(HEALTH_CALC_KEY);
1262 hash_lookup = simple_uhash(HEALTH_LOOKUP_KEY);
1263 hash_green = simple_uhash(HEALTH_GREEN_KEY);
1264 hash_red = simple_uhash(HEALTH_RED_KEY);
1265 hash_warn = simple_uhash(HEALTH_WARN_KEY);
1266 hash_crit = simple_uhash(HEALTH_CRIT_KEY);
1267 hash_exec = simple_uhash(HEALTH_EXEC_KEY);
1268 hash_every = simple_uhash(HEALTH_EVERY_KEY);
1269 hash_units = simple_hash(HEALTH_UNITS_KEY);
1270 hash_info = simple_hash(HEALTH_INFO_KEY);
1271 hash_recipient = simple_hash(HEALTH_RECIPIENT_KEY);
1272 hash_delay = simple_uhash(HEALTH_DELAY_KEY);
1275 snprintfz(buffer, HEALTH_CONF_MAX_LINE, "%s/%s", path, filename);
1276 FILE *fp = fopen(buffer, "r");
1278 error("Health configuration cannot read file '%s'.", buffer);
1283 RRDCALCTEMPLATE *rt = NULL;
1285 size_t line = 0, append = 0;
1287 while((s = fgets(&buffer[append], (int)(HEALTH_CONF_MAX_LINE - append), fp)) || append) {
1288 int stop_appending = !s;
1290 // info("Line %zu of file '%s/%s': '%s'", line, path, filename, s);
1293 // info("Trimmed line %zu of file '%s/%s': '%s'", line, path, filename, s);
1296 if(!stop_appending && s[append - 1] == '\\') {
1297 s[append - 1] = ' ';
1298 append = &s[append] - buffer;
1299 if(append < HEALTH_CONF_MAX_LINE)
1302 error("Health configuration has too long muli-line at line %zu of file '%s/%s'.", line, path, filename);
1308 while(*s && *s != ':') s++;
1310 error("Health configuration has invalid line %zu of file '%s/%s'. It does not contain a ':'. Ignoring it.", line, path, filename);
1318 value = trim(value);
1321 error("Health configuration has invalid line %zu of file '%s/%s'. Keyword is empty. Ignoring it.", line, path, filename);
1326 error("Health configuration has invalid line %zu of file '%s/%s'. value is empty. Ignoring it.", line, path, filename);
1330 // info("Health file '%s/%s', key '%s', value '%s'", path, filename, key, value);
1331 uint32_t hash = simple_uhash(key);
1333 if(hash == hash_alarm && !strcasecmp(key, HEALTH_ALARM_KEY)) {
1334 if(rc && !rrdcalc_add_alarm_from_config(&localhost, rc))
1335 rrdcalc_free(&localhost, rc);
1338 if (!rrdcalctemplate_add_template_from_config(&localhost, rt))
1339 rrdcalctemplate_free(&localhost, rt);
1343 rc = callocz(1, sizeof(RRDCALC));
1344 rc->next_event_id = 1;
1345 rc->name = strdupz(value);
1346 rc->hash = simple_hash(rc->name);
1347 rc->source = health_source_file(line, path, filename);
1351 rc->old_value = NAN;
1352 rc->delay_multiplier = 1.0;
1354 if(rrdvar_fix_name(rc->name))
1355 error("Health configuration renamed alarm '%s' to '%s'", value, rc->name);
1357 else if(hash == hash_template && !strcasecmp(key, HEALTH_TEMPLATE_KEY)) {
1359 if(!rrdcalc_add_alarm_from_config(&localhost, rc))
1360 rrdcalc_free(&localhost, rc);
1364 if(rt && !rrdcalctemplate_add_template_from_config(&localhost, rt))
1365 rrdcalctemplate_free(&localhost, rt);
1367 rt = callocz(1, sizeof(RRDCALCTEMPLATE));
1368 rt->name = strdupz(value);
1369 rt->hash_name = simple_hash(rt->name);
1370 rt->source = health_source_file(line, path, filename);
1373 rt->delay_multiplier = 1.0;
1375 if(rrdvar_fix_name(rt->name))
1376 error("Health configuration renamed template '%s' to '%s'", value, rt->name);
1379 if(hash == hash_on && !strcasecmp(key, HEALTH_ON_KEY)) {
1381 if(strcmp(rc->chart, value))
1382 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').",
1383 line, path, filename, rc->name, key, rc->chart, value, value);
1387 rc->chart = strdupz(value);
1388 rc->hash_chart = simple_hash(rc->chart);
1390 else if(hash == hash_lookup && !strcasecmp(key, HEALTH_LOOKUP_KEY)) {
1391 health_parse_db_lookup(line, path, filename, value, &rc->group, &rc->after, &rc->before,
1393 &rc->options, &rc->dimensions);
1395 else if(hash == hash_every && !strcasecmp(key, HEALTH_EVERY_KEY)) {
1396 if(!health_parse_duration(value, &rc->update_every))
1397 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' cannot parse duration: '%s'.",
1398 line, path, filename, rc->name, key, value);
1400 else if(hash == hash_green && !strcasecmp(key, HEALTH_GREEN_KEY)) {
1402 rc->green = strtold(value, &e);
1404 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' leaves this string unmatched: '%s'.",
1405 line, path, filename, rc->name, key, e);
1408 else if(hash == hash_red && !strcasecmp(key, HEALTH_RED_KEY)) {
1410 rc->red = strtold(value, &e);
1412 info("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' leaves this string unmatched: '%s'.",
1413 line, path, filename, rc->name, key, e);
1416 else if(hash == hash_calc && !strcasecmp(key, HEALTH_CALC_KEY)) {
1417 const char *failed_at = NULL;
1419 rc->calculation = expression_parse(value, &failed_at, &error);
1420 if(!rc->calculation) {
1421 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1422 line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
1425 else if(hash == hash_warn && !strcasecmp(key, HEALTH_WARN_KEY)) {
1426 const char *failed_at = NULL;
1428 rc->warning = expression_parse(value, &failed_at, &error);
1430 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1431 line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
1434 else if(hash == hash_crit && !strcasecmp(key, HEALTH_CRIT_KEY)) {
1435 const char *failed_at = NULL;
1437 rc->critical = expression_parse(value, &failed_at, &error);
1439 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1440 line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
1443 else if(hash == hash_exec && !strcasecmp(key, HEALTH_EXEC_KEY)) {
1445 if(strcmp(rc->exec, value))
1446 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').",
1447 line, path, filename, rc->name, key, rc->exec, value, value);
1451 rc->exec = strdupz(value);
1453 else if(hash == hash_recipient && !strcasecmp(key, HEALTH_RECIPIENT_KEY)) {
1455 if(strcmp(rc->recipient, value))
1456 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').",
1457 line, path, filename, rc->name, key, rc->recipient, value, value);
1459 freez(rc->recipient);
1461 rc->recipient = strdupz(value);
1463 else if(hash == hash_units && !strcasecmp(key, HEALTH_UNITS_KEY)) {
1465 if(strcmp(rc->units, value))
1466 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').",
1467 line, path, filename, rc->name, key, rc->units, value, value);
1471 rc->units = strdupz(value);
1472 strip_quotes(rc->units);
1474 else if(hash == hash_info && !strcasecmp(key, HEALTH_INFO_KEY)) {
1476 if(strcmp(rc->info, value))
1477 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').",
1478 line, path, filename, rc->name, key, rc->info, value, value);
1482 rc->info = strdupz(value);
1483 strip_quotes(rc->info);
1485 else if(hash == hash_delay && !strcasecmp(key, HEALTH_DELAY_KEY)) {
1486 health_parse_delay(line, path, filename, value, &rc->delay_up_duration, &rc->delay_down_duration, &rc->delay_max_duration, &rc->delay_multiplier);
1489 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' has unknown key '%s'.",
1490 line, path, filename, rc->name, key);
1494 if(hash == hash_on && !strcasecmp(key, HEALTH_ON_KEY)) {
1496 if(strcmp(rt->context, value))
1497 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').",
1498 line, path, filename, rt->name, key, rt->context, value, value);
1502 rt->context = strdupz(value);
1503 rt->hash_context = simple_hash(rt->context);
1505 else if(hash == hash_lookup && !strcasecmp(key, HEALTH_LOOKUP_KEY)) {
1506 health_parse_db_lookup(line, path, filename, value, &rt->group, &rt->after, &rt->before,
1508 &rt->options, &rt->dimensions);
1510 else if(hash == hash_every && !strcasecmp(key, HEALTH_EVERY_KEY)) {
1511 if(!health_parse_duration(value, &rt->update_every))
1512 info("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' cannot parse duration: '%s'.",
1513 line, path, filename, rt->name, key, value);
1515 else if(hash == hash_green && !strcasecmp(key, HEALTH_GREEN_KEY)) {
1517 rt->green = strtold(value, &e);
1519 info("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' leaves this string unmatched: '%s'.",
1520 line, path, filename, rt->name, key, e);
1523 else if(hash == hash_red && !strcasecmp(key, HEALTH_RED_KEY)) {
1525 rt->red = strtold(value, &e);
1527 info("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' leaves this string unmatched: '%s'.",
1528 line, path, filename, rt->name, key, e);
1531 else if(hash == hash_calc && !strcasecmp(key, HEALTH_CALC_KEY)) {
1532 const char *failed_at = NULL;
1534 rt->calculation = expression_parse(value, &failed_at, &error);
1535 if(!rt->calculation) {
1536 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1537 line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
1540 else if(hash == hash_warn && !strcasecmp(key, HEALTH_WARN_KEY)) {
1541 const char *failed_at = NULL;
1543 rt->warning = expression_parse(value, &failed_at, &error);
1545 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1546 line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
1549 else if(hash == hash_crit && !strcasecmp(key, HEALTH_CRIT_KEY)) {
1550 const char *failed_at = NULL;
1552 rt->critical = expression_parse(value, &failed_at, &error);
1554 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1555 line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
1558 else if(hash == hash_exec && !strcasecmp(key, HEALTH_EXEC_KEY)) {
1560 if(strcmp(rt->exec, value))
1561 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').",
1562 line, path, filename, rt->name, key, rt->exec, value, value);
1566 rt->exec = strdupz(value);
1568 else if(hash == hash_recipient && !strcasecmp(key, HEALTH_RECIPIENT_KEY)) {
1570 if(strcmp(rt->recipient, value))
1571 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').",
1572 line, path, filename, rt->name, key, rt->recipient, value, value);
1574 freez(rt->recipient);
1576 rt->recipient = strdupz(value);
1578 else if(hash == hash_units && !strcasecmp(key, HEALTH_UNITS_KEY)) {
1580 if(strcmp(rt->units, value))
1581 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').",
1582 line, path, filename, rt->name, key, rt->units, value, value);
1586 rt->units = strdupz(value);
1587 strip_quotes(rt->units);
1589 else if(hash == hash_info && !strcasecmp(key, HEALTH_INFO_KEY)) {
1591 if(strcmp(rt->info, value))
1592 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').",
1593 line, path, filename, rt->name, key, rt->info, value, value);
1597 rt->info = strdupz(value);
1598 strip_quotes(rt->info);
1600 else if(hash == hash_delay && !strcasecmp(key, HEALTH_DELAY_KEY)) {
1601 health_parse_delay(line, path, filename, value, &rt->delay_up_duration, &rt->delay_down_duration, &rt->delay_max_duration, &rt->delay_multiplier);
1604 error("Health configuration at line %zu of file '%s/%s' for template '%s' has unknown key '%s'.",
1605 line, path, filename, rt->name, key);
1609 error("Health configuration at line %zu of file '%s/%s' has unknown key '%s'. Expected either '" HEALTH_ALARM_KEY "' or '" HEALTH_TEMPLATE_KEY "'.",
1610 line, path, filename, key);
1614 if(rc && !rrdcalc_add_alarm_from_config(&localhost, rc))
1615 rrdcalc_free(&localhost, rc);
1617 if(rt && !rrdcalctemplate_add_template_from_config(&localhost, rt))
1618 rrdcalctemplate_free(&localhost, rt);
1624 void health_readdir(const char *path) {
1625 size_t pathlen = strlen(path);
1627 debug(D_HEALTH, "Health configuration reading directory '%s'", path);
1629 DIR *dir = opendir(path);
1631 error("Health configuration cannot open directory '%s'.", path);
1635 struct dirent *de = NULL;
1636 while ((de = readdir(dir))) {
1637 size_t len = strlen(de->d_name);
1639 if(de->d_type == DT_DIR
1641 (de->d_name[0] == '.' && de->d_name[1] == '\0')
1642 || (de->d_name[0] == '.' && de->d_name[1] == '.' && de->d_name[2] == '\0')
1644 debug(D_HEALTH, "Ignoring directory '%s'", de->d_name);
1648 else if(de->d_type == DT_DIR) {
1649 char *s = mallocz(pathlen + strlen(de->d_name) + 2);
1652 strcat(s, de->d_name);
1658 else if((de->d_type == DT_LNK || de->d_type == DT_REG || de->d_type == DT_UNKNOWN) &&
1659 len > 5 && !strcmp(&de->d_name[len - 5], ".conf")) {
1660 health_readfile(path, de->d_name);
1663 else debug(D_HEALTH, "Ignoring file '%s'", de->d_name);
1669 static inline char *health_config_dir(void) {
1670 char buffer[FILENAME_MAX + 1];
1671 snprintfz(buffer, FILENAME_MAX, "%s/health.d", config_get("global", "config directory", CONFIG_DIR));
1672 return config_get("health", "health configuration directory", buffer);
1675 void health_init(void) {
1676 debug(D_HEALTH, "Health configuration initializing");
1678 if(!(health_enabled = config_get_boolean("health", "enabled", 1))) {
1679 debug(D_HEALTH, "Health is disabled.");
1683 char *path = health_config_dir();
1686 char buffer[FILENAME_MAX + 1];
1687 snprintfz(buffer, FILENAME_MAX, "%s/alarm-notify.sh", config_get("global", "plugins directory", PLUGINS_DIR));
1688 health_default_exec = config_get("health", "script to execute on alarm", buffer);
1691 long n = config_get_number("health", "in memory max health log entries", (long)localhost.health_log.max);
1693 error("Health configuration has invalid max log entries %ld. Using default %u", n, localhost.health_log.max);
1694 config_set_number("health", "in memory max health log entries", (long)localhost.health_log.max);
1696 else localhost.health_log.max = (unsigned int)n;
1698 rrdhost_rwlock(&localhost);
1699 health_readdir(path);
1700 rrdhost_unlock(&localhost);
1703 // ----------------------------------------------------------------------------
1706 static inline void health_string2json(BUFFER *wb, const char *prefix, const char *label, const char *value, const char *suffix) {
1708 buffer_sprintf(wb, "%s\"%s\":\"%s\"%s", prefix, label, value, suffix);
1710 buffer_sprintf(wb, "%s\"%s\":null%s", prefix, label, suffix);
1713 static inline void health_alarm_entry2json_nolock(BUFFER *wb, ALARM_ENTRY *ae, RRDHOST *host) {
1714 buffer_sprintf(wb, "\n\t{\n"
1715 "\t\t\"hostname\": \"%s\",\n"
1716 "\t\t\"unique_id\": %u,\n"
1717 "\t\t\"alarm_id\": %u,\n"
1718 "\t\t\"alarm_event_id\": %u,\n"
1719 "\t\t\"name\": \"%s\",\n"
1720 "\t\t\"chart\": \"%s\",\n"
1721 "\t\t\"family\": \"%s\",\n"
1722 "\t\t\"processed\": %s,\n"
1723 "\t\t\"updated\": %s,\n"
1724 "\t\t\"exec_run\": %lu,\n"
1725 "\t\t\"exec_failed\": %s,\n"
1726 "\t\t\"exec\": \"%s\",\n"
1727 "\t\t\"recipient\": \"%s\",\n"
1728 "\t\t\"exec_code\": %d,\n"
1729 "\t\t\"source\": \"%s\",\n"
1730 "\t\t\"units\": \"%s\",\n"
1731 "\t\t\"info\": \"%s\",\n"
1732 "\t\t\"when\": %lu,\n"
1733 "\t\t\"duration\": %lu,\n"
1734 "\t\t\"non_clear_duration\": %lu,\n"
1735 "\t\t\"status\": \"%s\",\n"
1736 "\t\t\"old_status\": \"%s\",\n"
1737 "\t\t\"delay\": %d,\n"
1738 "\t\t\"delay_up_to_timestamp\": %lu,\n",
1746 (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_PROCESSED)?"true":"false",
1747 (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_UPDATED)?"true":"false",
1748 (unsigned long)ae->exec_run_timestamp,
1749 (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_EXEC_FAILED)?"true":"false",
1750 ae->exec?ae->exec:health_default_exec,
1751 ae->recipient?ae->recipient:health_default_recipient,
1754 ae->units?ae->units:"",
1755 ae->info?ae->info:"",
1756 (unsigned long)ae->when,
1757 (unsigned long)ae->duration,
1758 (unsigned long)ae->non_clear_duration,
1759 rrdcalc_status2string(ae->new_status),
1760 rrdcalc_status2string(ae->old_status),
1762 (unsigned long)ae->delay_up_to_timestamp
1765 buffer_strcat(wb, "\t\t\"value\":");
1766 buffer_rrd_value(wb, ae->new_value);
1767 buffer_strcat(wb, ",\n");
1769 buffer_strcat(wb, "\t\t\"old_value\":");
1770 buffer_rrd_value(wb, ae->old_value);
1771 buffer_strcat(wb, "\n");
1773 buffer_strcat(wb, "\t}");
1776 void health_alarm_log2json(RRDHOST *host, BUFFER *wb, uint32_t after) {
1777 pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
1779 buffer_strcat(wb, "[");
1781 unsigned int max = host->health_log.max;
1782 unsigned int count = 0;
1784 for(ae = host->health_log.alarms; ae && count < max ; count++, ae = ae->next) {
1785 if(ae->unique_id > after) {
1786 if(likely(count)) buffer_strcat(wb, ",");
1787 health_alarm_entry2json_nolock(wb, ae, host);
1791 buffer_strcat(wb, "\n]\n");
1793 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1796 static inline void health_rrdcalc2json_nolock(BUFFER *wb, RRDCALC *rc) {
1798 "\t\t\"%s.%s\": {\n"
1799 "\t\t\t\"name\": \"%s\",\n"
1800 "\t\t\t\"chart\": \"%s\",\n"
1801 "\t\t\t\"family\": \"%s\",\n"
1802 "\t\t\t\"active\": %s,\n"
1803 "\t\t\t\"exec\": \"%s\",\n"
1804 "\t\t\t\"recipient\": \"%s\",\n"
1805 "\t\t\t\"source\": \"%s\",\n"
1806 "\t\t\t\"units\": \"%s\",\n"
1807 "\t\t\t\"info\": \"%s\",\n"
1808 "\t\t\t\"status\": \"%s\",\n"
1809 "\t\t\t\"last_status_change\": %lu,\n"
1810 "\t\t\t\"last_updated\": %lu,\n"
1811 "\t\t\t\"next_update\": %lu,\n"
1812 "\t\t\t\"update_every\": %d,\n"
1813 "\t\t\t\"delay_up_duration\": %d,\n"
1814 "\t\t\t\"delay_down_duration\": %d,\n"
1815 "\t\t\t\"delay_max_duration\": %d,\n"
1816 "\t\t\t\"delay_multiplier\": %f,\n"
1817 "\t\t\t\"delay\": %d,\n"
1818 "\t\t\t\"delay_up_to_timestamp\": %lu,\n"
1819 , rc->chart, rc->name
1822 , (rc->rrdset && rc->rrdset->family)?rc->rrdset->family:""
1823 , (rc->rrdset)?"true":"false"
1824 , rc->exec?rc->exec:health_default_exec
1825 , rc->recipient?rc->recipient:health_default_recipient
1827 , rc->units?rc->units:""
1828 , rc->info?rc->info:""
1829 , rrdcalc_status2string(rc->status)
1830 , (unsigned long)rc->last_status_change
1831 , (unsigned long)rc->last_updated
1832 , (unsigned long)rc->next_update
1834 , rc->delay_up_duration
1835 , rc->delay_down_duration
1836 , rc->delay_max_duration
1837 , rc->delay_multiplier
1839 , rc->delay_up_to_timestamp
1842 if(RRDCALC_HAS_DB_LOOKUP(rc)) {
1843 if(rc->dimensions && *rc->dimensions)
1844 health_string2json(wb, "\t\t\t", "lookup_dimensions", rc->dimensions, ",\n");
1847 "\t\t\t\"db_after\": %lu,\n"
1848 "\t\t\t\"db_before\": %lu,\n"
1849 "\t\t\t\"lookup_method\": \"%s\",\n"
1850 "\t\t\t\"lookup_after\": %d,\n"
1851 "\t\t\t\"lookup_before\": %d,\n"
1852 "\t\t\t\"lookup_options\": \"",
1853 (unsigned long) rc->db_after,
1854 (unsigned long) rc->db_before,
1855 group_method2string(rc->group),
1859 buffer_data_options2string(wb, rc->options);
1860 buffer_strcat(wb, "\",\n");
1863 if(rc->calculation) {
1864 health_string2json(wb, "\t\t\t", "calc", rc->calculation->source, ",\n");
1865 health_string2json(wb, "\t\t\t", "calc_parsed", rc->calculation->parsed_as, ",\n");
1869 health_string2json(wb, "\t\t\t", "warn", rc->warning->source, ",\n");
1870 health_string2json(wb, "\t\t\t", "warn_parsed", rc->warning->parsed_as, ",\n");
1874 health_string2json(wb, "\t\t\t", "crit", rc->critical->source, ",\n");
1875 health_string2json(wb, "\t\t\t", "crit_parsed", rc->critical->parsed_as, ",\n");
1878 buffer_strcat(wb, "\t\t\t\"green\":");
1879 buffer_rrd_value(wb, rc->green);
1880 buffer_strcat(wb, ",\n");
1882 buffer_strcat(wb, "\t\t\t\"red\":");
1883 buffer_rrd_value(wb, rc->red);
1884 buffer_strcat(wb, ",\n");
1886 buffer_strcat(wb, "\t\t\t\"value\":");
1887 buffer_rrd_value(wb, rc->value);
1888 buffer_strcat(wb, "\n");
1890 buffer_strcat(wb, "\t\t}");
1893 //void health_rrdcalctemplate2json_nolock(BUFFER *wb, RRDCALCTEMPLATE *rt) {
1897 void health_alarms2json(RRDHOST *host, BUFFER *wb, int all) {
1900 rrdhost_rdlock(&localhost);
1901 buffer_sprintf(wb, "{\n\t\"hostname\": \"%s\","
1902 "\n\t\"latest_alarm_log_unique_id\": %u,"
1903 "\n\t\"status\": %s,"
1905 "\n\t\"alarms\": {\n",
1907 (host->health_log.next_log_id > 0)?(host->health_log.next_log_id - 1):0,
1908 health_enabled?"true":"false",
1909 (unsigned long)time(NULL));
1912 for(i = 0, rc = host->alarms; rc ; rc = rc->next) {
1913 if(unlikely(!rc->rrdset || !rc->rrdset->last_collected_time.tv_sec))
1916 if(likely(!all && !(rc->status == RRDCALC_STATUS_WARNING || rc->status == RRDCALC_STATUS_CRITICAL)))
1919 if(likely(i)) buffer_strcat(wb, ",\n");
1920 health_rrdcalc2json_nolock(wb, rc);
1924 // buffer_strcat(wb, "\n\t},\n\t\"templates\": {");
1925 // RRDCALCTEMPLATE *rt;
1926 // for(rt = host->templates; rt ; rt = rt->next)
1927 // health_rrdcalctemplate2json_nolock(wb, rt);
1929 buffer_strcat(wb, "\n\t}\n}\n");
1930 rrdhost_unlock(&localhost);
1934 // ----------------------------------------------------------------------------
1935 // re-load health configuration
1937 static inline void health_free_all_nolock(RRDHOST *host) {
1938 while(host->templates)
1939 rrdcalctemplate_free(host, host->templates);
1942 rrdcalc_free(host, host->alarms);
1945 void health_reload(void) {
1946 if(!health_enabled) {
1947 error("Health reload is requested, but health is not enabled.");
1951 char *path = health_config_dir();
1953 // free all running alarms
1954 rrdhost_rwlock(&localhost);
1955 health_free_all_nolock(&localhost);
1956 rrdhost_unlock(&localhost);
1958 // invalidate all previous entries in the alarm log
1960 for(t = localhost.health_log.alarms ; t ; t = t->next) {
1961 if(t->new_status != RRDCALC_STATUS_REMOVED)
1962 t->notifications |= HEALTH_ENTRY_NOTIFICATIONS_UPDATED;
1965 // reset all thresholds to all charts
1967 for(st = localhost.rrdset_root; st ; st = st->next) {
1972 // load the new alarms
1973 rrdhost_rwlock(&localhost);
1974 health_readdir(path);
1975 rrdhost_unlock(&localhost);
1977 // link the loaded alarms to their charts
1978 for(st = localhost.rrdset_root; st ; st = st->next) {
1979 rrdhost_rwlock(&localhost);
1981 rrdsetcalc_link_matching(st);
1982 rrdcalctemplate_link_matching(st);
1984 rrdhost_unlock(&localhost);
1989 // ----------------------------------------------------------------------------
1990 // health main thread and friends
1992 static inline int rrdcalc_value2status(calculated_number n) {
1993 if(isnan(n)) return RRDCALC_STATUS_UNDEFINED;
1994 if(n) return RRDCALC_STATUS_RAISED;
1995 return RRDCALC_STATUS_CLEAR;
1998 static inline void health_alarm_execute(RRDHOST *host, ALARM_ENTRY *ae) {
1999 ae->notifications |= HEALTH_ENTRY_NOTIFICATIONS_PROCESSED;
2001 if(ae->old_status == RRDCALC_STATUS_UNINITIALIZED && ae->new_status == RRDCALC_STATUS_CLEAR)
2004 char buffer[FILENAME_MAX + 1];
2007 const char *exec = ae->exec;
2008 if(!exec) exec = health_default_exec;
2010 const char *recipient = ae->recipient;
2011 if(!recipient) recipient = health_default_recipient;
2013 snprintfz(buffer, FILENAME_MAX, "exec %s '%s' '%s' '%u' '%u' '%u' '%lu' '%s' '%s' '%s' '%s' '%s' '%0.0Lf' '%0.0Lf' '%s' '%u' '%u' '%s' '%s'",
2020 (unsigned long)ae->when,
2022 ae->chart?ae->chart:"NOCAHRT",
2023 ae->family?ae->family:"NOFAMILY",
2024 rrdcalc_status2string(ae->new_status),
2025 rrdcalc_status2string(ae->old_status),
2028 ae->source?ae->source:"UNKNOWN",
2029 (uint32_t)ae->duration,
2030 (uint32_t)ae->non_clear_duration,
2031 ae->units?ae->units:"",
2032 ae->info?ae->info:""
2035 ae->notifications |= HEALTH_ENTRY_NOTIFICATIONS_EXEC_RUN;
2036 ae->exec_run_timestamp = time(NULL);
2038 debug(D_HEALTH, "executing command '%s'", buffer);
2039 FILE *fp = mypopen(buffer, &command_pid);
2041 error("HEALTH: Cannot popen(\"%s\", \"r\").", buffer);
2044 debug(D_HEALTH, "HEALTH reading from command");
2045 char *s = fgets(buffer, FILENAME_MAX, fp);
2047 debug(D_HEALTH, "HEALTH closing command");
2048 ae->exec_code = mypclose(fp, command_pid);
2049 debug(D_HEALTH, "done executing command - returned with code %d", ae->exec_code);
2051 if(ae->exec_code != 0)
2052 ae->notifications |= HEALTH_ENTRY_NOTIFICATIONS_EXEC_FAILED;
2055 static inline void health_process_notifications(RRDHOST *host, ALARM_ENTRY *ae) {
2056 info("Health alarm '%s.%s' = %0.2Lf - changed status from %s to %s",
2057 ae->chart?ae->chart:"NOCHART", ae->name,
2059 rrdcalc_status2string(ae->old_status),
2060 rrdcalc_status2string(ae->new_status)
2063 health_alarm_execute(host, ae);
2066 static inline void health_alarm_log_process(RRDHOST *host) {
2067 static uint32_t stop_at_id = 0;
2068 uint32_t first_waiting = (host->health_log.alarms)?host->health_log.alarms->unique_id:0;
2069 time_t now = time(NULL);
2071 pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
2074 for(ae = host->health_log.alarms; ae && ae->unique_id >= stop_at_id ; ae = ae->next) {
2076 !(ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_PROCESSED) &&
2077 !(ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_UPDATED)
2080 if(unlikely(ae->unique_id < first_waiting))
2081 first_waiting = ae->unique_id;
2083 if(likely(now > ae->delay_up_to_timestamp))
2084 health_process_notifications(host, ae);
2088 // remember this for the next iteration
2089 stop_at_id = first_waiting;
2091 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
2093 if(host->health_log.count <= host->health_log.max)
2096 // cleanup excess entries in the log
2097 pthread_rwlock_wrlock(&host->health_log.alarm_log_rwlock);
2099 ALARM_ENTRY *last = NULL;
2100 unsigned int count = host->health_log.max * 2 / 3;
2101 for(ae = host->health_log.alarms; ae && count ; count--, last = ae, ae = ae->next) ;
2103 if(ae && last && last->next == ae)
2109 debug(D_HEALTH, "Health removing alarm log entry with id: %u", ae->unique_id);
2111 ALARM_ENTRY *t = ae->next;
2117 freez(ae->recipient);
2124 host->health_log.count--;
2127 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
2130 static inline int rrdcalc_isrunnable(RRDCALC *rc, time_t now, time_t *next_run) {
2131 if (unlikely(!rc->rrdset)) {
2132 debug(D_HEALTH, "Health not running alarm '%s.%s'. It is not linked to a chart.", rc->chart?rc->chart:"NOCHART", rc->name);
2136 if (unlikely(!rc->rrdset->last_collected_time.tv_sec)) {
2137 debug(D_HEALTH, "Health not running alarm '%s.%s'. Chart is not yet collected.", rc->chart?rc->chart:"NOCHART", rc->name);
2141 if (unlikely(!rc->update_every)) {
2142 debug(D_HEALTH, "Health not running alarm '%s.%s'. It does not have an update frequency", rc->chart?rc->chart:"NOCHART", rc->name);
2146 if (unlikely(rc->next_update > now)) {
2147 if (unlikely(*next_run > rc->next_update))
2148 *next_run = rc->next_update;
2150 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));
2155 // we should check that the DB lookup is possible
2157 // - the duration of the chart includes the required timeframe
2158 // we SHOULD NOT check the dimensions - there might be alarms that refer non-existing dimensions (e.g. cpu steal)
2163 void *health_main(void *ptr) {
2166 info("HEALTH thread created with task id %d", gettid());
2168 if(pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, NULL) != 0)
2169 error("Cannot set pthread cancel type to DEFERRED.");
2171 if(pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL) != 0)
2172 error("Cannot set pthread cancel state to ENABLE.");
2174 int min_run_every = (int)config_get_number("health", "run at least every seconds", 10);
2175 if(min_run_every < 1) min_run_every = 1;
2177 BUFFER *wb = buffer_create(100);
2179 unsigned int loop = 0;
2180 while(health_enabled) {
2182 debug(D_HEALTH, "Health monitoring iteration no %u started", loop);
2184 int oldstate, runnable = 0;
2185 time_t now = time(NULL);
2186 time_t next_run = now + min_run_every;
2189 if (unlikely(pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate) != 0))
2190 error("Cannot set pthread cancel state to DISABLE.");
2192 rrdhost_rdlock(&localhost);
2194 // the first loop is to lookup values from the db
2195 for (rc = localhost.alarms; rc; rc = rc->next) {
2196 if (unlikely(!rrdcalc_isrunnable(rc, now, &next_run)))
2200 rc->old_value = rc->value;
2202 // 1. if there is database lookup, do it
2203 // 2. if there is calculation expression, run it
2205 if (unlikely(RRDCALC_HAS_DB_LOOKUP(rc))) {
2206 time_t old_db_timestamp = rc->db_before;
2207 int value_is_null = 0;
2209 int ret = rrd2value(rc->rrdset, wb, &rc->value,
2210 rc->dimensions, 1, rc->after, rc->before, rc->group,
2211 rc->options, &rc->db_after, &rc->db_before, &value_is_null);
2213 if (unlikely(ret != 200)) {
2214 // database lookup failed
2217 debug(D_HEALTH, "Health alarm '%s.%s': database lookup returned error %d", rc->chart?rc->chart:"NOCHART", rc->name, ret);
2219 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_ERROR))) {
2220 rc->rrdcalc_flags |= RRDCALC_FLAG_DB_ERROR;
2221 error("Health alarm '%s.%s': database lookup returned error %d", rc->chart?rc->chart:"NOCHART", rc->name, ret);
2224 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_ERROR))
2225 rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_ERROR;
2227 if (unlikely(old_db_timestamp == rc->db_before)) {
2228 // database is stale
2230 debug(D_HEALTH, "Health alarm '%s.%s': database is stale", rc->chart?rc->chart:"NOCHART", rc->name);
2232 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_STALE))) {
2233 rc->rrdcalc_flags |= RRDCALC_FLAG_DB_STALE;
2234 error("Health alarm '%s.%s': database is stale", rc->chart?rc->chart:"NOCHART", rc->name);
2237 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_STALE))
2238 rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_STALE;
2240 if (unlikely(value_is_null)) {
2241 // collected value is null
2245 debug(D_HEALTH, "Health alarm '%s.%s': database lookup returned empty value (possibly value is not collected yet)",
2246 rc->chart?rc->chart:"NOCHART", rc->name);
2248 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_NAN))) {
2249 rc->rrdcalc_flags |= RRDCALC_FLAG_DB_NAN;
2250 error("Health alarm '%s.%s': database lookup returned empty value (possibly value is not collected yet)",
2251 rc->chart?rc->chart:"NOCHART", rc->name);
2254 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_NAN))
2255 rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_NAN;
2257 debug(D_HEALTH, "Health alarm '%s.%s': database lookup gave value "
2258 CALCULATED_NUMBER_FORMAT, rc->chart?rc->chart:"NOCHART", rc->name, rc->value);
2261 if(unlikely(rc->calculation)) {
2262 if (unlikely(!expression_evaluate(rc->calculation))) {
2263 // calculation failed
2267 debug(D_HEALTH, "Health alarm '%s.%s': failed to evaluate calculation with error: %s",
2268 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->calculation->error_msg));
2270 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_CALC_ERROR))) {
2271 rc->rrdcalc_flags |= RRDCALC_FLAG_CALC_ERROR;
2272 error("Health alarm '%s.%s': failed to evaluate calculation with error: %s",
2273 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->calculation->error_msg));
2277 if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_CALC_ERROR))
2278 rc->rrdcalc_flags &= ~RRDCALC_FLAG_CALC_ERROR;
2280 debug(D_HEALTH, "Health alarm '%s.%s': calculation expression gave value "
2281 CALCULATED_NUMBER_FORMAT
2282 ": %s (source: %s)",
2283 rc->chart?rc->chart:"NOCHART", rc->name,
2284 rc->calculation->result,
2285 buffer_tostring(rc->calculation->error_msg),
2289 rc->value = rc->calculation->result;
2293 rrdhost_unlock(&localhost);
2296 rrdhost_rdlock(&localhost);
2298 for (rc = localhost.alarms; rc; rc = rc->next) {
2299 if (unlikely(!rrdcalc_isrunnable(rc, now, &next_run)))
2302 int warning_status = RRDCALC_STATUS_UNDEFINED;
2303 int critical_status = RRDCALC_STATUS_UNDEFINED;
2305 if(likely(rc->warning)) {
2306 if(unlikely(!expression_evaluate(rc->warning))) {
2307 // calculation failed
2309 debug(D_HEALTH, "Health alarm '%s.%s': warning expression failed with error: %s",
2310 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->warning->error_msg));
2312 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_WARN_ERROR))) {
2313 rc->rrdcalc_flags |= RRDCALC_FLAG_WARN_ERROR;
2314 error("Health alarm '%s.%s': warning expression failed with error: %s",
2315 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->warning->error_msg));
2319 if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_WARN_ERROR))
2320 rc->rrdcalc_flags &= ~RRDCALC_FLAG_WARN_ERROR;
2322 debug(D_HEALTH, "Health alarm '%s.%s': warning expression gave value "
2323 CALCULATED_NUMBER_FORMAT
2324 ": %s (source: %s)",
2325 rc->chart?rc->chart:"NOCHART", rc->name,
2326 rc->warning->result,
2327 buffer_tostring(rc->warning->error_msg),
2331 warning_status = rrdcalc_value2status(rc->warning->result);
2335 if(likely(rc->critical)) {
2336 if(unlikely(!expression_evaluate(rc->critical))) {
2337 // calculation failed
2339 debug(D_HEALTH, "Health alarm '%s.%s': critical expression failed with error: %s",
2340 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->critical->error_msg));
2342 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_CRIT_ERROR))) {
2343 rc->rrdcalc_flags |= RRDCALC_FLAG_CRIT_ERROR;
2344 error("Health alarm '%s.%s': critical expression failed with error: %s",
2345 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->critical->error_msg));
2349 if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_CRIT_ERROR))
2350 rc->rrdcalc_flags &= ~RRDCALC_FLAG_CRIT_ERROR;
2352 debug(D_HEALTH, "Health alarm '%s.%s': critical expression gave value "
2353 CALCULATED_NUMBER_FORMAT
2354 ": %s (source: %s)",
2355 rc->chart?rc->chart:"NOCHART", rc->name,
2356 rc->critical->result,
2357 buffer_tostring(rc->critical->error_msg),
2361 critical_status = rrdcalc_value2status(rc->critical->result);
2365 int status = RRDCALC_STATUS_UNDEFINED;
2367 switch(warning_status) {
2368 case RRDCALC_STATUS_CLEAR:
2369 status = RRDCALC_STATUS_CLEAR;
2372 case RRDCALC_STATUS_RAISED:
2373 status = RRDCALC_STATUS_WARNING;
2380 switch(critical_status) {
2381 case RRDCALC_STATUS_CLEAR:
2382 if(status == RRDCALC_STATUS_UNDEFINED)
2383 status = RRDCALC_STATUS_CLEAR;
2386 case RRDCALC_STATUS_RAISED:
2387 status = RRDCALC_STATUS_CRITICAL;
2394 if(status != rc->status) {
2397 if(now > rc->delay_up_to_timestamp) {
2398 rc->delay_up_current = rc->delay_up_duration;
2399 rc->delay_down_current = rc->delay_down_duration;
2401 rc->delay_up_to_timestamp = 0;
2404 rc->delay_up_current = (int)(rc->delay_up_current * rc->delay_multiplier);
2405 if(rc->delay_up_current > rc->delay_max_duration) rc->delay_up_current = rc->delay_max_duration;
2407 rc->delay_down_current = (int)(rc->delay_down_current * rc->delay_multiplier);
2408 if(rc->delay_down_current > rc->delay_max_duration) rc->delay_down_current = rc->delay_max_duration;
2411 if(status > rc->status)
2412 delay = rc->delay_up_current;
2414 delay = rc->delay_down_current;
2416 if(now + delay < rc->delay_up_to_timestamp)
2417 delay = rc->delay_up_to_timestamp - now;
2419 rc->delay_last = delay;
2420 rc->delay_up_to_timestamp = now + delay;
2421 health_alarm_log(&localhost, rc->id, rc->next_event_id++, now, rc->name, rc->rrdset->id, rc->rrdset->family, rc->exec, rc->recipient, now - rc->last_status_change, rc->old_value, rc->value, rc->status, status, rc->source, rc->units, rc->info, rc->delay_last);
2422 rc->last_status_change = now;
2423 rc->status = status;
2426 rc->last_updated = now;
2427 rc->next_update = now + rc->update_every;
2429 if (next_run > rc->next_update)
2430 next_run = rc->next_update;
2433 rrdhost_unlock(&localhost);
2436 if (unlikely(pthread_setcancelstate(oldstate, NULL) != 0))
2437 error("Cannot set pthread cancel state to RESTORE (%d).", oldstate);
2439 // execute notifications
2441 health_alarm_log_process(&localhost);
2444 if(now < next_run) {
2445 debug(D_HEALTH, "Health monitoring iteration no %u done. Next iteration in %d secs",
2446 loop, (int) (next_run - now));
2447 sleep_usec(1000000 * (unsigned long long) (next_run - now));
2450 debug(D_HEALTH, "Health monitoring iteration no %u done. Next iteration now", loop);
2456 info("HEALTH thread exiting");