3 #define RRDVAR_MAX_LENGTH 1024
5 struct health_options {
6 const char *health_default_exec;
7 const char *health_default_recipient;
8 const char *log_filename;
9 size_t log_entries_written;
13 static struct health_options health = {
14 .health_default_exec = PLUGINS_DIR "/alarm-notify.sh",
15 .health_default_recipient = "root",
16 .log_filename = VARLIB_DIR "/health/alarm_log.db",
17 .log_entries_written = 0,
21 int health_enabled = 1;
23 // ----------------------------------------------------------------------------
24 // health alarm log load/save
25 // no need for locking - only one thread is reading / writing the alarms log
27 static inline int health_alarm_log_open(void) {
29 fclose(health.log_fp);
31 health.log_fp = fopen(health.log_filename, "a");
34 if (setvbuf(health.log_fp, NULL, _IOLBF, 0) != 0)
35 error("Health: cannot set line buffering on health log file.");
39 error("Health: cannot open health log file '%s'. Health data will be lost in case of netdata or server crash.", health.log_filename);
43 static inline void health_alarm_log_close(void) {
45 fclose(health.log_fp);
50 static inline void health_log_rotate(void) {
51 static size_t rotate_every = 0;
53 if(unlikely(rotate_every == 0)) {
54 rotate_every = (size_t)config_get_number("health", "rotate log every lines", 2000);
55 if(rotate_every < 100) rotate_every = 100;
58 if(unlikely(health.log_entries_written > rotate_every)) {
59 health_alarm_log_close();
61 char old_filename[FILENAME_MAX + 1];
62 snprintfz(old_filename, FILENAME_MAX, "%s.old", health.log_filename);
64 if(unlink(old_filename) == -1 && errno != ENOENT)
65 error("Health: cannot remove old alarms log file '%s'", old_filename);
67 if(link(health.log_filename, old_filename) == -1 && errno != ENOENT)
68 error("Health: cannot move file '%s' to '%s'.", health.log_filename, old_filename);
70 if(unlink(health.log_filename) == -1 && errno != ENOENT)
71 error("Health: cannot remove old alarms log file '%s'", health.log_filename);
73 // open it with truncate
74 health.log_fp = fopen(health.log_filename, "w");
77 fclose(health.log_fp);
79 error("Health: cannot truncate health log '%s'", health.log_filename);
83 health.log_entries_written = 0;
84 health_alarm_log_open();
88 static inline void health_alarm_log_save(RRDHOST *host, ALARM_ENTRY *ae) {
91 if(likely(health.log_fp)) {
92 if(unlikely(fprintf(health.log_fp
94 "\t%08x\t%08x\t%08x\t%08x\t%08x"
97 "\t%s\t%s\t%s\t%s\t%s\t%s\t%s\t%s"
101 , (ae->flags & HEALTH_ENTRY_FLAG_SAVED)?'U':'A'
111 , (uint32_t)ae->duration
112 , (uint32_t)ae->non_clear_duration
113 , (uint32_t)ae->flags
114 , (uint32_t)ae->exec_run_timestamp
115 , (uint32_t)ae->delay_up_to_timestamp
117 , (ae->name)?ae->name:""
118 , (ae->chart)?ae->chart:""
119 , (ae->family)?ae->family:""
120 , (ae->exec)?ae->exec:""
121 , (ae->recipient)?ae->recipient:""
122 , (ae->source)?ae->source:""
123 , (ae->units)?ae->units:""
124 , (ae->info)?ae->info:""
131 , (long double)ae->new_value
132 , (long double)ae->old_value
134 error("Health: failed to save alarm log entry. Health data may be lost in case of abnormal restart.");
136 ae->flags |= HEALTH_ENTRY_FLAG_SAVED;
137 health.log_entries_written++;
142 static inline ssize_t health_alarm_log_read(RRDHOST *host, FILE *fp, const char *filename) {
143 static uint32_t max_unique_id = 0, max_alarm_id = 0;
147 char *s, *buf = mallocz(65536 + 1);
148 size_t line = 0, len = 0;
149 ssize_t loaded = 0, updated = 0, errored = 0, duplicate = 0;
151 pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
153 while((s = fgets_trim_len(buf, 65536, fp, &len))) {
154 health.log_entries_written++;
157 int max_entries = 30, entries = 0;
158 char *pointers[max_entries];
160 pointers[entries++] = s++;
162 if(unlikely(*s == '\t')) {
164 pointers[entries++] = ++s;
165 if(entries >= max_entries) {
166 error("Health: line %zu of file '%s' has more than %d entries. Ignoring excessive entries.", line, filename, max_entries);
173 if(likely(*pointers[0] == 'U' || *pointers[0] == 'A')) {
174 ALARM_ENTRY *ae = NULL;
177 error("Health: line %zu of file '%s' should have at least 26 entries, but it has %d. Ignoring it.", line, filename, entries);
182 // check that we have valid ids
183 uint32_t unique_id = (uint32_t)strtoul(pointers[2], NULL, 16);
185 error("Health: line %zu of file '%s' states alarm entry with invalid unique id %u (%s). Ignoring it.", line, filename, unique_id, pointers[2]);
190 uint32_t alarm_id = (uint32_t)strtoul(pointers[3], NULL, 16);
192 error("Health: line %zu of file '%s' states alarm entry for invalid alarm id %u (%s). Ignoring it.", line, filename, alarm_id, pointers[3]);
197 if(unlikely(*pointers[0] == 'A')) {
198 // make sure it is properly numbered
199 if(unlikely(host->health_log.alarms && unique_id < host->health_log.alarms->unique_id)) {
200 error("Health: line %zu of file '%s' has alarm log entry with %u in wrong order. Ignoring it.", line, filename, unique_id);
205 ae = callocz(1, sizeof(ALARM_ENTRY));
207 else if(unlikely(*pointers[0] == 'U')) {
209 for(ae = host->health_log.alarms; ae; ae = ae->next) {
210 if(unlikely(unique_id == ae->unique_id)) {
211 if(unlikely(*pointers[0] == 'A')) {
212 error("Health: line %zu of file '%s' adds duplicate alarm log entry with unique id %u. Using the later."
213 , line, filename, unique_id);
219 else if(unlikely(unique_id > ae->unique_id)) {
220 // no need to continue
221 // the linked list is sorted
227 // if not found, skip this line
229 // error("Health: line %zu of file '%s' updates alarm log entry with unique id %u, but it is not found.", line, filename, unique_id);
234 // check for a possible host missmatch
235 //if(strcmp(pointers[1], host->hostname))
236 // error("Health: line %zu of file '%s' provides an alarm for host '%s' but this is named '%s'.", line, filename, pointers[1], host->hostname);
238 ae->unique_id = unique_id;
239 ae->alarm_id = alarm_id;
240 ae->alarm_event_id = (uint32_t)strtoul(pointers[4], NULL, 16);
241 ae->updated_by_id = (uint32_t)strtoul(pointers[5], NULL, 16);
242 ae->updates_id = (uint32_t)strtoul(pointers[6], NULL, 16);
244 ae->when = (uint32_t)strtoul(pointers[7], NULL, 16);
245 ae->duration = (uint32_t)strtoul(pointers[8], NULL, 16);
246 ae->non_clear_duration = (uint32_t)strtoul(pointers[9], NULL, 16);
248 ae->flags = (uint32_t)strtoul(pointers[10], NULL, 16);
249 ae->flags |= HEALTH_ENTRY_FLAG_SAVED;
251 ae->exec_run_timestamp = (uint32_t)strtoul(pointers[11], NULL, 16);
252 ae->delay_up_to_timestamp = (uint32_t)strtoul(pointers[12], NULL, 16);
254 if(unlikely(ae->name)) freez(ae->name);
255 ae->name = strdupz(pointers[13]);
256 ae->hash_name = simple_hash(ae->name);
258 if(unlikely(ae->chart)) freez(ae->chart);
259 ae->chart = strdupz(pointers[14]);
260 ae->hash_chart = simple_hash(ae->chart);
262 if(unlikely(ae->family)) freez(ae->family);
263 ae->family = strdupz(pointers[15]);
265 if(unlikely(ae->exec)) freez(ae->exec);
266 ae->exec = strdupz(pointers[16]);
267 if(!*ae->exec) { freez(ae->exec); ae->exec = NULL; }
269 if(unlikely(ae->recipient)) freez(ae->recipient);
270 ae->recipient = strdupz(pointers[17]);
271 if(!*ae->recipient) { freez(ae->recipient); ae->recipient = NULL; }
273 if(unlikely(ae->source)) freez(ae->source);
274 ae->source = strdupz(pointers[18]);
275 if(!*ae->source) { freez(ae->source); ae->source = NULL; }
277 if(unlikely(ae->units)) freez(ae->units);
278 ae->units = strdupz(pointers[19]);
279 if(!*ae->units) { freez(ae->units); ae->units = NULL; }
281 if(unlikely(ae->info)) freez(ae->info);
282 ae->info = strdupz(pointers[20]);
283 if(!*ae->info) { freez(ae->info); ae->info = NULL; }
285 ae->exec_code = atoi(pointers[21]);
286 ae->new_status = atoi(pointers[22]);
287 ae->old_status = atoi(pointers[23]);
288 ae->delay = atoi(pointers[24]);
290 ae->new_value = strtold(pointers[25], NULL);
291 ae->old_value = strtold(pointers[26], NULL);
293 // add it to host if not already there
294 if(unlikely(*pointers[0] == 'A')) {
295 ae->next = host->health_log.alarms;
296 host->health_log.alarms = ae;
301 if(unlikely(ae->unique_id > max_unique_id))
302 max_unique_id = ae->unique_id;
304 if(unlikely(ae->alarm_id >= max_alarm_id))
305 max_alarm_id = ae->alarm_id;
308 error("Health: line %zu of file '%s' is invalid (unrecognized entry type '%s').", line, filename, pointers[0]);
313 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
317 if(!max_unique_id) max_unique_id = (uint32_t)time(NULL);
318 if(!max_alarm_id) max_alarm_id = (uint32_t)time(NULL);
320 host->health_log.next_log_id = max_unique_id + 1;
321 host->health_log.next_alarm_id = max_alarm_id + 1;
323 debug(D_HEALTH, "Health: loaded file '%s' with %zd new alarm entries, updated %zd alarms, errors %zd entries, duplicate %zd", filename, loaded, updated, errored, duplicate);
327 static inline void health_alarm_log_load(RRDHOST *host) {
328 health_alarm_log_close();
330 char filename[FILENAME_MAX + 1];
331 snprintfz(filename, FILENAME_MAX, "%s.old", health.log_filename);
332 FILE *fp = fopen(filename, "r");
334 error("Health: cannot open health file: %s", filename);
336 health_alarm_log_read(host, fp, filename);
340 health.log_entries_written = 0;
341 fp = fopen(health.log_filename, "r");
343 error("Health: cannot open health file: %s", health.log_filename);
345 health_alarm_log_read(host, fp, health.log_filename);
349 health_alarm_log_open();
353 // ----------------------------------------------------------------------------
354 // health alarm log management
356 static inline void health_alarm_log(RRDHOST *host,
357 uint32_t alarm_id, uint32_t alarm_event_id,
359 const char *name, const char *chart, const char *family,
360 const char *exec, const char *recipient, time_t duration,
361 calculated_number old_value, calculated_number new_value,
362 int old_status, int new_status,
368 debug(D_HEALTH, "Health adding alarm log entry with id: %u", host->health_log.next_log_id);
370 ALARM_ENTRY *ae = callocz(1, sizeof(ALARM_ENTRY));
371 ae->name = strdupz(name);
372 ae->hash_name = simple_hash(ae->name);
375 ae->chart = strdupz(chart);
376 ae->hash_chart = simple_hash(ae->chart);
380 ae->family = strdupz(family);
382 if(exec) ae->exec = strdupz(exec);
383 if(recipient) ae->recipient = strdupz(recipient);
384 if(source) ae->source = strdupz(source);
385 if(units) ae->units = strdupz(units);
386 if(info) ae->info = strdupz(info);
388 ae->unique_id = host->health_log.next_log_id++;
389 ae->alarm_id = alarm_id;
390 ae->alarm_event_id = alarm_event_id;
392 ae->old_value = old_value;
393 ae->new_value = new_value;
394 ae->old_status = old_status;
395 ae->new_status = new_status;
396 ae->duration = duration;
398 ae->delay_up_to_timestamp = when + delay;
400 if(ae->old_status == RRDCALC_STATUS_WARNING || ae->old_status == RRDCALC_STATUS_CRITICAL)
401 ae->non_clear_duration += ae->duration;
404 pthread_rwlock_wrlock(&host->health_log.alarm_log_rwlock);
405 ae->next = host->health_log.alarms;
406 host->health_log.alarms = ae;
407 host->health_log.count++;
408 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
410 // match previous alarms
411 pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
413 for(t = host->health_log.alarms ; t ; t = t->next) {
414 if(t != ae && t->alarm_id == ae->alarm_id) {
415 if(!(t->flags & HEALTH_ENTRY_FLAG_UPDATED) && !t->updated_by_id) {
416 t->flags |= HEALTH_ENTRY_FLAG_UPDATED;
417 t->updated_by_id = ae->unique_id;
418 ae->updates_id = t->unique_id;
420 if((t->new_status == RRDCALC_STATUS_WARNING || t->new_status == RRDCALC_STATUS_CRITICAL) &&
421 (t->old_status == RRDCALC_STATUS_WARNING || t->old_status == RRDCALC_STATUS_CRITICAL))
422 ae->non_clear_duration += t->non_clear_duration;
424 health_alarm_log_save(host, t);
427 // no need to continue
431 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
433 health_alarm_log_save(host, ae);
436 // ----------------------------------------------------------------------------
439 static inline int rrdvar_fix_name(char *variable) {
442 if (!isalnum(*variable) && *variable != '.' && *variable != '_') {
453 int rrdvar_compare(void* a, void* b) {
454 if(((RRDVAR *)a)->hash < ((RRDVAR *)b)->hash) return -1;
455 else if(((RRDVAR *)a)->hash > ((RRDVAR *)b)->hash) return 1;
456 else return strcmp(((RRDVAR *)a)->name, ((RRDVAR *)b)->name);
459 static inline RRDVAR *rrdvar_index_add(avl_tree_lock *tree, RRDVAR *rv) {
460 RRDVAR *ret = (RRDVAR *)avl_insert_lock(tree, (avl *)(rv));
462 debug(D_VARIABLES, "Request to insert RRDVAR '%s' into index failed. Already exists.", rv->name);
467 static inline RRDVAR *rrdvar_index_del(avl_tree_lock *tree, RRDVAR *rv) {
468 RRDVAR *ret = (RRDVAR *)avl_remove_lock(tree, (avl *)(rv));
470 error("Request to remove RRDVAR '%s' from index failed. Not Found.", rv->name);
475 static inline RRDVAR *rrdvar_index_find(avl_tree_lock *tree, const char *name, uint32_t hash) {
477 tmp.name = (char *)name;
478 tmp.hash = (hash)?hash:simple_hash(tmp.name);
480 return (RRDVAR *)avl_search_lock(tree, (avl *)&tmp);
483 static inline void rrdvar_free(RRDHOST *host, avl_tree_lock *tree, RRDVAR *rv) {
489 debug(D_VARIABLES, "Deleting variable '%s'", rv->name);
490 if(unlikely(!rrdvar_index_del(tree, rv)))
491 error("Attempted to delete variable '%s' from host '%s', but it is not found.", rv->name, host->hostname);
498 static inline RRDVAR *rrdvar_create_and_index(const char *scope, avl_tree_lock *tree, const char *name, int type, void *value) {
499 char *variable = strdupz(name);
500 rrdvar_fix_name(variable);
501 uint32_t hash = simple_hash(variable);
503 RRDVAR *rv = rrdvar_index_find(tree, variable, hash);
505 debug(D_VARIABLES, "Variable '%s' not found in scope '%s'. Creating a new one.", variable, scope);
507 rv = callocz(1, sizeof(RRDVAR));
513 RRDVAR *ret = rrdvar_index_add(tree, rv);
514 if(unlikely(ret != rv)) {
515 debug(D_VARIABLES, "Variable '%s' in scope '%s' already exists", variable, scope);
516 rrdvar_free(NULL, NULL, rv);
520 debug(D_VARIABLES, "Variable '%s' created in scope '%s'", variable, scope);
523 debug(D_VARIABLES, "Variable '%s' is already found in scope '%s'.", variable, scope);
529 // it must return NULL - not the existing variable - or double-free will happen
536 // ----------------------------------------------------------------------------
539 RRDVAR *rrdvar_custom_host_variable_create(RRDHOST *host, const char *name) {
540 calculated_number *v = callocz(1, sizeof(calculated_number));
542 RRDVAR *rv = rrdvar_create_and_index("host", &host->variables_root_index, name, RRDVAR_TYPE_CALCULATED_ALLOCATED, v);
545 error("Requested variable '%s' already exists - possibly 2 plugins will be updating it at the same time", name);
547 char *variable = strdupz(name);
548 rrdvar_fix_name(variable);
549 uint32_t hash = simple_hash(variable);
551 rv = rrdvar_index_find(&host->variables_root_index, variable, hash);
557 void rrdvar_custom_host_variable_destroy(RRDHOST *host, const char *name) {
558 char *variable = strdupz(name);
559 rrdvar_fix_name(variable);
560 uint32_t hash = simple_hash(variable);
562 RRDVAR *rv = rrdvar_index_find(&host->variables_root_index, variable, hash);
566 error("Attempted to remove variable '%s' from host '%s', but it does not exist.", name, host->hostname);
570 if(rv->type != RRDVAR_TYPE_CALCULATED_ALLOCATED) {
571 error("Attempted to remove variable '%s' from host '%s', but it does not a custom allocated variable.", name, host->hostname);
575 if(!rrdvar_index_del(&host->variables_root_index, rv)) {
576 error("Attempted to remove variable '%s' from host '%s', but it cannot be found.", name, host->hostname);
585 void rrdvar_custom_host_variable_set(RRDVAR *rv, calculated_number value) {
586 if(rv->type != RRDVAR_TYPE_CALCULATED_ALLOCATED)
587 error("requested to set variable '%s' to value " CALCULATED_NUMBER_FORMAT " but the variable is not a custom one.", rv->name, value);
589 calculated_number *v = rv->value;
594 // ----------------------------------------------------------------------------
597 static calculated_number rrdvar2number(RRDVAR *rv) {
599 case RRDVAR_TYPE_CALCULATED_ALLOCATED:
600 case RRDVAR_TYPE_CALCULATED: {
601 calculated_number *n = (calculated_number *)rv->value;
605 case RRDVAR_TYPE_TIME_T: {
606 time_t *n = (time_t *)rv->value;
610 case RRDVAR_TYPE_COLLECTED: {
611 collected_number *n = (collected_number *)rv->value;
615 case RRDVAR_TYPE_TOTAL: {
616 total_number *n = (total_number *)rv->value;
620 case RRDVAR_TYPE_INT: {
621 int *n = (int *)rv->value;
626 error("I don't know how to convert RRDVAR type %d to calculated_number", rv->type);
631 int health_variable_lookup(const char *variable, uint32_t hash, RRDCALC *rc, calculated_number *result) {
632 RRDSET *st = rc->rrdset;
637 rv = rrdvar_index_find(&st->variables_root_index, variable, hash);
639 *result = rrdvar2number(rv);
643 rv = rrdvar_index_find(&st->rrdfamily->variables_root_index, variable, hash);
645 *result = rrdvar2number(rv);
649 rv = rrdvar_index_find(&st->rrdhost->variables_root_index, variable, hash);
651 *result = rrdvar2number(rv);
658 // ----------------------------------------------------------------------------
661 struct variable2json_helper {
666 static void single_variable2json(void *entry, void *data) {
667 struct variable2json_helper *helper = (struct variable2json_helper *)data;
668 RRDVAR *rv = (RRDVAR *)entry;
669 calculated_number value = rrdvar2number(rv);
671 if(unlikely(isnan(value) || isinf(value)))
672 buffer_sprintf(helper->buf, "%s\n\t\t\"%s\": null", helper->counter?",":"", rv->name);
674 buffer_sprintf(helper->buf, "%s\n\t\t\"%s\": %0.5Lf", helper->counter?",":"", rv->name, (long double)value);
679 void health_api_v1_chart_variables2json(RRDSET *st, BUFFER *buf) {
680 struct variable2json_helper helper = {
685 buffer_sprintf(buf, "{\n\t\"chart\": \"%s.%s\",\n\t\"chart_name\": \"%s.%s\",\n\t\"chart_variables\": {", st->type, st->id, st->type, st->name);
686 avl_traverse_lock(&st->variables_root_index, single_variable2json, (void *)&helper);
687 buffer_sprintf(buf, "\n\t},\n\t\"family\": \"%s\",\n\t\"family_variables\": {", st->family);
689 avl_traverse_lock(&st->rrdfamily->variables_root_index, single_variable2json, (void *)&helper);
690 buffer_sprintf(buf, "\n\t},\n\t\"host\": \"%s\",\n\t\"host_variables\": {", st->rrdhost->hostname);
692 avl_traverse_lock(&st->rrdhost->variables_root_index, single_variable2json, (void *)&helper);
693 buffer_strcat(buf, "\n\t}\n}\n");
697 // ----------------------------------------------------------------------------
698 // RRDDIMVAR management
699 // DIMENSION VARIABLES
701 #define RRDDIMVAR_ID_MAX 1024
703 static inline void rrddimvar_free_variables(RRDDIMVAR *rs) {
704 RRDDIM *rd = rs->rrddim;
705 RRDSET *st = rd->rrdset;
707 // CHART VARIABLES FOR THIS DIMENSION
709 rrdvar_free(st->rrdhost, &st->variables_root_index, rs->var_local_id);
710 rs->var_local_id = NULL;
712 rrdvar_free(st->rrdhost, &st->variables_root_index, rs->var_local_name);
713 rs->var_local_name = NULL;
715 // FAMILY VARIABLES FOR THIS DIMENSION
717 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->var_family_id);
718 rs->var_family_id = NULL;
720 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->var_family_name);
721 rs->var_family_name = NULL;
723 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->var_family_contextid);
724 rs->var_family_contextid = NULL;
726 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->var_family_contextname);
727 rs->var_family_contextname = NULL;
729 // HOST VARIABLES FOR THIS DIMENSION
731 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->var_host_chartidid);
732 rs->var_host_chartidid = NULL;
734 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->var_host_chartidname);
735 rs->var_host_chartidname = NULL;
737 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->var_host_chartnameid);
738 rs->var_host_chartnameid = NULL;
740 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->var_host_chartnamename);
741 rs->var_host_chartnamename = NULL;
751 freez(rs->key_fullidid);
752 rs->key_fullidid = NULL;
754 freez(rs->key_fullidname);
755 rs->key_fullidname = NULL;
757 freez(rs->key_contextid);
758 rs->key_contextid = NULL;
760 freez(rs->key_contextname);
761 rs->key_contextname = NULL;
763 freez(rs->key_fullnameid);
764 rs->key_fullnameid = NULL;
766 freez(rs->key_fullnamename);
767 rs->key_fullnamename = NULL;
770 static inline void rrddimvar_create_variables(RRDDIMVAR *rs) {
771 rrddimvar_free_variables(rs);
773 RRDDIM *rd = rs->rrddim;
774 RRDSET *st = rd->rrdset;
776 char buffer[RRDDIMVAR_ID_MAX + 1];
780 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s%s%s", rs->prefix, rd->id, rs->suffix);
781 rs->key_id = strdupz(buffer);
783 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s%s%s", rs->prefix, rd->name, rs->suffix);
784 rs->key_name = strdupz(buffer);
786 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->id, rs->key_id);
787 rs->key_fullidid = strdupz(buffer);
789 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->id, rs->key_name);
790 rs->key_fullidname = strdupz(buffer);
792 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->context, rs->key_id);
793 rs->key_contextid = strdupz(buffer);
795 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->context, rs->key_name);
796 rs->key_contextname = strdupz(buffer);
798 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->name, rs->key_id);
799 rs->key_fullnameid = strdupz(buffer);
801 snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->name, rs->key_name);
802 rs->key_fullnamename = strdupz(buffer);
804 // CHART VARIABLES FOR THIS DIMENSION
805 // -----------------------------------
807 // dimensions are available as:
811 rs->var_local_id = rrdvar_create_and_index("local", &st->variables_root_index, rs->key_id, rs->type, rs->value);
812 rs->var_local_name = rrdvar_create_and_index("local", &st->variables_root_index, rs->key_name, rs->type, rs->value);
814 // FAMILY VARIABLES FOR THIS DIMENSION
815 // -----------------------------------
817 // dimensions are available as:
818 // - $id (only the first, when multiple overlap)
819 // - $name (only the first, when multiple overlap)
820 // - $chart-context.id
821 // - $chart-context.name
823 rs->var_family_id = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->key_id, rs->type, rs->value);
824 rs->var_family_name = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->key_name, rs->type, rs->value);
825 rs->var_family_contextid = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->key_contextid, rs->type, rs->value);
826 rs->var_family_contextname = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->key_contextname, rs->type, rs->value);
828 // HOST VARIABLES FOR THIS DIMENSION
829 // -----------------------------------
831 // dimensions are available as:
835 // - $chart-name.name
837 rs->var_host_chartidid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->key_fullidid, rs->type, rs->value);
838 rs->var_host_chartidname = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->key_fullidname, rs->type, rs->value);
839 rs->var_host_chartnameid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->key_fullnameid, rs->type, rs->value);
840 rs->var_host_chartnamename = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->key_fullnamename, rs->type, rs->value);
843 RRDDIMVAR *rrddimvar_create(RRDDIM *rd, int type, const char *prefix, const char *suffix, void *value, uint32_t options) {
844 RRDSET *st = rd->rrdset;
846 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:"");
848 if(!prefix) prefix = "";
849 if(!suffix) suffix = "";
851 RRDDIMVAR *rs = (RRDDIMVAR *)callocz(1, sizeof(RRDDIMVAR));
853 rs->prefix = strdupz(prefix);
854 rs->suffix = strdupz(suffix);
858 rs->options = options;
861 rs->next = rd->variables;
864 rrddimvar_create_variables(rs);
869 void rrddimvar_rename_all(RRDDIM *rd) {
870 RRDSET *st = rd->rrdset;
871 debug(D_VARIABLES, "RRDDIMSET rename for chart id '%s' name '%s', dimension id '%s', name '%s'", st->id, st->name, rd->id, rd->name);
873 RRDDIMVAR *rs, *next = rd->variables;
876 rrddimvar_create_variables(rs);
880 void rrddimvar_free(RRDDIMVAR *rs) {
881 RRDDIM *rd = rs->rrddim;
882 RRDSET *st = rd->rrdset;
883 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);
885 rrddimvar_free_variables(rs);
887 if(rd->variables == rs) {
888 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);
889 rd->variables = rs->next;
892 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);
894 for (t = rd->variables; t && t->next != rs; t = t->next) ;
895 if(!t) error("RRDDIMVAR '%s' not found in dimension '%s/%s' variables linked list", rs->key_name, st->id, rd->id);
896 else t->next = rs->next;
904 // ----------------------------------------------------------------------------
905 // RRDSETVAR management
908 static inline void rrdsetvar_free_variables(RRDSETVAR *rs) {
909 RRDSET *st = rs->rrdset;
913 rrdvar_free(st->rrdhost, &st->variables_root_index, rs->var_local);
914 rs->var_local = NULL;
918 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->var_family);
919 rs->var_family = NULL;
921 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->var_host);
926 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->var_family_name);
927 rs->var_family_name = NULL;
929 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->var_host_name);
930 rs->var_host_name = NULL;
934 freez(rs->key_fullid);
935 rs->key_fullid = NULL;
937 freez(rs->key_fullname);
938 rs->key_fullname = NULL;
941 static inline void rrdsetvar_create_variables(RRDSETVAR *rs) {
942 rrdsetvar_free_variables(rs);
944 RRDSET *st = rs->rrdset;
948 char buffer[RRDVAR_MAX_LENGTH + 1];
949 snprintfz(buffer, RRDVAR_MAX_LENGTH, "%s.%s", st->id, rs->variable);
950 rs->key_fullid = strdupz(buffer);
952 snprintfz(buffer, RRDVAR_MAX_LENGTH, "%s.%s", st->name, rs->variable);
953 rs->key_fullname = strdupz(buffer);
957 rs->var_local = rrdvar_create_and_index("local", &st->variables_root_index, rs->variable, rs->type, rs->value);
961 rs->var_family = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->key_fullid, rs->type, rs->value);
962 rs->var_family_name = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->key_fullname, rs->type, rs->value);
966 rs->var_host = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->key_fullid, rs->type, rs->value);
967 rs->var_host_name = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->key_fullname, rs->type, rs->value);
971 RRDSETVAR *rrdsetvar_create(RRDSET *st, const char *variable, int type, void *value, uint32_t options) {
972 debug(D_VARIABLES, "RRDVARSET create for chart id '%s' name '%s' with variable name '%s'", st->id, st->name, variable);
973 RRDSETVAR *rs = (RRDSETVAR *)callocz(1, sizeof(RRDSETVAR));
975 rs->variable = strdupz(variable);
978 rs->options = options;
981 rs->next = st->variables;
984 rrdsetvar_create_variables(rs);
989 void rrdsetvar_rename_all(RRDSET *st) {
990 debug(D_VARIABLES, "RRDSETVAR rename for chart id '%s' name '%s'", st->id, st->name);
992 RRDSETVAR *rs, *next = st->variables;
995 rrdsetvar_create_variables(rs);
998 rrdsetcalc_link_matching(st);
1001 void rrdsetvar_free(RRDSETVAR *rs) {
1002 RRDSET *st = rs->rrdset;
1003 debug(D_VARIABLES, "RRDSETVAR free for chart id '%s' name '%s', variable '%s'", st->id, st->name, rs->variable);
1005 if(st->variables == rs) {
1006 st->variables = rs->next;
1010 for (t = st->variables; t && t->next != rs; t = t->next);
1011 if(!t) error("RRDSETVAR '%s' not found in chart '%s' variables linked list", rs->key_fullname, st->id);
1012 else t->next = rs->next;
1015 rrdsetvar_free_variables(rs);
1017 freez(rs->variable);
1021 // ----------------------------------------------------------------------------
1022 // RRDCALC management
1024 static inline const char *rrdcalc_status2string(int status) {
1026 case RRDCALC_STATUS_REMOVED:
1029 case RRDCALC_STATUS_UNDEFINED:
1032 case RRDCALC_STATUS_UNINITIALIZED:
1033 return "UNINITIALIZED";
1035 case RRDCALC_STATUS_CLEAR:
1038 case RRDCALC_STATUS_RAISED:
1041 case RRDCALC_STATUS_WARNING:
1044 case RRDCALC_STATUS_CRITICAL:
1048 error("Unknown alarm status %d", status);
1053 static void rrdsetcalc_link(RRDSET *st, RRDCALC *rc) {
1054 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);
1056 rc->last_status_change = time(NULL);
1059 rc->rrdset_next = st->alarms;
1060 rc->rrdset_prev = NULL;
1063 rc->rrdset_next->rrdset_prev = rc;
1067 if(rc->update_every < rc->rrdset->update_every) {
1068 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);
1069 rc->update_every = rc->rrdset->update_every;
1072 if(!isnan(rc->green) && isnan(st->green)) {
1073 debug(D_HEALTH, "Health alarm '%s.%s' green threshold set from %Lf to %Lf.", rc->rrdset->id, rc->name, rc->rrdset->green, rc->green);
1074 st->green = rc->green;
1077 if(!isnan(rc->red) && isnan(st->red)) {
1078 debug(D_HEALTH, "Health alarm '%s.%s' red threshold set from %Lf to %Lf.", rc->rrdset->id, rc->name, rc->rrdset->red, rc->red);
1082 rc->local = rrdvar_create_and_index("local", &st->variables_root_index, rc->name, RRDVAR_TYPE_CALCULATED, &rc->value);
1083 rc->family = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rc->name, RRDVAR_TYPE_CALCULATED, &rc->value);
1085 char fullname[RRDVAR_MAX_LENGTH + 1];
1086 snprintfz(fullname, RRDVAR_MAX_LENGTH, "%s.%s", st->id, rc->name);
1087 rc->hostid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, fullname, RRDVAR_TYPE_CALCULATED, &rc->value);
1089 snprintfz(fullname, RRDVAR_MAX_LENGTH, "%s.%s", st->name, rc->name);
1090 rc->hostname = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, fullname, RRDVAR_TYPE_CALCULATED, &rc->value);
1092 if(!rc->units) rc->units = strdupz(st->units);
1095 time_t now = time(NULL);
1096 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);
1100 static inline int rrdcalc_is_matching_this_rrdset(RRDCALC *rc, RRDSET *st) {
1101 if( (rc->hash_chart == st->hash && !strcmp(rc->chart, st->id)) ||
1102 (rc->hash_chart == st->hash_name && !strcmp(rc->chart, st->name)))
1108 // this has to be called while the RRDHOST is locked
1109 inline void rrdsetcalc_link_matching(RRDSET *st) {
1110 // debug(D_HEALTH, "find matching alarms for chart '%s'", st->id);
1113 for(rc = st->rrdhost->alarms; rc ; rc = rc->next) {
1114 if(unlikely(rc->rrdset))
1117 if(unlikely(rrdcalc_is_matching_this_rrdset(rc, st)))
1118 rrdsetcalc_link(st, rc);
1122 // this has to be called while the RRDHOST is locked
1123 inline void rrdsetcalc_unlink(RRDCALC *rc) {
1124 RRDSET *st = rc->rrdset;
1127 debug(D_HEALTH, "Requested to unlink RRDCALC '%s.%s' which is not linked to any RRDSET", rc->chart?rc->chart:"NOCHART", rc->name);
1128 error("Requested to unlink RRDCALC '%s.%s' which is not linked to any RRDSET", rc->chart?rc->chart:"NOCHART", rc->name);
1133 time_t now = time(NULL);
1134 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);
1137 RRDHOST *host = st->rrdhost;
1139 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);
1143 rc->rrdset_prev->rrdset_next = rc->rrdset_next;
1146 rc->rrdset_next->rrdset_prev = rc->rrdset_prev;
1148 if(st->alarms == rc)
1149 st->alarms = rc->rrdset_next;
1151 rc->rrdset_prev = rc->rrdset_next = NULL;
1153 rrdvar_free(st->rrdhost, &st->variables_root_index, rc->local);
1156 rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rc->family);
1159 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rc->hostid);
1162 rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rc->hostname);
1163 rc->hostname = NULL;
1167 // RRDCALC will remain in RRDHOST
1168 // so that if the matching chart is found in the future
1169 // it will be applied automatically
1172 RRDCALC *rrdcalc_find(RRDSET *st, const char *name) {
1174 uint32_t hash = simple_hash(name);
1176 for( rc = st->alarms; rc ; rc = rc->rrdset_next ) {
1177 if(unlikely(rc->hash == hash && !strcmp(rc->name, name)))
1184 static inline int rrdcalc_exists(RRDHOST *host, const char *chart, const char *name, uint32_t hash_chart, uint32_t hash_name) {
1187 if(unlikely(!chart)) {
1188 error("attempt to find RRDCALC '%s' without giving a chart name", name);
1192 if(unlikely(!hash_chart)) hash_chart = simple_hash(chart);
1193 if(unlikely(!hash_name)) hash_name = simple_hash(name);
1195 // make sure it does not already exist
1196 for(rc = host->alarms; rc ; rc = rc->next) {
1197 if (unlikely(rc->chart && rc->hash == hash_name && rc->hash_chart == hash_chart && !strcmp(name, rc->name) && !strcmp(chart, rc->chart))) {
1198 debug(D_HEALTH, "Health alarm '%s.%s' already exists in host '%s'.", chart, name, host->hostname);
1199 error("Health alarm '%s.%s' already exists in host '%s'.", chart, name, host->hostname);
1207 static inline uint32_t rrdcalc_get_unique_id(RRDHOST *host, const char *chart, const char *name, uint32_t *next_event_id) {
1209 uint32_t hash_chart = simple_hash(chart);
1210 uint32_t hash_name = simple_hash(name);
1212 // re-use old IDs, by looking them up in the alarm log
1214 for(ae = host->health_log.alarms; ae ;ae = ae->next) {
1215 if(unlikely(ae->hash_name == hash_name && ae->hash_chart == hash_chart && !strcmp(name, ae->name) && !strcmp(chart, ae->chart))) {
1216 if(next_event_id) *next_event_id = ae->alarm_event_id + 1;
1217 return ae->alarm_id;
1222 return host->health_log.next_alarm_id++;
1225 static inline void rrdcalc_create_part2(RRDHOST *host, RRDCALC *rc) {
1226 rrdhost_check_rdlock(host);
1228 if(rc->calculation) {
1229 rc->calculation->status = &rc->status;
1230 rc->calculation->this = &rc->value;
1231 rc->calculation->after = &rc->db_after;
1232 rc->calculation->before = &rc->db_before;
1233 rc->calculation->rrdcalc = rc;
1237 rc->warning->status = &rc->status;
1238 rc->warning->this = &rc->value;
1239 rc->warning->after = &rc->db_after;
1240 rc->warning->before = &rc->db_before;
1241 rc->warning->rrdcalc = rc;
1245 rc->critical->status = &rc->status;
1246 rc->critical->this = &rc->value;
1247 rc->critical->after = &rc->db_after;
1248 rc->critical->before = &rc->db_before;
1249 rc->critical->rrdcalc = rc;
1252 // link it to the host
1253 if(likely(host->alarms)) {
1256 for(t = host->alarms; t && t->next ; t = t->next) ;
1263 // link it to its chart
1265 for(st = host->rrdset_root; st ; st = st->next) {
1266 if(rrdcalc_is_matching_this_rrdset(rc, st)) {
1267 rrdsetcalc_link(st, rc);
1273 static inline RRDCALC *rrdcalc_create(RRDHOST *host, RRDCALCTEMPLATE *rt, const char *chart) {
1275 debug(D_HEALTH, "Health creating dynamic alarm (from template) '%s.%s'", chart, rt->name);
1277 if(rrdcalc_exists(host, chart, rt->name, 0, 0))
1280 RRDCALC *rc = callocz(1, sizeof(RRDCALC));
1281 rc->next_event_id = 1;
1282 rc->id = rrdcalc_get_unique_id(host, chart, rt->name, &rc->next_event_id);
1283 rc->name = strdupz(rt->name);
1284 rc->hash = simple_hash(rc->name);
1285 rc->chart = strdupz(chart);
1286 rc->hash_chart = simple_hash(rc->chart);
1288 if(rt->dimensions) rc->dimensions = strdupz(rt->dimensions);
1290 rc->green = rt->green;
1293 rc->old_value = NAN;
1295 rc->delay_up_duration = rt->delay_up_duration;
1296 rc->delay_down_duration = rt->delay_down_duration;
1297 rc->delay_max_duration = rt->delay_max_duration;
1298 rc->delay_multiplier = rt->delay_multiplier;
1300 rc->group = rt->group;
1301 rc->after = rt->after;
1302 rc->before = rt->before;
1303 rc->update_every = rt->update_every;
1304 rc->options = rt->options;
1306 if(rt->exec) rc->exec = strdupz(rt->exec);
1307 if(rt->recipient) rc->recipient = strdupz(rt->recipient);
1308 if(rt->source) rc->source = strdupz(rt->source);
1309 if(rt->units) rc->units = strdupz(rt->units);
1310 if(rt->info) rc->info = strdupz(rt->info);
1312 if(rt->calculation) {
1313 rc->calculation = expression_parse(rt->calculation->source, NULL, NULL);
1314 if(!rc->calculation)
1315 error("Health alarm '%s.%s': failed to parse calculation expression '%s'", chart, rt->name, rt->calculation->source);
1318 rc->warning = expression_parse(rt->warning->source, NULL, NULL);
1320 error("Health alarm '%s.%s': failed to re-parse warning expression '%s'", chart, rt->name, rt->warning->source);
1323 rc->critical = expression_parse(rt->critical->source, NULL, NULL);
1325 error("Health alarm '%s.%s': failed to re-parse critical expression '%s'", chart, rt->name, rt->critical->source);
1328 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",
1329 (rc->chart)?rc->chart:"NOCHART",
1331 (rc->exec)?rc->exec:"DEFAULT",
1332 (rc->recipient)?rc->recipient:"DEFAULT",
1339 (rc->dimensions)?rc->dimensions:"NONE",
1341 (rc->calculation)?rc->calculation->parsed_as:"NONE",
1342 (rc->warning)?rc->warning->parsed_as:"NONE",
1343 (rc->critical)?rc->critical->parsed_as:"NONE",
1345 rc->delay_up_duration,
1346 rc->delay_down_duration,
1347 rc->delay_max_duration,
1348 rc->delay_multiplier
1351 rrdcalc_create_part2(host, rc);
1355 void rrdcalc_free(RRDHOST *host, RRDCALC *rc) {
1358 debug(D_HEALTH, "Health removing alarm '%s.%s' of host '%s'", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);
1360 // unlink it from RRDSET
1361 if(rc->rrdset) rrdsetcalc_unlink(rc);
1363 // unlink it from RRDHOST
1364 if(unlikely(rc == host->alarms))
1365 host->alarms = rc->next;
1367 else if(likely(host->alarms)) {
1368 RRDCALC *t, *last = host->alarms;
1369 for(t = last->next; t && t != rc; last = t, t = t->next) ;
1370 if(last->next == rc)
1371 last->next = rc->next;
1373 error("Cannot unlink alarm '%s.%s' from host '%s': not found", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);
1376 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);
1378 expression_free(rc->calculation);
1379 expression_free(rc->warning);
1380 expression_free(rc->critical);
1385 freez(rc->dimensions);
1387 freez(rc->recipient);
1394 // ----------------------------------------------------------------------------
1395 // RRDCALCTEMPLATE management
1397 void rrdcalctemplate_link_matching(RRDSET *st) {
1398 RRDCALCTEMPLATE *rt;
1400 for(rt = st->rrdhost->templates; rt ; rt = rt->next) {
1401 if(rt->hash_context == st->hash_context && !strcmp(rt->context, st->context)) {
1402 RRDCALC *rc = rrdcalc_create(st->rrdhost, rt, st->id);
1404 error("Health tried to create alarm from template '%s', but it failed", rt->name);
1406 #ifdef NETDATA_INTERNAL_CHECKS
1407 else if(rc->rrdset != st)
1408 error("Health alarm '%s.%s' should be linked to chart '%s', but it is not", rc->chart?rc->chart:"NOCHART", rc->name, st->id);
1414 static inline void rrdcalctemplate_free(RRDHOST *host, RRDCALCTEMPLATE *rt) {
1415 debug(D_HEALTH, "Health removing template '%s' of host '%s'", rt->name, host->hostname);
1417 if(host->templates) {
1418 if(host->templates == rt) {
1419 host->templates = rt->next;
1422 RRDCALCTEMPLATE *t, *last = host->templates;
1423 for (t = last->next; t && t != rt; last = t, t = t->next ) ;
1424 if(last && last->next == rt) {
1425 last->next = rt->next;
1429 error("Cannot find RRDCALCTEMPLATE '%s' linked in host '%s'", rt->name, host->hostname);
1433 expression_free(rt->calculation);
1434 expression_free(rt->warning);
1435 expression_free(rt->critical);
1439 freez(rt->recipient);
1444 freez(rt->dimensions);
1448 // ----------------------------------------------------------------------------
1449 // load health configuration
1451 #define HEALTH_CONF_MAX_LINE 4096
1453 #define HEALTH_ALARM_KEY "alarm"
1454 #define HEALTH_TEMPLATE_KEY "template"
1455 #define HEALTH_ON_KEY "on"
1456 #define HEALTH_LOOKUP_KEY "lookup"
1457 #define HEALTH_CALC_KEY "calc"
1458 #define HEALTH_EVERY_KEY "every"
1459 #define HEALTH_GREEN_KEY "green"
1460 #define HEALTH_RED_KEY "red"
1461 #define HEALTH_WARN_KEY "warn"
1462 #define HEALTH_CRIT_KEY "crit"
1463 #define HEALTH_EXEC_KEY "exec"
1464 #define HEALTH_RECIPIENT_KEY "to"
1465 #define HEALTH_UNITS_KEY "units"
1466 #define HEALTH_INFO_KEY "info"
1467 #define HEALTH_DELAY_KEY "delay"
1469 static inline int rrdcalc_add_alarm_from_config(RRDHOST *host, RRDCALC *rc) {
1471 error("Health configuration for alarm '%s' does not have a chart", rc->name);
1475 if(!rc->update_every) {
1476 error("Health configuration for alarm '%s.%s' has no frequency (parameter 'every'). Ignoring it.", rc->chart?rc->chart:"NOCHART", rc->name);
1480 if(!RRDCALC_HAS_DB_LOOKUP(rc) && !rc->warning && !rc->critical) {
1481 error("Health configuration for alarm '%s.%s' is useless (no calculation, no warning and no critical evaluation)", rc->chart?rc->chart:"NOCHART", rc->name);
1485 if (rrdcalc_exists(host, rc->chart, rc->name, rc->hash_chart, rc->hash))
1488 rc->id = rrdcalc_get_unique_id(&localhost, rc->chart, rc->name, &rc->next_event_id);
1490 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",
1491 rc->chart?rc->chart:"NOCHART",
1494 (rc->exec)?rc->exec:"DEFAULT",
1495 (rc->recipient)?rc->recipient:"DEFAULT",
1502 (rc->dimensions)?rc->dimensions:"NONE",
1504 (rc->calculation)?rc->calculation->parsed_as:"NONE",
1505 (rc->warning)?rc->warning->parsed_as:"NONE",
1506 (rc->critical)?rc->critical->parsed_as:"NONE",
1508 rc->delay_up_duration,
1509 rc->delay_down_duration,
1510 rc->delay_max_duration,
1511 rc->delay_multiplier
1514 rrdcalc_create_part2(host, rc);
1518 static inline int rrdcalctemplate_add_template_from_config(RRDHOST *host, RRDCALCTEMPLATE *rt) {
1519 if(unlikely(!rt->context)) {
1520 error("Health configuration for template '%s' does not have a context", rt->name);
1524 if(unlikely(!rt->update_every)) {
1525 error("Health configuration for template '%s' has no frequency (parameter 'every'). Ignoring it.", rt->name);
1529 if(unlikely(!RRDCALCTEMPLATE_HAS_CALCULATION(rt) && !rt->warning && !rt->critical)) {
1530 error("Health configuration for template '%s' is useless (no calculation, no warning and no critical evaluation)", rt->name);
1534 RRDCALCTEMPLATE *t, *last = NULL;
1535 for (t = host->templates; t ; last = t, t = t->next) {
1536 if(unlikely(t->hash_name == rt->hash_name && !strcmp(t->name, rt->name))) {
1537 error("Health configuration template '%s' already exists for host '%s'.", rt->name, host->hostname);
1542 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",
1544 (rt->context)?rt->context:"NONE",
1545 (rt->exec)?rt->exec:"DEFAULT",
1546 (rt->recipient)?rt->recipient:"DEFAULT",
1553 (rt->dimensions)?rt->dimensions:"NONE",
1555 (rt->calculation)?rt->calculation->parsed_as:"NONE",
1556 (rt->warning)?rt->warning->parsed_as:"NONE",
1557 (rt->critical)?rt->critical->parsed_as:"NONE",
1559 rt->delay_up_duration,
1560 rt->delay_down_duration,
1561 rt->delay_max_duration,
1562 rt->delay_multiplier
1569 rt->next = host->templates;
1570 host->templates = rt;
1576 static inline int health_parse_duration(char *string, int *result) {
1577 // make sure it is a number
1578 if(!*string || !(isdigit(*string) || *string == '+' || *string == '-')) {
1584 calculated_number n = strtold(string, &e);
1588 *result = (int) (n * 86400 * 365);
1591 *result = (int) (n * 86400 * 30);
1594 *result = (int) (n * 86400 * 7);
1597 *result = (int) (n * 86400);
1600 *result = (int) (n * 3600);
1603 *result = (int) (n * 60);
1608 *result = (int) (n);
1618 static inline int health_parse_delay(
1619 size_t line, const char *path, const char *file, char *string,
1620 int *delay_up_duration,
1621 int *delay_down_duration,
1622 int *delay_max_duration,
1623 float *delay_multiplier) {
1626 char given_down = 0;
1628 char given_multiplier = 0;
1634 while(*s && !isspace(*s)) s++;
1635 while(*s && isspace(*s)) *s++ = '\0';
1640 while(*s && !isspace(*s)) s++;
1641 while(*s && isspace(*s)) *s++ = '\0';
1643 if(!strcasecmp(key, "up")) {
1644 if (!health_parse_duration(value, delay_up_duration)) {
1645 error("Health configuration at line %zu of file '%s/%s': invalid value '%s' for '%s' keyword",
1646 line, path, file, value, key);
1650 else if(!strcasecmp(key, "down")) {
1651 if (!health_parse_duration(value, delay_down_duration)) {
1652 error("Health configuration at line %zu of file '%s/%s': invalid value '%s' for '%s' keyword",
1653 line, path, file, value, key);
1655 else given_down = 1;
1657 else if(!strcasecmp(key, "multiplier")) {
1658 *delay_multiplier = strtof(value, NULL);
1659 if(isnan(*delay_multiplier) || isinf(*delay_multiplier) || islessequal(*delay_multiplier, 0)) {
1660 error("Health configuration at line %zu of file '%s/%s': invalid value '%s' for '%s' keyword",
1661 line, path, file, value, key);
1663 else given_multiplier = 1;
1665 else if(!strcasecmp(key, "max")) {
1666 if (!health_parse_duration(value, delay_max_duration)) {
1667 error("Health configuration at line %zu of file '%s/%s': invalid value '%s' for '%s' keyword",
1668 line, path, file, value, key);
1673 error("Health configuration at line %zu of file '%s/%s': unknown keyword '%s'",
1674 line, path, file, key);
1679 *delay_up_duration = 0;
1682 *delay_down_duration = 0;
1684 if(!given_multiplier)
1685 *delay_multiplier = 1.0;
1688 if((*delay_max_duration) < (*delay_up_duration) * (*delay_multiplier))
1689 *delay_max_duration = (*delay_up_duration) * (*delay_multiplier);
1691 if((*delay_max_duration) < (*delay_down_duration) * (*delay_multiplier))
1692 *delay_max_duration = (*delay_down_duration) * (*delay_multiplier);
1698 static inline int health_parse_db_lookup(
1699 size_t line, const char *path, const char *file, char *string,
1700 int *group_method, int *after, int *before, int *every,
1701 uint32_t *options, char **dimensions
1703 debug(D_HEALTH, "Health configuration parsing database lookup %zu@%s/%s: %s", line, path, file, string);
1705 if(*dimensions) freez(*dimensions);
1712 char *s = string, *key;
1714 // first is the group method
1716 while(*s && !isspace(*s)) s++;
1717 while(*s && isspace(*s)) *s++ = '\0';
1719 error("Health configuration invalid chart calculation at line %zu of file '%s/%s': expected group method followed by the 'after' time, but got '%s'",
1720 line, path, file, key);
1724 if((*group_method = web_client_api_request_v1_data_group(key, -1)) == -1) {
1725 error("Health configuration at line %zu of file '%s/%s': invalid group method '%s'",
1726 line, path, file, key);
1730 // then is the 'after' time
1732 while(*s && !isspace(*s)) s++;
1733 while(*s && isspace(*s)) *s++ = '\0';
1735 if(!health_parse_duration(key, after)) {
1736 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' after group method",
1737 line, path, file, key);
1742 *every = abs(*after);
1744 // now we may have optional parameters
1747 while(*s && !isspace(*s)) s++;
1748 while(*s && isspace(*s)) *s++ = '\0';
1751 if(!strcasecmp(key, "at")) {
1753 while(*s && !isspace(*s)) s++;
1754 while(*s && isspace(*s)) *s++ = '\0';
1756 if (!health_parse_duration(value, before)) {
1757 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' for '%s' keyword",
1758 line, path, file, value, key);
1761 else if(!strcasecmp(key, HEALTH_EVERY_KEY)) {
1763 while(*s && !isspace(*s)) s++;
1764 while(*s && isspace(*s)) *s++ = '\0';
1766 if (!health_parse_duration(value, every)) {
1767 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' for '%s' keyword",
1768 line, path, file, value, key);
1771 else if(!strcasecmp(key, "absolute") || !strcasecmp(key, "abs") || !strcasecmp(key, "absolute_sum")) {
1772 *options |= RRDR_OPTION_ABSOLUTE;
1774 else if(!strcasecmp(key, "min2max")) {
1775 *options |= RRDR_OPTION_MIN2MAX;
1777 else if(!strcasecmp(key, "null2zero")) {
1778 *options |= RRDR_OPTION_NULL2ZERO;
1780 else if(!strcasecmp(key, "percentage")) {
1781 *options |= RRDR_OPTION_PERCENTAGE;
1783 else if(!strcasecmp(key, "unaligned")) {
1784 *options |= RRDR_OPTION_NOT_ALIGNED;
1786 else if(!strcasecmp(key, "of")) {
1787 if(*s && strcasecmp(s, "all"))
1788 *dimensions = strdupz(s);
1792 error("Health configuration at line %zu of file '%s/%s': unknown keyword '%s'",
1793 line, path, file, key);
1800 static inline char *tabs2spaces(char *s) {
1803 if(unlikely(*t == '\t')) *t = ' ';
1810 static inline char *health_source_file(size_t line, const char *path, const char *filename) {
1811 char buffer[FILENAME_MAX + 1];
1812 snprintfz(buffer, FILENAME_MAX, "%zu@%s/%s", line, path, filename);
1813 return strdupz(buffer);
1816 static inline void strip_quotes(char *s) {
1818 if(*s == '\'' || *s == '"') *s = ' ';
1823 int health_readfile(const char *path, const char *filename) {
1824 debug(D_HEALTH, "Health configuration reading file '%s/%s'", path, filename);
1826 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;
1827 char buffer[HEALTH_CONF_MAX_LINE + 1];
1829 if(unlikely(!hash_alarm)) {
1830 hash_alarm = simple_uhash(HEALTH_ALARM_KEY);
1831 hash_template = simple_uhash(HEALTH_TEMPLATE_KEY);
1832 hash_on = simple_uhash(HEALTH_ON_KEY);
1833 hash_calc = simple_uhash(HEALTH_CALC_KEY);
1834 hash_lookup = simple_uhash(HEALTH_LOOKUP_KEY);
1835 hash_green = simple_uhash(HEALTH_GREEN_KEY);
1836 hash_red = simple_uhash(HEALTH_RED_KEY);
1837 hash_warn = simple_uhash(HEALTH_WARN_KEY);
1838 hash_crit = simple_uhash(HEALTH_CRIT_KEY);
1839 hash_exec = simple_uhash(HEALTH_EXEC_KEY);
1840 hash_every = simple_uhash(HEALTH_EVERY_KEY);
1841 hash_units = simple_hash(HEALTH_UNITS_KEY);
1842 hash_info = simple_hash(HEALTH_INFO_KEY);
1843 hash_recipient = simple_hash(HEALTH_RECIPIENT_KEY);
1844 hash_delay = simple_uhash(HEALTH_DELAY_KEY);
1847 snprintfz(buffer, HEALTH_CONF_MAX_LINE, "%s/%s", path, filename);
1848 FILE *fp = fopen(buffer, "r");
1850 error("Health configuration cannot read file '%s'.", buffer);
1855 RRDCALCTEMPLATE *rt = NULL;
1857 size_t line = 0, append = 0;
1859 while((s = fgets(&buffer[append], (int)(HEALTH_CONF_MAX_LINE - append), fp)) || append) {
1860 int stop_appending = !s;
1866 if(!stop_appending && s[append - 1] == '\\') {
1867 s[append - 1] = ' ';
1868 append = &s[append] - buffer;
1869 if(append < HEALTH_CONF_MAX_LINE)
1872 error("Health configuration has too long muli-line at line %zu of file '%s/%s'.", line, path, filename);
1878 while(*s && *s != ':') s++;
1880 error("Health configuration has invalid line %zu of file '%s/%s'. It does not contain a ':'. Ignoring it.", line, path, filename);
1888 value = trim(value);
1891 error("Health configuration has invalid line %zu of file '%s/%s'. Keyword is empty. Ignoring it.", line, path, filename);
1896 error("Health configuration has invalid line %zu of file '%s/%s'. value is empty. Ignoring it.", line, path, filename);
1900 uint32_t hash = simple_uhash(key);
1902 if(hash == hash_alarm && !strcasecmp(key, HEALTH_ALARM_KEY)) {
1903 if(rc && !rrdcalc_add_alarm_from_config(&localhost, rc))
1904 rrdcalc_free(&localhost, rc);
1907 if (!rrdcalctemplate_add_template_from_config(&localhost, rt))
1908 rrdcalctemplate_free(&localhost, rt);
1912 rc = callocz(1, sizeof(RRDCALC));
1913 rc->next_event_id = 1;
1914 rc->name = tabs2spaces(strdupz(value));
1915 rc->hash = simple_hash(rc->name);
1916 rc->source = health_source_file(line, path, filename);
1920 rc->old_value = NAN;
1921 rc->delay_multiplier = 1.0;
1923 if(rrdvar_fix_name(rc->name))
1924 error("Health configuration renamed alarm '%s' to '%s'", value, rc->name);
1926 else if(hash == hash_template && !strcasecmp(key, HEALTH_TEMPLATE_KEY)) {
1928 if(!rrdcalc_add_alarm_from_config(&localhost, rc))
1929 rrdcalc_free(&localhost, rc);
1933 if(rt && !rrdcalctemplate_add_template_from_config(&localhost, rt))
1934 rrdcalctemplate_free(&localhost, rt);
1936 rt = callocz(1, sizeof(RRDCALCTEMPLATE));
1937 rt->name = tabs2spaces(strdupz(value));
1938 rt->hash_name = simple_hash(rt->name);
1939 rt->source = health_source_file(line, path, filename);
1942 rt->delay_multiplier = 1.0;
1944 if(rrdvar_fix_name(rt->name))
1945 error("Health configuration renamed template '%s' to '%s'", value, rt->name);
1948 if(hash == hash_on && !strcasecmp(key, HEALTH_ON_KEY)) {
1950 if(strcmp(rc->chart, value))
1951 error("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').",
1952 line, path, filename, rc->name, key, rc->chart, value, value);
1956 rc->chart = tabs2spaces(strdupz(value));
1957 rc->hash_chart = simple_hash(rc->chart);
1959 else if(hash == hash_lookup && !strcasecmp(key, HEALTH_LOOKUP_KEY)) {
1960 health_parse_db_lookup(line, path, filename, value, &rc->group, &rc->after, &rc->before,
1962 &rc->options, &rc->dimensions);
1964 else if(hash == hash_every && !strcasecmp(key, HEALTH_EVERY_KEY)) {
1965 if(!health_parse_duration(value, &rc->update_every))
1966 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' cannot parse duration: '%s'.",
1967 line, path, filename, rc->name, key, value);
1969 else if(hash == hash_green && !strcasecmp(key, HEALTH_GREEN_KEY)) {
1971 rc->green = strtold(value, &e);
1973 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' leaves this string unmatched: '%s'.",
1974 line, path, filename, rc->name, key, e);
1977 else if(hash == hash_red && !strcasecmp(key, HEALTH_RED_KEY)) {
1979 rc->red = strtold(value, &e);
1981 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' leaves this string unmatched: '%s'.",
1982 line, path, filename, rc->name, key, e);
1985 else if(hash == hash_calc && !strcasecmp(key, HEALTH_CALC_KEY)) {
1986 const char *failed_at = NULL;
1988 rc->calculation = expression_parse(value, &failed_at, &error);
1989 if(!rc->calculation) {
1990 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1991 line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
1994 else if(hash == hash_warn && !strcasecmp(key, HEALTH_WARN_KEY)) {
1995 const char *failed_at = NULL;
1997 rc->warning = expression_parse(value, &failed_at, &error);
1999 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
2000 line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
2003 else if(hash == hash_crit && !strcasecmp(key, HEALTH_CRIT_KEY)) {
2004 const char *failed_at = NULL;
2006 rc->critical = expression_parse(value, &failed_at, &error);
2008 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
2009 line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
2012 else if(hash == hash_exec && !strcasecmp(key, HEALTH_EXEC_KEY)) {
2014 if(strcmp(rc->exec, value))
2015 error("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').",
2016 line, path, filename, rc->name, key, rc->exec, value, value);
2020 rc->exec = tabs2spaces(strdupz(value));
2022 else if(hash == hash_recipient && !strcasecmp(key, HEALTH_RECIPIENT_KEY)) {
2024 if(strcmp(rc->recipient, value))
2025 error("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').",
2026 line, path, filename, rc->name, key, rc->recipient, value, value);
2028 freez(rc->recipient);
2030 rc->recipient = tabs2spaces(strdupz(value));
2032 else if(hash == hash_units && !strcasecmp(key, HEALTH_UNITS_KEY)) {
2034 if(strcmp(rc->units, value))
2035 error("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').",
2036 line, path, filename, rc->name, key, rc->units, value, value);
2040 rc->units = tabs2spaces(strdupz(value));
2041 strip_quotes(rc->units);
2043 else if(hash == hash_info && !strcasecmp(key, HEALTH_INFO_KEY)) {
2045 if(strcmp(rc->info, value))
2046 error("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').",
2047 line, path, filename, rc->name, key, rc->info, value, value);
2051 rc->info = tabs2spaces(strdupz(value));
2052 strip_quotes(rc->info);
2054 else if(hash == hash_delay && !strcasecmp(key, HEALTH_DELAY_KEY)) {
2055 health_parse_delay(line, path, filename, value, &rc->delay_up_duration, &rc->delay_down_duration, &rc->delay_max_duration, &rc->delay_multiplier);
2058 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' has unknown key '%s'.",
2059 line, path, filename, rc->name, key);
2063 if(hash == hash_on && !strcasecmp(key, HEALTH_ON_KEY)) {
2065 if(strcmp(rt->context, value))
2066 error("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').",
2067 line, path, filename, rt->name, key, rt->context, value, value);
2071 rt->context = tabs2spaces(strdupz(value));
2072 rt->hash_context = simple_hash(rt->context);
2074 else if(hash == hash_lookup && !strcasecmp(key, HEALTH_LOOKUP_KEY)) {
2075 health_parse_db_lookup(line, path, filename, value, &rt->group, &rt->after, &rt->before,
2077 &rt->options, &rt->dimensions);
2079 else if(hash == hash_every && !strcasecmp(key, HEALTH_EVERY_KEY)) {
2080 if(!health_parse_duration(value, &rt->update_every))
2081 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' cannot parse duration: '%s'.",
2082 line, path, filename, rt->name, key, value);
2084 else if(hash == hash_green && !strcasecmp(key, HEALTH_GREEN_KEY)) {
2086 rt->green = strtold(value, &e);
2088 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' leaves this string unmatched: '%s'.",
2089 line, path, filename, rt->name, key, e);
2092 else if(hash == hash_red && !strcasecmp(key, HEALTH_RED_KEY)) {
2094 rt->red = strtold(value, &e);
2096 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' leaves this string unmatched: '%s'.",
2097 line, path, filename, rt->name, key, e);
2100 else if(hash == hash_calc && !strcasecmp(key, HEALTH_CALC_KEY)) {
2101 const char *failed_at = NULL;
2103 rt->calculation = expression_parse(value, &failed_at, &error);
2104 if(!rt->calculation) {
2105 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
2106 line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
2109 else if(hash == hash_warn && !strcasecmp(key, HEALTH_WARN_KEY)) {
2110 const char *failed_at = NULL;
2112 rt->warning = expression_parse(value, &failed_at, &error);
2114 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
2115 line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
2118 else if(hash == hash_crit && !strcasecmp(key, HEALTH_CRIT_KEY)) {
2119 const char *failed_at = NULL;
2121 rt->critical = expression_parse(value, &failed_at, &error);
2123 error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
2124 line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
2127 else if(hash == hash_exec && !strcasecmp(key, HEALTH_EXEC_KEY)) {
2129 if(strcmp(rt->exec, value))
2130 error("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').",
2131 line, path, filename, rt->name, key, rt->exec, value, value);
2135 rt->exec = tabs2spaces(strdupz(value));
2137 else if(hash == hash_recipient && !strcasecmp(key, HEALTH_RECIPIENT_KEY)) {
2139 if(strcmp(rt->recipient, value))
2140 error("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').",
2141 line, path, filename, rt->name, key, rt->recipient, value, value);
2143 freez(rt->recipient);
2145 rt->recipient = tabs2spaces(strdupz(value));
2147 else if(hash == hash_units && !strcasecmp(key, HEALTH_UNITS_KEY)) {
2149 if(strcmp(rt->units, value))
2150 error("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').",
2151 line, path, filename, rt->name, key, rt->units, value, value);
2155 rt->units = tabs2spaces(strdupz(value));
2156 strip_quotes(rt->units);
2158 else if(hash == hash_info && !strcasecmp(key, HEALTH_INFO_KEY)) {
2160 if(strcmp(rt->info, value))
2161 error("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').",
2162 line, path, filename, rt->name, key, rt->info, value, value);
2166 rt->info = tabs2spaces(strdupz(value));
2167 strip_quotes(rt->info);
2169 else if(hash == hash_delay && !strcasecmp(key, HEALTH_DELAY_KEY)) {
2170 health_parse_delay(line, path, filename, value, &rt->delay_up_duration, &rt->delay_down_duration, &rt->delay_max_duration, &rt->delay_multiplier);
2173 error("Health configuration at line %zu of file '%s/%s' for template '%s' has unknown key '%s'.",
2174 line, path, filename, rt->name, key);
2178 error("Health configuration at line %zu of file '%s/%s' has unknown key '%s'. Expected either '" HEALTH_ALARM_KEY "' or '" HEALTH_TEMPLATE_KEY "'.",
2179 line, path, filename, key);
2183 if(rc && !rrdcalc_add_alarm_from_config(&localhost, rc))
2184 rrdcalc_free(&localhost, rc);
2186 if(rt && !rrdcalctemplate_add_template_from_config(&localhost, rt))
2187 rrdcalctemplate_free(&localhost, rt);
2193 void health_readdir(const char *path) {
2194 size_t pathlen = strlen(path);
2196 debug(D_HEALTH, "Health configuration reading directory '%s'", path);
2198 DIR *dir = opendir(path);
2200 error("Health configuration cannot open directory '%s'.", path);
2204 struct dirent *de = NULL;
2205 while ((de = readdir(dir))) {
2206 size_t len = strlen(de->d_name);
2208 if(de->d_type == DT_DIR
2210 (de->d_name[0] == '.' && de->d_name[1] == '\0')
2211 || (de->d_name[0] == '.' && de->d_name[1] == '.' && de->d_name[2] == '\0')
2213 debug(D_HEALTH, "Ignoring directory '%s'", de->d_name);
2217 else if(de->d_type == DT_DIR) {
2218 char *s = mallocz(pathlen + strlen(de->d_name) + 2);
2221 strcat(s, de->d_name);
2227 else if((de->d_type == DT_LNK || de->d_type == DT_REG || de->d_type == DT_UNKNOWN) &&
2228 len > 5 && !strcmp(&de->d_name[len - 5], ".conf")) {
2229 health_readfile(path, de->d_name);
2232 else debug(D_HEALTH, "Ignoring file '%s'", de->d_name);
2238 static inline char *health_config_dir(void) {
2239 char buffer[FILENAME_MAX + 1];
2240 snprintfz(buffer, FILENAME_MAX, "%s/health.d", config_get("global", "config directory", CONFIG_DIR));
2241 return config_get("health", "health configuration directory", buffer);
2244 void health_init(void) {
2245 debug(D_HEALTH, "Health configuration initializing");
2247 if(!(health_enabled = config_get_boolean("health", "enabled", 1))) {
2248 debug(D_HEALTH, "Health is disabled.");
2252 char *pathname = config_get("health", "health db directory", VARLIB_DIR "/health");
2253 if(mkdir(pathname, 0770) == -1 && errno != EEXIST)
2254 fatal("Cannot create directory '%s'.", pathname);
2256 char filename[FILENAME_MAX + 1];
2257 snprintfz(filename, FILENAME_MAX, "%s/health-log.db", pathname);
2258 health.log_filename = config_get("health", "health db file", filename);
2260 health_alarm_log_load(&localhost);
2261 health_alarm_log_open();
2263 char *path = health_config_dir();
2266 char buffer[FILENAME_MAX + 1];
2267 snprintfz(buffer, FILENAME_MAX, "%s/alarm-notify.sh", config_get("global", "plugins directory", PLUGINS_DIR));
2268 health.health_default_exec = config_get("health", "script to execute on alarm", buffer);
2271 long n = config_get_number("health", "in memory max health log entries", (long)localhost.health_log.max);
2273 error("Health configuration has invalid max log entries %ld. Using default %u", n, localhost.health_log.max);
2274 config_set_number("health", "in memory max health log entries", (long)localhost.health_log.max);
2276 else localhost.health_log.max = (unsigned int)n;
2278 rrdhost_rwlock(&localhost);
2279 health_readdir(path);
2280 rrdhost_unlock(&localhost);
2283 // ----------------------------------------------------------------------------
2286 static inline void health_string2json(BUFFER *wb, const char *prefix, const char *label, const char *value, const char *suffix) {
2288 buffer_sprintf(wb, "%s\"%s\":\"%s\"%s", prefix, label, value, suffix);
2290 buffer_sprintf(wb, "%s\"%s\":null%s", prefix, label, suffix);
2293 static inline void health_alarm_entry2json_nolock(BUFFER *wb, ALARM_ENTRY *ae, RRDHOST *host) {
2294 buffer_sprintf(wb, "\n\t{\n"
2295 "\t\t\"hostname\": \"%s\",\n"
2296 "\t\t\"unique_id\": %u,\n"
2297 "\t\t\"alarm_id\": %u,\n"
2298 "\t\t\"alarm_event_id\": %u,\n"
2299 "\t\t\"name\": \"%s\",\n"
2300 "\t\t\"chart\": \"%s\",\n"
2301 "\t\t\"family\": \"%s\",\n"
2302 "\t\t\"processed\": %s,\n"
2303 "\t\t\"updated\": %s,\n"
2304 "\t\t\"exec_run\": %lu,\n"
2305 "\t\t\"exec_failed\": %s,\n"
2306 "\t\t\"exec\": \"%s\",\n"
2307 "\t\t\"recipient\": \"%s\",\n"
2308 "\t\t\"exec_code\": %d,\n"
2309 "\t\t\"source\": \"%s\",\n"
2310 "\t\t\"units\": \"%s\",\n"
2311 "\t\t\"info\": \"%s\",\n"
2312 "\t\t\"when\": %lu,\n"
2313 "\t\t\"duration\": %lu,\n"
2314 "\t\t\"non_clear_duration\": %lu,\n"
2315 "\t\t\"status\": \"%s\",\n"
2316 "\t\t\"old_status\": \"%s\",\n"
2317 "\t\t\"delay\": %d,\n"
2318 "\t\t\"delay_up_to_timestamp\": %lu,\n"
2319 "\t\t\"updated_by_id\": %u,\n"
2320 "\t\t\"updates_id\": %u,\n",
2328 (ae->flags & HEALTH_ENTRY_FLAG_PROCESSED)?"true":"false",
2329 (ae->flags & HEALTH_ENTRY_FLAG_UPDATED)?"true":"false",
2330 (unsigned long)ae->exec_run_timestamp,
2331 (ae->flags & HEALTH_ENTRY_FLAG_EXEC_FAILED)?"true":"false",
2332 ae->exec?ae->exec:health.health_default_exec,
2333 ae->recipient?ae->recipient:health.health_default_recipient,
2336 ae->units?ae->units:"",
2337 ae->info?ae->info:"",
2338 (unsigned long)ae->when,
2339 (unsigned long)ae->duration,
2340 (unsigned long)ae->non_clear_duration,
2341 rrdcalc_status2string(ae->new_status),
2342 rrdcalc_status2string(ae->old_status),
2344 (unsigned long)ae->delay_up_to_timestamp,
2349 buffer_strcat(wb, "\t\t\"value\":");
2350 buffer_rrd_value(wb, ae->new_value);
2351 buffer_strcat(wb, ",\n");
2353 buffer_strcat(wb, "\t\t\"old_value\":");
2354 buffer_rrd_value(wb, ae->old_value);
2355 buffer_strcat(wb, "\n");
2357 buffer_strcat(wb, "\t}");
2360 void health_alarm_log2json(RRDHOST *host, BUFFER *wb, uint32_t after) {
2361 pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
2363 buffer_strcat(wb, "[");
2365 unsigned int max = host->health_log.max;
2366 unsigned int count = 0;
2368 for(ae = host->health_log.alarms; ae && count < max ; count++, ae = ae->next) {
2369 if(ae->unique_id > after) {
2370 if(likely(count)) buffer_strcat(wb, ",");
2371 health_alarm_entry2json_nolock(wb, ae, host);
2375 buffer_strcat(wb, "\n]\n");
2377 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
2380 static inline void health_rrdcalc2json_nolock(BUFFER *wb, RRDCALC *rc) {
2382 "\t\t\"%s.%s\": {\n"
2383 "\t\t\t\"id\": %lu,\n"
2384 "\t\t\t\"name\": \"%s\",\n"
2385 "\t\t\t\"chart\": \"%s\",\n"
2386 "\t\t\t\"family\": \"%s\",\n"
2387 "\t\t\t\"active\": %s,\n"
2388 "\t\t\t\"exec\": \"%s\",\n"
2389 "\t\t\t\"recipient\": \"%s\",\n"
2390 "\t\t\t\"source\": \"%s\",\n"
2391 "\t\t\t\"units\": \"%s\",\n"
2392 "\t\t\t\"info\": \"%s\",\n"
2393 "\t\t\t\"status\": \"%s\",\n"
2394 "\t\t\t\"last_status_change\": %lu,\n"
2395 "\t\t\t\"last_updated\": %lu,\n"
2396 "\t\t\t\"next_update\": %lu,\n"
2397 "\t\t\t\"update_every\": %d,\n"
2398 "\t\t\t\"delay_up_duration\": %d,\n"
2399 "\t\t\t\"delay_down_duration\": %d,\n"
2400 "\t\t\t\"delay_max_duration\": %d,\n"
2401 "\t\t\t\"delay_multiplier\": %f,\n"
2402 "\t\t\t\"delay\": %d,\n"
2403 "\t\t\t\"delay_up_to_timestamp\": %lu,\n"
2404 , rc->chart, rc->name
2405 , (unsigned long)rc->id
2408 , (rc->rrdset && rc->rrdset->family)?rc->rrdset->family:""
2409 , (rc->rrdset)?"true":"false"
2410 , rc->exec?rc->exec:health.health_default_exec
2411 , rc->recipient?rc->recipient:health.health_default_recipient
2413 , rc->units?rc->units:""
2414 , rc->info?rc->info:""
2415 , rrdcalc_status2string(rc->status)
2416 , (unsigned long)rc->last_status_change
2417 , (unsigned long)rc->last_updated
2418 , (unsigned long)rc->next_update
2420 , rc->delay_up_duration
2421 , rc->delay_down_duration
2422 , rc->delay_max_duration
2423 , rc->delay_multiplier
2425 , (unsigned long)rc->delay_up_to_timestamp
2428 if(RRDCALC_HAS_DB_LOOKUP(rc)) {
2429 if(rc->dimensions && *rc->dimensions)
2430 health_string2json(wb, "\t\t\t", "lookup_dimensions", rc->dimensions, ",\n");
2433 "\t\t\t\"db_after\": %lu,\n"
2434 "\t\t\t\"db_before\": %lu,\n"
2435 "\t\t\t\"lookup_method\": \"%s\",\n"
2436 "\t\t\t\"lookup_after\": %d,\n"
2437 "\t\t\t\"lookup_before\": %d,\n"
2438 "\t\t\t\"lookup_options\": \"",
2439 (unsigned long) rc->db_after,
2440 (unsigned long) rc->db_before,
2441 group_method2string(rc->group),
2445 buffer_data_options2string(wb, rc->options);
2446 buffer_strcat(wb, "\",\n");
2449 if(rc->calculation) {
2450 health_string2json(wb, "\t\t\t", "calc", rc->calculation->source, ",\n");
2451 health_string2json(wb, "\t\t\t", "calc_parsed", rc->calculation->parsed_as, ",\n");
2455 health_string2json(wb, "\t\t\t", "warn", rc->warning->source, ",\n");
2456 health_string2json(wb, "\t\t\t", "warn_parsed", rc->warning->parsed_as, ",\n");
2460 health_string2json(wb, "\t\t\t", "crit", rc->critical->source, ",\n");
2461 health_string2json(wb, "\t\t\t", "crit_parsed", rc->critical->parsed_as, ",\n");
2464 buffer_strcat(wb, "\t\t\t\"green\":");
2465 buffer_rrd_value(wb, rc->green);
2466 buffer_strcat(wb, ",\n");
2468 buffer_strcat(wb, "\t\t\t\"red\":");
2469 buffer_rrd_value(wb, rc->red);
2470 buffer_strcat(wb, ",\n");
2472 buffer_strcat(wb, "\t\t\t\"value\":");
2473 buffer_rrd_value(wb, rc->value);
2474 buffer_strcat(wb, "\n");
2476 buffer_strcat(wb, "\t\t}");
2479 //void health_rrdcalctemplate2json_nolock(BUFFER *wb, RRDCALCTEMPLATE *rt) {
2483 void health_alarms2json(RRDHOST *host, BUFFER *wb, int all) {
2486 rrdhost_rdlock(&localhost);
2487 buffer_sprintf(wb, "{\n\t\"hostname\": \"%s\","
2488 "\n\t\"latest_alarm_log_unique_id\": %u,"
2489 "\n\t\"status\": %s,"
2491 "\n\t\"alarms\": {\n",
2493 (host->health_log.next_log_id > 0)?(host->health_log.next_log_id - 1):0,
2494 health_enabled?"true":"false",
2495 (unsigned long)time(NULL));
2498 for(i = 0, rc = host->alarms; rc ; rc = rc->next) {
2499 if(unlikely(!rc->rrdset || !rc->rrdset->last_collected_time.tv_sec))
2502 if(likely(!all && !(rc->status == RRDCALC_STATUS_WARNING || rc->status == RRDCALC_STATUS_CRITICAL)))
2505 if(likely(i)) buffer_strcat(wb, ",\n");
2506 health_rrdcalc2json_nolock(wb, rc);
2510 // buffer_strcat(wb, "\n\t},\n\t\"templates\": {");
2511 // RRDCALCTEMPLATE *rt;
2512 // for(rt = host->templates; rt ; rt = rt->next)
2513 // health_rrdcalctemplate2json_nolock(wb, rt);
2515 buffer_strcat(wb, "\n\t}\n}\n");
2516 rrdhost_unlock(&localhost);
2520 // ----------------------------------------------------------------------------
2521 // re-load health configuration
2523 static inline void health_free_all_nolock(RRDHOST *host) {
2524 while(host->templates)
2525 rrdcalctemplate_free(host, host->templates);
2528 rrdcalc_free(host, host->alarms);
2531 void health_reload(void) {
2532 if(!health_enabled) {
2533 error("Health reload is requested, but health is not enabled.");
2537 char *path = health_config_dir();
2539 // free all running alarms
2540 rrdhost_rwlock(&localhost);
2541 health_free_all_nolock(&localhost);
2542 rrdhost_unlock(&localhost);
2544 // invalidate all previous entries in the alarm log
2546 for(t = localhost.health_log.alarms ; t ; t = t->next) {
2547 if(t->new_status != RRDCALC_STATUS_REMOVED)
2548 t->flags |= HEALTH_ENTRY_FLAG_UPDATED;
2551 // reset all thresholds to all charts
2553 for(st = localhost.rrdset_root; st ; st = st->next) {
2558 // load the new alarms
2559 rrdhost_rwlock(&localhost);
2560 health_readdir(path);
2561 rrdhost_unlock(&localhost);
2563 // link the loaded alarms to their charts
2564 for(st = localhost.rrdset_root; st ; st = st->next) {
2565 rrdhost_rwlock(&localhost);
2567 rrdsetcalc_link_matching(st);
2568 rrdcalctemplate_link_matching(st);
2570 rrdhost_unlock(&localhost);
2574 // ----------------------------------------------------------------------------
2575 // health main thread and friends
2577 static inline int rrdcalc_value2status(calculated_number n) {
2578 if(isnan(n)) return RRDCALC_STATUS_UNDEFINED;
2579 if(n) return RRDCALC_STATUS_RAISED;
2580 return RRDCALC_STATUS_CLEAR;
2583 static inline void health_alarm_execute(RRDHOST *host, ALARM_ENTRY *ae) {
2584 ae->flags |= HEALTH_ENTRY_FLAG_PROCESSED;
2586 if(unlikely(ae->new_status < RRDCALC_STATUS_CLEAR)) {
2587 // do not send notifications for internal statuses
2591 // find the previous notification for the same alarm
2592 // which we have run the exec script
2594 for(t = ae->next; t ;t = t->next) {
2595 if(t->alarm_id == ae->alarm_id && t->flags & HEALTH_ENTRY_FLAG_EXEC_RUN)
2600 // we have executed this alarm notification in the past
2601 if (t && t->new_status == ae->new_status) {
2602 // don't send the same notification again
2603 debug(D_HEALTH, "Health not sending again notification for alarm '%s.%s' status %s", ae->chart, ae->name,
2604 rrdcalc_status2string(ae->new_status));
2609 // we have not executed this alarm notification in the past
2610 if(unlikely(ae->old_status == RRDCALC_STATUS_UNINITIALIZED && ae->new_status == RRDCALC_STATUS_CLEAR)) {
2611 debug(D_HEALTH, "Health not sending notification for first initialization of alarm '%s.%s' status %s", ae->chart, ae->name, rrdcalc_status2string(ae->new_status));
2616 char buffer[FILENAME_MAX + 1];
2619 const char *exec = ae->exec;
2620 if(!exec) exec = health.health_default_exec;
2622 const char *recipient = ae->recipient;
2623 if(!recipient) recipient = health.health_default_recipient;
2625 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'",
2632 (unsigned long)ae->when,
2634 ae->chart?ae->chart:"NOCAHRT",
2635 ae->family?ae->family:"NOFAMILY",
2636 rrdcalc_status2string(ae->new_status),
2637 rrdcalc_status2string(ae->old_status),
2640 ae->source?ae->source:"UNKNOWN",
2641 (uint32_t)ae->duration,
2642 (uint32_t)ae->non_clear_duration,
2643 ae->units?ae->units:"",
2644 ae->info?ae->info:""
2647 ae->flags |= HEALTH_ENTRY_FLAG_EXEC_RUN;
2648 ae->exec_run_timestamp = time(NULL);
2650 debug(D_HEALTH, "executing command '%s'", buffer);
2651 FILE *fp = mypopen(buffer, &command_pid);
2653 error("HEALTH: Cannot popen(\"%s\", \"r\").", buffer);
2656 debug(D_HEALTH, "HEALTH reading from command");
2657 char *s = fgets(buffer, FILENAME_MAX, fp);
2659 ae->exec_code = mypclose(fp, command_pid);
2660 debug(D_HEALTH, "done executing command - returned with code %d", ae->exec_code);
2662 if(ae->exec_code != 0)
2663 ae->flags |= HEALTH_ENTRY_FLAG_EXEC_FAILED;
2666 health_alarm_log_save(host, ae);
2670 static inline void health_process_notifications(RRDHOST *host, ALARM_ENTRY *ae) {
2671 debug(D_HEALTH, "Health alarm '%s.%s' = %0.2Lf - changed status from %s to %s",
2672 ae->chart?ae->chart:"NOCHART", ae->name,
2674 rrdcalc_status2string(ae->old_status),
2675 rrdcalc_status2string(ae->new_status)
2678 health_alarm_execute(host, ae);
2681 static inline void health_alarm_log_process(RRDHOST *host) {
2682 static uint32_t stop_at_id = 0;
2683 uint32_t first_waiting = (host->health_log.alarms)?host->health_log.alarms->unique_id:0;
2684 time_t now = time(NULL);
2686 pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
2689 for(ae = host->health_log.alarms; ae && ae->unique_id >= stop_at_id ; ae = ae->next) {
2691 !(ae->flags & HEALTH_ENTRY_FLAG_PROCESSED) &&
2692 !(ae->flags & HEALTH_ENTRY_FLAG_UPDATED)
2695 if(unlikely(ae->unique_id < first_waiting))
2696 first_waiting = ae->unique_id;
2698 if(likely(now >= ae->delay_up_to_timestamp))
2699 health_process_notifications(host, ae);
2703 // remember this for the next iteration
2704 stop_at_id = first_waiting;
2706 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
2708 if(host->health_log.count <= host->health_log.max)
2711 // cleanup excess entries in the log
2712 pthread_rwlock_wrlock(&host->health_log.alarm_log_rwlock);
2714 ALARM_ENTRY *last = NULL;
2715 unsigned int count = host->health_log.max * 2 / 3;
2716 for(ae = host->health_log.alarms; ae && count ; count--, last = ae, ae = ae->next) ;
2718 if(ae && last && last->next == ae)
2724 debug(D_HEALTH, "Health removing alarm log entry with id: %u", ae->unique_id);
2726 ALARM_ENTRY *t = ae->next;
2732 freez(ae->recipient);
2739 host->health_log.count--;
2742 pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
2745 static inline int rrdcalc_isrunnable(RRDCALC *rc, time_t now, time_t *next_run) {
2746 if(unlikely(!rc->rrdset)) {
2747 debug(D_HEALTH, "Health not running alarm '%s.%s'. It is not linked to a chart.", rc->chart?rc->chart:"NOCHART", rc->name);
2751 if(unlikely(rc->next_update > now)) {
2752 if (unlikely(*next_run > rc->next_update)) {
2753 // update the next_run time of the main loop
2754 // to run this alarm precisely the time required
2755 *next_run = rc->next_update;
2758 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));
2762 if(unlikely(!rc->update_every)) {
2763 debug(D_HEALTH, "Health not running alarm '%s.%s'. It does not have an update frequency", rc->chart?rc->chart:"NOCHART", rc->name);
2767 if(unlikely(!rc->rrdset->last_collected_time.tv_sec || rc->rrdset->counter_done < 2)) {
2768 debug(D_HEALTH, "Health not running alarm '%s.%s'. Chart is not fully collected yet.", rc->chart?rc->chart:"NOCHART", rc->name);
2772 int update_every = rc->rrdset->update_every;
2773 time_t first = rrdset_first_entry_t(rc->rrdset);
2774 time_t last = rrdset_last_entry_t(rc->rrdset);
2776 if(unlikely(now + update_every < first /* || now - update_every > last */)) {
2778 , "Health not examining alarm '%s.%s' yet (wanted time is out of bounds - we need %lu but got %lu - %lu)."
2779 , rc->chart ? rc->chart : "NOCHART", rc->name, (unsigned long) now, (unsigned long) first
2780 , (unsigned long) last);
2784 if(RRDCALC_HAS_DB_LOOKUP(rc)) {
2785 time_t needed = now + rc->before + rc->after;
2787 if(needed + update_every < first || needed - update_every > last) {
2789 , "Health not examining alarm '%s.%s' yet (not enough data yet - we need %lu but got %lu - %lu)."
2790 , rc->chart ? rc->chart : "NOCHART", rc->name, (unsigned long) needed, (unsigned long) first
2791 , (unsigned long) last);
2799 void *health_main(void *ptr) {
2802 info("HEALTH thread created with task id %d", gettid());
2804 if(pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, NULL) != 0)
2805 error("Cannot set pthread cancel type to DEFERRED.");
2807 if(pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL) != 0)
2808 error("Cannot set pthread cancel state to ENABLE.");
2810 int min_run_every = (int)config_get_number("health", "run at least every seconds", 10);
2811 if(min_run_every < 1) min_run_every = 1;
2813 BUFFER *wb = buffer_create(100);
2815 unsigned int loop = 0;
2816 while(health_enabled && !netdata_exit) {
2818 debug(D_HEALTH, "Health monitoring iteration no %u started", loop);
2820 int oldstate, runnable = 0;
2821 time_t now = time(NULL);
2822 time_t next_run = now + min_run_every;
2825 if(unlikely(pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate) != 0))
2826 error("Cannot set pthread cancel state to DISABLE.");
2828 rrdhost_rdlock(&localhost);
2830 // the first loop is to lookup values from the db
2831 for(rc = localhost.alarms; rc; rc = rc->next) {
2832 if(unlikely(!rrdcalc_isrunnable(rc, now, &next_run))) {
2833 if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_RUNNABLE))
2834 rc->rrdcalc_flags &= ~RRDCALC_FLAG_RUNNABLE;
2839 rc->old_value = rc->value;
2840 rc->rrdcalc_flags |= RRDCALC_FLAG_RUNNABLE;
2842 // 1. if there is database lookup, do it
2843 // 2. if there is calculation expression, run it
2845 if (unlikely(RRDCALC_HAS_DB_LOOKUP(rc))) {
2846 /* time_t old_db_timestamp = rc->db_before; */
2847 int value_is_null = 0;
2849 int ret = rrd2value(rc->rrdset, wb, &rc->value,
2850 rc->dimensions, 1, rc->after, rc->before, rc->group,
2851 rc->options, &rc->db_after, &rc->db_before, &value_is_null);
2853 if (unlikely(ret != 200)) {
2854 // database lookup failed
2857 debug(D_HEALTH, "Health alarm '%s.%s': database lookup returned error %d", rc->chart?rc->chart:"NOCHART", rc->name, ret);
2859 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_ERROR))) {
2860 rc->rrdcalc_flags |= RRDCALC_FLAG_DB_ERROR;
2861 error("Health alarm '%s.%s': database lookup returned error %d", rc->chart?rc->chart:"NOCHART", rc->name, ret);
2864 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_ERROR))
2865 rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_ERROR;
2867 /* - RRDCALC_FLAG_DB_STALE not currently used
2868 if (unlikely(old_db_timestamp == rc->db_before)) {
2869 // database is stale
2871 debug(D_HEALTH, "Health alarm '%s.%s': database is stale", rc->chart?rc->chart:"NOCHART", rc->name);
2873 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_STALE))) {
2874 rc->rrdcalc_flags |= RRDCALC_FLAG_DB_STALE;
2875 error("Health alarm '%s.%s': database is stale", rc->chart?rc->chart:"NOCHART", rc->name);
2878 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_STALE))
2879 rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_STALE;
2882 if (unlikely(value_is_null)) {
2883 // collected value is null
2887 debug(D_HEALTH, "Health alarm '%s.%s': database lookup returned empty value (possibly value is not collected yet)",
2888 rc->chart?rc->chart:"NOCHART", rc->name);
2890 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_NAN))) {
2891 rc->rrdcalc_flags |= RRDCALC_FLAG_DB_NAN;
2892 error("Health alarm '%s.%s': database lookup returned empty value (possibly value is not collected yet)",
2893 rc->chart?rc->chart:"NOCHART", rc->name);
2896 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_NAN))
2897 rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_NAN;
2899 debug(D_HEALTH, "Health alarm '%s.%s': database lookup gave value "
2900 CALCULATED_NUMBER_FORMAT, rc->chart?rc->chart:"NOCHART", rc->name, rc->value);
2903 if(unlikely(rc->calculation)) {
2904 if (unlikely(!expression_evaluate(rc->calculation))) {
2905 // calculation failed
2909 debug(D_HEALTH, "Health alarm '%s.%s': expression '%s' failed: %s",
2910 rc->chart?rc->chart:"NOCHART", rc->name, rc->calculation->parsed_as, buffer_tostring(rc->calculation->error_msg));
2912 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_CALC_ERROR))) {
2913 rc->rrdcalc_flags |= RRDCALC_FLAG_CALC_ERROR;
2914 error("Health alarm '%s.%s': expression '%s' failed: %s",
2915 rc->chart?rc->chart:"NOCHART", rc->name, rc->calculation->parsed_as, buffer_tostring(rc->calculation->error_msg));
2919 if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_CALC_ERROR))
2920 rc->rrdcalc_flags &= ~RRDCALC_FLAG_CALC_ERROR;
2922 debug(D_HEALTH, "Health alarm '%s.%s': expression '%s' gave value "
2923 CALCULATED_NUMBER_FORMAT
2924 ": %s (source: %s)",
2925 rc->chart?rc->chart:"NOCHART", rc->name,
2926 rc->calculation->parsed_as,
2927 rc->calculation->result,
2928 buffer_tostring(rc->calculation->error_msg),
2932 rc->value = rc->calculation->result;
2936 rrdhost_unlock(&localhost);
2938 if(unlikely(runnable && !netdata_exit)) {
2939 rrdhost_rdlock(&localhost);
2941 for(rc = localhost.alarms; rc; rc = rc->next) {
2942 if(unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_RUNNABLE)))
2945 int warning_status = RRDCALC_STATUS_UNDEFINED;
2946 int critical_status = RRDCALC_STATUS_UNDEFINED;
2948 if(likely(rc->warning)) {
2949 if(unlikely(!expression_evaluate(rc->warning))) {
2950 // calculation failed
2952 debug(D_HEALTH, "Health alarm '%s.%s': warning expression failed with error: %s",
2953 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->warning->error_msg));
2955 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_WARN_ERROR))) {
2956 rc->rrdcalc_flags |= RRDCALC_FLAG_WARN_ERROR;
2957 error("Health alarm '%s.%s': warning expression failed with error: %s",
2958 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->warning->error_msg));
2962 if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_WARN_ERROR))
2963 rc->rrdcalc_flags &= ~RRDCALC_FLAG_WARN_ERROR;
2965 debug(D_HEALTH, "Health alarm '%s.%s': warning expression gave value "
2966 CALCULATED_NUMBER_FORMAT
2967 ": %s (source: %s)",
2968 rc->chart?rc->chart:"NOCHART", rc->name,
2969 rc->warning->result,
2970 buffer_tostring(rc->warning->error_msg),
2974 warning_status = rrdcalc_value2status(rc->warning->result);
2978 if(likely(rc->critical)) {
2979 if(unlikely(!expression_evaluate(rc->critical))) {
2980 // calculation failed
2982 debug(D_HEALTH, "Health alarm '%s.%s': critical expression failed with error: %s",
2983 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->critical->error_msg));
2985 if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_CRIT_ERROR))) {
2986 rc->rrdcalc_flags |= RRDCALC_FLAG_CRIT_ERROR;
2987 error("Health alarm '%s.%s': critical expression failed with error: %s",
2988 rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->critical->error_msg));
2992 if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_CRIT_ERROR))
2993 rc->rrdcalc_flags &= ~RRDCALC_FLAG_CRIT_ERROR;
2995 debug(D_HEALTH, "Health alarm '%s.%s': critical expression gave value "
2996 CALCULATED_NUMBER_FORMAT
2997 ": %s (source: %s)",
2998 rc->chart?rc->chart:"NOCHART", rc->name,
2999 rc->critical->result,
3000 buffer_tostring(rc->critical->error_msg),
3004 critical_status = rrdcalc_value2status(rc->critical->result);
3008 int status = RRDCALC_STATUS_UNDEFINED;
3010 switch(warning_status) {
3011 case RRDCALC_STATUS_CLEAR:
3012 status = RRDCALC_STATUS_CLEAR;
3015 case RRDCALC_STATUS_RAISED:
3016 status = RRDCALC_STATUS_WARNING;
3023 switch(critical_status) {
3024 case RRDCALC_STATUS_CLEAR:
3025 if(status == RRDCALC_STATUS_UNDEFINED)
3026 status = RRDCALC_STATUS_CLEAR;
3029 case RRDCALC_STATUS_RAISED:
3030 status = RRDCALC_STATUS_CRITICAL;
3037 if(status != rc->status) {
3040 if(now > rc->delay_up_to_timestamp) {
3041 rc->delay_up_current = rc->delay_up_duration;
3042 rc->delay_down_current = rc->delay_down_duration;
3044 rc->delay_up_to_timestamp = 0;
3047 rc->delay_up_current = (int)(rc->delay_up_current * rc->delay_multiplier);
3048 if(rc->delay_up_current > rc->delay_max_duration) rc->delay_up_current = rc->delay_max_duration;
3050 rc->delay_down_current = (int)(rc->delay_down_current * rc->delay_multiplier);
3051 if(rc->delay_down_current > rc->delay_max_duration) rc->delay_down_current = rc->delay_max_duration;
3054 if(status > rc->status)
3055 delay = rc->delay_up_current;
3057 delay = rc->delay_down_current;
3059 // COMMENTED: because we do need to send raising alarms
3060 // if(now + delay < rc->delay_up_to_timestamp)
3061 // delay = (int)(rc->delay_up_to_timestamp - now);
3063 rc->delay_last = delay;
3064 rc->delay_up_to_timestamp = now + delay;
3065 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);
3066 rc->last_status_change = now;
3067 rc->status = status;
3070 rc->last_updated = now;
3071 rc->next_update = now + rc->update_every;
3073 if (next_run > rc->next_update)
3074 next_run = rc->next_update;
3077 rrdhost_unlock(&localhost);
3080 if (unlikely(pthread_setcancelstate(oldstate, NULL) != 0))
3081 error("Cannot set pthread cancel state to RESTORE (%d).", oldstate);
3083 if(unlikely(netdata_exit))
3086 // execute notifications
3088 health_alarm_log_process(&localhost);
3090 if(unlikely(netdata_exit))
3094 if(now < next_run) {
3095 debug(D_HEALTH, "Health monitoring iteration no %u done. Next iteration in %d secs",
3096 loop, (int) (next_run - now));
3097 sleep_usec(1000000 * (unsigned long long) (next_run - now));
3100 debug(D_HEALTH, "Health monitoring iteration no %u done. Next iteration now", loop);
3106 info("HEALTH thread exiting");