rrdcalc.c 30 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845
  1. // SPDX-License-Identifier: GPL-3.0-or-later
  2. #include "rrd.h"
  3. // ----------------------------------------------------------------------------
  4. // RRDCALC helpers
  5. void rrdcalc_flags_to_json_array(BUFFER *wb, const char *key, RRDCALC_FLAGS flags) {
  6. buffer_json_member_add_array(wb, key);
  7. if(flags & RRDCALC_FLAG_DB_ERROR)
  8. buffer_json_add_array_item_string(wb, "DB_ERROR");
  9. if(flags & RRDCALC_FLAG_DB_NAN)
  10. buffer_json_add_array_item_string(wb, "DB_NAN");
  11. if(flags & RRDCALC_FLAG_CALC_ERROR)
  12. buffer_json_add_array_item_string(wb, "CALC_ERROR");
  13. if(flags & RRDCALC_FLAG_WARN_ERROR)
  14. buffer_json_add_array_item_string(wb, "WARN_ERROR");
  15. if(flags & RRDCALC_FLAG_CRIT_ERROR)
  16. buffer_json_add_array_item_string(wb, "CRIT_ERROR");
  17. if(flags & RRDCALC_FLAG_RUNNABLE)
  18. buffer_json_add_array_item_string(wb, "RUNNABLE");
  19. if(flags & RRDCALC_FLAG_DISABLED)
  20. buffer_json_add_array_item_string(wb, "DISABLED");
  21. if(flags & RRDCALC_FLAG_SILENCED)
  22. buffer_json_add_array_item_string(wb, "SILENCED");
  23. if(flags & RRDCALC_FLAG_RUN_ONCE)
  24. buffer_json_add_array_item_string(wb, "RUN_ONCE");
  25. if(flags & RRDCALC_FLAG_FROM_TEMPLATE)
  26. buffer_json_add_array_item_string(wb, "FROM_TEMPLATE");
  27. buffer_json_array_close(wb);
  28. }
  29. inline const char *rrdcalc_status2string(RRDCALC_STATUS status) {
  30. switch(status) {
  31. case RRDCALC_STATUS_REMOVED:
  32. return "REMOVED";
  33. case RRDCALC_STATUS_UNDEFINED:
  34. return "UNDEFINED";
  35. case RRDCALC_STATUS_UNINITIALIZED:
  36. return "UNINITIALIZED";
  37. case RRDCALC_STATUS_CLEAR:
  38. return "CLEAR";
  39. case RRDCALC_STATUS_RAISED:
  40. return "RAISED";
  41. case RRDCALC_STATUS_WARNING:
  42. return "WARNING";
  43. case RRDCALC_STATUS_CRITICAL:
  44. return "CRITICAL";
  45. default:
  46. netdata_log_error("Unknown alarm status %d", status);
  47. return "UNKNOWN";
  48. }
  49. }
  50. uint32_t rrdcalc_get_unique_id(RRDHOST *host, STRING *chart, STRING *name, uint32_t *next_event_id, uuid_t *config_hash_id) {
  51. rw_spinlock_read_lock(&host->health_log.spinlock);
  52. // re-use old IDs, by looking them up in the alarm log
  53. ALARM_ENTRY *ae = NULL;
  54. for(ae = host->health_log.alarms; ae ;ae = ae->next) {
  55. if(unlikely(name == ae->name && chart == ae->chart && !uuid_memcmp(&ae->config_hash_id, config_hash_id))) {
  56. if(next_event_id) *next_event_id = ae->alarm_event_id + 1;
  57. break;
  58. }
  59. }
  60. uint32_t alarm_id;
  61. if(ae)
  62. alarm_id = ae->alarm_id;
  63. else {
  64. alarm_id = sql_get_alarm_id(host, chart, name, next_event_id, config_hash_id);
  65. if (!alarm_id) {
  66. if (unlikely(!host->health_log.next_alarm_id))
  67. host->health_log.next_alarm_id = (uint32_t)now_realtime_sec();
  68. alarm_id = host->health_log.next_alarm_id++;
  69. }
  70. }
  71. rw_spinlock_read_unlock(&host->health_log.spinlock);
  72. return alarm_id;
  73. }
  74. // ----------------------------------------------------------------------------
  75. // RRDCALC replacing info text variables with RRDSET labels
  76. static STRING *rrdcalc_replace_variables_with_rrdset_labels(const char *line, RRDCALC *rc) {
  77. if (!line || !*line)
  78. return NULL;
  79. size_t pos = 0;
  80. char *temp = strdupz(line);
  81. char var[RRDCALC_VAR_MAX];
  82. char *m, *lbl_value = NULL;
  83. while ((m = strchr(temp + pos, '$')) && *(m+1) == '{') {
  84. int i = 0;
  85. char *e = m;
  86. while (*e) {
  87. var[i++] = *e;
  88. if (*e == '}' || i == RRDCALC_VAR_MAX - 1)
  89. break;
  90. e++;
  91. }
  92. var[i] = '\0';
  93. pos = m - temp + 1;
  94. if (!strcmp(var, RRDCALC_VAR_FAMILY)) {
  95. char *buf = find_and_replace(temp, var, (rc->rrdset && rc->rrdset->family) ? rrdset_family(rc->rrdset) : "", m);
  96. freez(temp);
  97. temp = buf;
  98. }
  99. else if (!strncmp(var, RRDCALC_VAR_LABEL, RRDCALC_VAR_LABEL_LEN)) {
  100. char label_val[RRDCALC_VAR_MAX + RRDCALC_VAR_LABEL_LEN + 1] = { 0 };
  101. strcpy(label_val, var+RRDCALC_VAR_LABEL_LEN);
  102. label_val[i - RRDCALC_VAR_LABEL_LEN - 1] = '\0';
  103. if(likely(rc->rrdset && rc->rrdset->rrdlabels)) {
  104. rrdlabels_get_value_strdup_or_null(rc->rrdset->rrdlabels, &lbl_value, label_val);
  105. if (lbl_value) {
  106. char *buf = find_and_replace(temp, var, lbl_value, m);
  107. freez(temp);
  108. temp = buf;
  109. freez(lbl_value);
  110. }
  111. }
  112. }
  113. }
  114. STRING *ret = string_strdupz(temp);
  115. freez(temp);
  116. return ret;
  117. }
  118. void rrdcalc_update_info_using_rrdset_labels(RRDCALC *rc) {
  119. if(!rc->rrdset || !rc->original_info || !rc->rrdset->rrdlabels) return;
  120. size_t labels_version = dictionary_version(rc->rrdset->rrdlabels);
  121. if(rc->labels_version != labels_version) {
  122. STRING *old = rc->info;
  123. rc->info = rrdcalc_replace_variables_with_rrdset_labels(rrdcalc_original_info(rc), rc);
  124. string_freez(old);
  125. rc->labels_version = labels_version;
  126. }
  127. }
  128. // ----------------------------------------------------------------------------
  129. // RRDCALC index management for RRDSET
  130. // the dictionary requires a unique key for every item
  131. // we use {chart id}.{alert name} for both the RRDHOST and RRDSET alert indexes.
  132. #define RRDCALC_MAX_KEY_SIZE 1024
  133. static size_t rrdcalc_key(char *dst, size_t dst_len, const char *chart, const char *alert) {
  134. return snprintfz(dst, dst_len, "%s/%s", chart, alert);
  135. }
  136. const RRDCALC_ACQUIRED *rrdcalc_from_rrdset_get(RRDSET *st, const char *alert_name) {
  137. char key[RRDCALC_MAX_KEY_SIZE + 1];
  138. size_t key_len = rrdcalc_key(key, RRDCALC_MAX_KEY_SIZE, rrdset_id(st), alert_name);
  139. const RRDCALC_ACQUIRED *rca = (const RRDCALC_ACQUIRED *)dictionary_get_and_acquire_item_advanced(st->rrdhost->rrdcalc_root_index, key, (ssize_t)(key_len + 1));
  140. if(!rca) {
  141. key_len = rrdcalc_key(key, RRDCALC_MAX_KEY_SIZE, rrdset_name(st), alert_name);
  142. rca = (const RRDCALC_ACQUIRED *)dictionary_get_and_acquire_item_advanced(st->rrdhost->rrdcalc_root_index, key, (ssize_t)(key_len + 1));
  143. }
  144. return rca;
  145. }
  146. void rrdcalc_from_rrdset_release(RRDSET *st, const RRDCALC_ACQUIRED *rca) {
  147. if(!rca) return;
  148. dictionary_acquired_item_release(st->rrdhost->rrdcalc_root_index, (const DICTIONARY_ITEM *)rca);
  149. }
  150. RRDCALC *rrdcalc_acquired_to_rrdcalc(const RRDCALC_ACQUIRED *rca) {
  151. if(rca)
  152. return dictionary_acquired_item_value((const DICTIONARY_ITEM *)rca);
  153. return NULL;
  154. }
  155. // ----------------------------------------------------------------------------
  156. // RRDCALC managing the linking with RRDSET
  157. static void rrdcalc_link_to_rrdset(RRDSET *st, RRDCALC *rc) {
  158. RRDHOST *host = st->rrdhost;
  159. netdata_log_debug(D_HEALTH, "Health linking alarm '%s.%s' to chart '%s' of host '%s'", rrdcalc_chart_name(rc), rrdcalc_name(rc), rrdset_id(st), rrdhost_hostname(host));
  160. rc->last_status_change_value = rc->value;
  161. rc->last_status_change = now_realtime_sec();
  162. rc->rrdset = st;
  163. rw_spinlock_write_lock(&st->alerts.spinlock);
  164. DOUBLE_LINKED_LIST_APPEND_ITEM_UNSAFE(st->alerts.base, rc, prev, next);
  165. rw_spinlock_write_unlock(&st->alerts.spinlock);
  166. if(rc->update_every < rc->rrdset->update_every) {
  167. netdata_log_error("Health alarm '%s.%s' has update every %d, less than chart update every %d. Setting alarm update frequency to %d.", rrdset_id(rc->rrdset), rrdcalc_name(rc), rc->update_every, rc->rrdset->update_every, rc->rrdset->update_every);
  168. rc->update_every = rc->rrdset->update_every;
  169. }
  170. if(!isnan(rc->green) && isnan(st->green)) {
  171. netdata_log_debug(D_HEALTH, "Health alarm '%s.%s' green threshold set from " NETDATA_DOUBLE_FORMAT_AUTO
  172. " to " NETDATA_DOUBLE_FORMAT_AUTO ".", rrdset_id(rc->rrdset), rrdcalc_name(rc), rc->rrdset->green, rc->green);
  173. st->green = rc->green;
  174. }
  175. if(!isnan(rc->red) && isnan(st->red)) {
  176. netdata_log_debug(D_HEALTH, "Health alarm '%s.%s' red threshold set from " NETDATA_DOUBLE_FORMAT_AUTO " to " NETDATA_DOUBLE_FORMAT_AUTO
  177. ".", rrdset_id(rc->rrdset), rrdcalc_name(rc), rc->rrdset->red, rc->red);
  178. st->red = rc->red;
  179. }
  180. char buf[RRDVAR_MAX_LENGTH + 1];
  181. snprintfz(buf, RRDVAR_MAX_LENGTH, "%s.%s", rrdset_name(st), rrdcalc_name(rc));
  182. STRING *rrdset_name_rrdcalc_name = string_strdupz(buf);
  183. snprintfz(buf, RRDVAR_MAX_LENGTH, "%s.%s", rrdset_id(st), rrdcalc_name(rc));
  184. STRING *rrdset_id_rrdcalc_name = string_strdupz(buf);
  185. rc->rrdvar_local = rrdvar_add_and_acquire(
  186. "local",
  187. st->rrdvars,
  188. rc->name,
  189. RRDVAR_TYPE_CALCULATED,
  190. RRDVAR_FLAG_RRDCALC_LOCAL_VAR,
  191. &rc->value);
  192. rc->rrdvar_family = rrdvar_add_and_acquire(
  193. "family",
  194. rrdfamily_rrdvars_dict(st->rrdfamily),
  195. rc->name,
  196. RRDVAR_TYPE_CALCULATED,
  197. RRDVAR_FLAG_RRDCALC_FAMILY_VAR,
  198. &rc->value);
  199. rc->rrdvar_host_chart_name = rrdvar_add_and_acquire(
  200. "host",
  201. host->rrdvars,
  202. rrdset_name_rrdcalc_name,
  203. RRDVAR_TYPE_CALCULATED,
  204. RRDVAR_FLAG_RRDCALC_HOST_CHARTNAME_VAR,
  205. &rc->value);
  206. rc->rrdvar_host_chart_id = rrdvar_add_and_acquire(
  207. "host",
  208. host->rrdvars,
  209. rrdset_id_rrdcalc_name,
  210. RRDVAR_TYPE_CALCULATED,
  211. RRDVAR_FLAG_RRDCALC_HOST_CHARTID_VAR | ((rc->rrdvar_host_chart_name) ? 0 : RRDVAR_FLAG_RRDCALC_HOST_CHARTNAME_VAR),
  212. &rc->value);
  213. string_freez(rrdset_id_rrdcalc_name);
  214. string_freez(rrdset_name_rrdcalc_name);
  215. if(!rc->units)
  216. rc->units = string_dup(st->units);
  217. rrdvar_store_for_chart(host, st);
  218. rrdcalc_update_info_using_rrdset_labels(rc);
  219. time_t now = now_realtime_sec();
  220. ALARM_ENTRY *ae = health_create_alarm_entry(
  221. host,
  222. rc->id,
  223. rc->next_event_id++,
  224. rc->config_hash_id,
  225. now,
  226. rc->name,
  227. rc->rrdset->id,
  228. rc->rrdset->context,
  229. rc->rrdset->name,
  230. rc->rrdset->family,
  231. rc->classification,
  232. rc->component,
  233. rc->type,
  234. rc->exec,
  235. rc->recipient,
  236. now - rc->last_status_change,
  237. rc->old_value,
  238. rc->value,
  239. RRDCALC_STATUS_REMOVED,
  240. rc->status,
  241. rc->source,
  242. rc->units,
  243. rc->info,
  244. 0,
  245. rrdcalc_isrepeating(rc)?HEALTH_ENTRY_FLAG_IS_REPEATING:0);
  246. rc->ae = ae;
  247. health_alarm_log_add_entry(host, ae);
  248. }
  249. static void rrdcalc_unlink_from_rrdset(RRDCALC *rc, bool having_ll_wrlock) {
  250. RRDSET *st = rc->rrdset;
  251. if(!st) {
  252. netdata_log_debug(D_HEALTH, "Requested to unlink RRDCALC '%s.%s' which is not linked to any RRDSET", rrdcalc_chart_name(rc), rrdcalc_name(rc));
  253. netdata_log_error("Requested to unlink RRDCALC '%s.%s' which is not linked to any RRDSET", rrdcalc_chart_name(rc), rrdcalc_name(rc));
  254. return;
  255. }
  256. RRDHOST *host = st->rrdhost;
  257. time_t now = now_realtime_sec();
  258. if (likely(rc->status != RRDCALC_STATUS_REMOVED)) {
  259. ALARM_ENTRY *ae = health_create_alarm_entry(
  260. host,
  261. rc->id,
  262. rc->next_event_id++,
  263. rc->config_hash_id,
  264. now,
  265. rc->name,
  266. rc->rrdset->id,
  267. rc->rrdset->context,
  268. rc->rrdset->name,
  269. rc->rrdset->family,
  270. rc->classification,
  271. rc->component,
  272. rc->type,
  273. rc->exec,
  274. rc->recipient,
  275. now - rc->last_status_change,
  276. rc->old_value,
  277. rc->value,
  278. rc->status,
  279. RRDCALC_STATUS_REMOVED,
  280. rc->source,
  281. rc->units,
  282. rc->info,
  283. 0,
  284. 0);
  285. rc->ae = ae;
  286. health_alarm_log_add_entry(host, ae);
  287. }
  288. netdata_log_debug(D_HEALTH, "Health unlinking alarm '%s.%s' from chart '%s' of host '%s'", rrdcalc_chart_name(rc), rrdcalc_name(rc), rrdset_id(st), rrdhost_hostname(host));
  289. // unlink it
  290. if(!having_ll_wrlock)
  291. rw_spinlock_write_lock(&st->alerts.spinlock);
  292. DOUBLE_LINKED_LIST_REMOVE_ITEM_UNSAFE(st->alerts.base, rc, prev, next);
  293. if(!having_ll_wrlock)
  294. rw_spinlock_write_unlock(&st->alerts.spinlock);
  295. rc->rrdset = NULL;
  296. rrdvar_release_and_del(st->rrdvars, rc->rrdvar_local);
  297. rc->rrdvar_local = NULL;
  298. rrdvar_release_and_del(rrdfamily_rrdvars_dict(st->rrdfamily), rc->rrdvar_family);
  299. rc->rrdvar_family = NULL;
  300. rrdvar_release_and_del(host->rrdvars, rc->rrdvar_host_chart_id);
  301. rc->rrdvar_host_chart_id = NULL;
  302. rrdvar_release_and_del(host->rrdvars, rc->rrdvar_host_chart_name);
  303. rc->rrdvar_host_chart_name = NULL;
  304. // RRDCALC will remain in RRDHOST
  305. // so that if the matching chart is found in the future
  306. // it will be applied automatically
  307. }
  308. static inline bool rrdcalc_check_if_it_matches_rrdset(RRDCALC *rc, RRDSET *st) {
  309. if ( (rc->chart != st->id)
  310. && (rc->chart != st->name))
  311. return false;
  312. if (rc->module_pattern && !simple_pattern_matches_string(rc->module_pattern, st->module_name))
  313. return false;
  314. if (rc->plugin_pattern && !simple_pattern_matches_string(rc->plugin_pattern, st->module_name))
  315. return false;
  316. if (st->rrdhost->rrdlabels && rc->host_labels_pattern && !rrdlabels_match_simple_pattern_parsed(
  317. st->rrdhost->rrdlabels, rc->host_labels_pattern, '=', NULL))
  318. return false;
  319. if (st->rrdlabels && rc->chart_labels_pattern && !rrdlabels_match_simple_pattern_parsed(
  320. st->rrdlabels, rc->chart_labels_pattern, '=', NULL))
  321. return false;
  322. return true;
  323. }
  324. void rrdcalc_link_matching_alerts_to_rrdset(RRDSET *st) {
  325. RRDHOST *host = st->rrdhost;
  326. // netdata_log_debug(D_HEALTH, "find matching alarms for chart '%s'", st->id);
  327. RRDCALC *rc;
  328. foreach_rrdcalc_in_rrdhost_read(host, rc) {
  329. if(rc->rrdset)
  330. continue;
  331. if(unlikely(rrdcalc_check_if_it_matches_rrdset(rc, st)))
  332. rrdcalc_link_to_rrdset(st, rc);
  333. }
  334. foreach_rrdcalc_in_rrdhost_done(rc);
  335. }
  336. static inline int rrdcalc_check_and_link_rrdset_callback(RRDSET *st, void *rrdcalc) {
  337. RRDCALC *rc = rrdcalc;
  338. if(unlikely(rrdcalc_check_if_it_matches_rrdset(rc, st))) {
  339. rrdcalc_link_to_rrdset(st, rc);
  340. return -1;
  341. }
  342. return 0;
  343. }
  344. // ----------------------------------------------------------------------------
  345. // RRDCALC rrdhost index management - constructor
  346. struct rrdcalc_constructor {
  347. RRDHOST *rrdhost; // the host we operate upon
  348. RRDCALC *from_config; // points to the original RRDCALC, as loaded from the config
  349. RRDCALCTEMPLATE *from_rrdcalctemplate; // the template this alert is generated from
  350. RRDSET *rrdset; // when this comes from rrdcalctemplate, we have a matching rrdset
  351. const char *overwrite_alert_name; // when we have a dimension foreach, the alert is renamed
  352. const char *overwrite_dimensions; // when we have a dimension foreach, the dimensions filter is renamed
  353. enum {
  354. RRDCALC_REACT_NONE,
  355. RRDCALC_REACT_NEW,
  356. } react_action;
  357. bool existing_from_template;
  358. };
  359. static void rrdcalc_rrdhost_insert_callback(const DICTIONARY_ITEM *item __maybe_unused, void *rrdcalc, void *constructor_data) {
  360. RRDCALC *rc = rrdcalc;
  361. struct rrdcalc_constructor *ctr = constructor_data;
  362. RRDHOST *host = ctr->rrdhost;
  363. rc->key = string_strdupz(dictionary_acquired_item_name(item));
  364. if(ctr->from_rrdcalctemplate) {
  365. rc->run_flags |= RRDCALC_FLAG_FROM_TEMPLATE;
  366. RRDCALCTEMPLATE *rt = ctr->from_rrdcalctemplate;
  367. RRDSET *st = ctr->rrdset;
  368. rc->next_event_id = 1;
  369. rc->name = (ctr->overwrite_alert_name) ? string_strdupz(ctr->overwrite_alert_name) : string_dup(rt->name);
  370. rc->chart = string_dup(st->id);
  371. uuid_copy(rc->config_hash_id, rt->config_hash_id);
  372. rc->dimensions = (ctr->overwrite_dimensions) ? string_strdupz(ctr->overwrite_dimensions) : string_dup(rt->dimensions);
  373. rc->foreach_dimension = NULL;
  374. rc->foreach_dimension_pattern = NULL;
  375. rc->green = rt->green;
  376. rc->red = rt->red;
  377. rc->value = NAN;
  378. rc->old_value = NAN;
  379. rc->delay_up_duration = rt->delay_up_duration;
  380. rc->delay_down_duration = rt->delay_down_duration;
  381. rc->delay_max_duration = rt->delay_max_duration;
  382. rc->delay_multiplier = rt->delay_multiplier;
  383. rc->last_repeat = 0;
  384. rc->times_repeat = 0;
  385. rc->warn_repeat_every = rt->warn_repeat_every;
  386. rc->crit_repeat_every = rt->crit_repeat_every;
  387. rc->group = rt->group;
  388. rc->after = rt->after;
  389. rc->before = rt->before;
  390. rc->update_every = rt->update_every;
  391. rc->options = rt->options;
  392. rc->exec = string_dup(rt->exec);
  393. rc->recipient = string_dup(rt->recipient);
  394. rc->source = string_dup(rt->source);
  395. rc->units = string_dup(rt->units);
  396. rc->info = string_dup(rt->info);
  397. rc->original_info = string_dup(rt->info);
  398. rc->classification = string_dup(rt->classification);
  399. rc->component = string_dup(rt->component);
  400. rc->type = string_dup(rt->type);
  401. if(rt->calculation) {
  402. rc->calculation = expression_parse(rt->calculation->source, NULL, NULL);
  403. if(!rc->calculation)
  404. netdata_log_error("Health alarm '%s.%s': failed to parse calculation expression '%s'", rrdset_id(st), rrdcalctemplate_name(rt), rt->calculation->source);
  405. }
  406. if(rt->warning) {
  407. rc->warning = expression_parse(rt->warning->source, NULL, NULL);
  408. if(!rc->warning)
  409. netdata_log_error("Health alarm '%s.%s': failed to re-parse warning expression '%s'", rrdset_id(st), rrdcalctemplate_name(rt), rt->warning->source);
  410. }
  411. if(rt->critical) {
  412. rc->critical = expression_parse(rt->critical->source, NULL, NULL);
  413. if(!rc->critical)
  414. netdata_log_error("Health alarm '%s.%s': failed to re-parse critical expression '%s'", rrdset_id(st), rrdcalctemplate_name(rt), rt->critical->source);
  415. }
  416. }
  417. else if(ctr->from_config) {
  418. // dictionary has already copied all the members values and pointers
  419. // no need for additional work in this case
  420. ;
  421. }
  422. rc->id = rrdcalc_get_unique_id(host, rc->chart, rc->name, &rc->next_event_id, &rc->config_hash_id);
  423. if(rc->calculation) {
  424. rc->calculation->status = &rc->status;
  425. rc->calculation->myself = &rc->value;
  426. rc->calculation->after = &rc->db_after;
  427. rc->calculation->before = &rc->db_before;
  428. rc->calculation->rrdcalc = rc;
  429. }
  430. if(rc->warning) {
  431. rc->warning->status = &rc->status;
  432. rc->warning->myself = &rc->value;
  433. rc->warning->after = &rc->db_after;
  434. rc->warning->before = &rc->db_before;
  435. rc->warning->rrdcalc = rc;
  436. }
  437. if(rc->critical) {
  438. rc->critical->status = &rc->status;
  439. rc->critical->myself = &rc->value;
  440. rc->critical->after = &rc->db_after;
  441. rc->critical->before = &rc->db_before;
  442. rc->critical->rrdcalc = rc;
  443. }
  444. netdata_log_debug(D_HEALTH, "Health added alarm '%s.%s': exec '%s', recipient '%s', green " NETDATA_DOUBLE_FORMAT_AUTO
  445. ", red " NETDATA_DOUBLE_FORMAT_AUTO
  446. ", lookup: group %d, after %d, before %d, options %u, dimensions '%s', for each dimension '%s', update every %d, calculation '%s', warning '%s', critical '%s', source '%s', delay up %d, delay down %d, delay max %d, delay_multiplier %f, warn_repeat_every %u, crit_repeat_every %u",
  447. rrdcalc_chart_name(rc),
  448. rrdcalc_name(rc),
  449. (rc->exec)?rrdcalc_exec(rc):"DEFAULT",
  450. (rc->recipient)?rrdcalc_recipient(rc):"DEFAULT",
  451. rc->green,
  452. rc->red,
  453. (int)rc->group,
  454. rc->after,
  455. rc->before,
  456. rc->options,
  457. (rc->dimensions)?rrdcalc_dimensions(rc):"NONE",
  458. (rc->foreach_dimension)?rrdcalc_foreachdim(rc):"NONE",
  459. rc->update_every,
  460. (rc->calculation)?rc->calculation->parsed_as:"NONE",
  461. (rc->warning)?rc->warning->parsed_as:"NONE",
  462. (rc->critical)?rc->critical->parsed_as:"NONE",
  463. rrdcalc_source(rc),
  464. rc->delay_up_duration,
  465. rc->delay_down_duration,
  466. rc->delay_max_duration,
  467. rc->delay_multiplier,
  468. rc->warn_repeat_every,
  469. rc->crit_repeat_every
  470. );
  471. ctr->react_action = RRDCALC_REACT_NEW;
  472. }
  473. static bool rrdcalc_rrdhost_conflict_callback(const DICTIONARY_ITEM *item __maybe_unused, void *rrdcalc, void *rrdcalc_new __maybe_unused, void *constructor_data ) {
  474. RRDCALC *rc = rrdcalc;
  475. struct rrdcalc_constructor *ctr = constructor_data;
  476. if(rc->run_flags & RRDCALC_FLAG_FROM_TEMPLATE)
  477. ctr->existing_from_template = true;
  478. else
  479. ctr->existing_from_template = false;
  480. ctr->react_action = RRDCALC_REACT_NONE;
  481. return false;
  482. }
  483. static void rrdcalc_rrdhost_react_callback(const DICTIONARY_ITEM *item __maybe_unused, void *rrdcalc, void *constructor_data) {
  484. RRDCALC *rc = rrdcalc;
  485. struct rrdcalc_constructor *ctr = constructor_data;
  486. RRDHOST *host = ctr->rrdhost;
  487. if(ctr->react_action == RRDCALC_REACT_NEW) {
  488. if(ctr->rrdset)
  489. rrdcalc_link_to_rrdset(ctr->rrdset, rc);
  490. else if (ctr->from_rrdcalctemplate)
  491. rrdcontext_foreach_instance_with_rrdset_in_context(host, string2str(ctr->from_rrdcalctemplate->context), rrdcalc_check_and_link_rrdset_callback, rc);
  492. }
  493. }
  494. // ----------------------------------------------------------------------------
  495. // RRDCALC rrdhost index management - destructor
  496. static void rrdcalc_free_internals(RRDCALC *rc) {
  497. if(unlikely(!rc)) return;
  498. expression_free(rc->calculation);
  499. expression_free(rc->warning);
  500. expression_free(rc->critical);
  501. string_freez(rc->key);
  502. string_freez(rc->name);
  503. string_freez(rc->chart);
  504. string_freez(rc->dimensions);
  505. string_freez(rc->foreach_dimension);
  506. string_freez(rc->exec);
  507. string_freez(rc->recipient);
  508. string_freez(rc->source);
  509. string_freez(rc->units);
  510. string_freez(rc->info);
  511. string_freez(rc->original_info);
  512. string_freez(rc->classification);
  513. string_freez(rc->component);
  514. string_freez(rc->type);
  515. string_freez(rc->host_labels);
  516. string_freez(rc->module_match);
  517. string_freez(rc->plugin_match);
  518. string_freez(rc->chart_labels);
  519. simple_pattern_free(rc->foreach_dimension_pattern);
  520. simple_pattern_free(rc->host_labels_pattern);
  521. simple_pattern_free(rc->module_pattern);
  522. simple_pattern_free(rc->plugin_pattern);
  523. simple_pattern_free(rc->chart_labels_pattern);
  524. }
  525. static void rrdcalc_rrdhost_delete_callback(const DICTIONARY_ITEM *item __maybe_unused, void *rrdcalc, void *rrdhost __maybe_unused) {
  526. RRDCALC *rc = rrdcalc;
  527. //RRDHOST *host = rrdhost;
  528. if(unlikely(rc->rrdset))
  529. rrdcalc_unlink_from_rrdset(rc, false);
  530. // any destruction actions that require other locks
  531. // have to be placed in rrdcalc_del(), because the object is actually locked for deletion
  532. rrdcalc_free_internals(rc);
  533. }
  534. // ----------------------------------------------------------------------------
  535. // RRDCALC rrdhost index management - index API
  536. void rrdcalc_rrdhost_index_init(RRDHOST *host) {
  537. if(!host->rrdcalc_root_index) {
  538. host->rrdcalc_root_index = dictionary_create_advanced(DICT_OPTION_DONT_OVERWRITE_VALUE | DICT_OPTION_FIXED_SIZE,
  539. &dictionary_stats_category_rrdhealth, sizeof(RRDCALC));
  540. dictionary_register_insert_callback(host->rrdcalc_root_index, rrdcalc_rrdhost_insert_callback, NULL);
  541. dictionary_register_conflict_callback(host->rrdcalc_root_index, rrdcalc_rrdhost_conflict_callback, NULL);
  542. dictionary_register_react_callback(host->rrdcalc_root_index, rrdcalc_rrdhost_react_callback, NULL);
  543. dictionary_register_delete_callback(host->rrdcalc_root_index, rrdcalc_rrdhost_delete_callback, host);
  544. }
  545. }
  546. void rrdcalc_rrdhost_index_destroy(RRDHOST *host) {
  547. dictionary_destroy(host->rrdcalc_root_index);
  548. host->rrdcalc_root_index = NULL;
  549. }
  550. void rrdcalc_add_from_rrdcalctemplate(RRDHOST *host, RRDCALCTEMPLATE *rt, RRDSET *st, const char *overwrite_alert_name, const char *overwrite_dimensions) {
  551. char key[RRDCALC_MAX_KEY_SIZE + 1];
  552. size_t key_len = rrdcalc_key(key, RRDCALC_MAX_KEY_SIZE, rrdset_id(st),
  553. overwrite_alert_name?overwrite_alert_name:string2str(rt->name));
  554. struct rrdcalc_constructor tmp = {
  555. .rrdhost = host,
  556. .from_config = NULL,
  557. .from_rrdcalctemplate = rt,
  558. .rrdset = st,
  559. .overwrite_alert_name = overwrite_alert_name,
  560. .overwrite_dimensions = overwrite_dimensions,
  561. .react_action = RRDCALC_REACT_NONE,
  562. .existing_from_template = false,
  563. };
  564. dictionary_set_advanced(host->rrdcalc_root_index, key, (ssize_t)(key_len + 1), NULL, sizeof(RRDCALC), &tmp);
  565. if(tmp.react_action != RRDCALC_REACT_NEW && tmp.existing_from_template == false)
  566. netdata_log_error("RRDCALC: from template '%s' on chart '%s' with key '%s', failed to be added to host '%s'. It is manually configured.",
  567. string2str(rt->name), rrdset_id(st), key, rrdhost_hostname(host));
  568. }
  569. int rrdcalc_add_from_config(RRDHOST *host, RRDCALC *rc) {
  570. if(!rc->chart) {
  571. netdata_log_error("Health configuration for alarm '%s' does not have a chart", rrdcalc_name(rc));
  572. return 0;
  573. }
  574. if(!rc->update_every) {
  575. netdata_log_error("Health configuration for alarm '%s.%s' has no frequency (parameter 'every'). Ignoring it.", rrdcalc_chart_name(rc), rrdcalc_name(rc));
  576. return 0;
  577. }
  578. if(!RRDCALC_HAS_DB_LOOKUP(rc) && !rc->calculation && !rc->warning && !rc->critical) {
  579. netdata_log_error("Health configuration for alarm '%s.%s' is useless (no db lookup, no calculation, no warning and no critical expressions)", rrdcalc_chart_name(rc), rrdcalc_name(rc));
  580. return 0;
  581. }
  582. char key[RRDCALC_MAX_KEY_SIZE + 1];
  583. size_t key_len = rrdcalc_key(key, RRDCALC_MAX_KEY_SIZE, string2str(rc->chart), string2str(rc->name));
  584. struct rrdcalc_constructor tmp = {
  585. .rrdhost = host,
  586. .from_config = rc,
  587. .from_rrdcalctemplate = NULL,
  588. .rrdset = NULL,
  589. .react_action = RRDCALC_REACT_NONE,
  590. };
  591. int ret = 1;
  592. RRDCALC *t = dictionary_set_advanced(host->rrdcalc_root_index, key, (ssize_t)(key_len + 1), rc, sizeof(RRDCALC), &tmp);
  593. if(tmp.react_action == RRDCALC_REACT_NEW) {
  594. // we copied rc into the dictionary, so we have to free the container here
  595. freez(rc);
  596. rc = t;
  597. // since we loaded this config from configuration, we need to check if we can link it to alarms
  598. RRDSET *st;
  599. rrdset_foreach_read(st, host) {
  600. if (unlikely(rrdcalc_check_and_link_rrdset_callback(st, rc) == -1))
  601. break;
  602. }
  603. rrdset_foreach_done(st);
  604. }
  605. else {
  606. netdata_log_error(
  607. "RRDCALC: from config '%s' on chart '%s' failed to be added to host '%s'. It already exists.",
  608. string2str(rc->name),
  609. string2str(rc->chart),
  610. rrdhost_hostname(host));
  611. ret = 0;
  612. // free all of it, internals and the container
  613. rrdcalc_free_unused_rrdcalc_loaded_from_config(rc);
  614. }
  615. return ret;
  616. }
  617. static void rrdcalc_unlink_and_delete(RRDHOST *host, RRDCALC *rc, bool having_ll_wrlock) {
  618. if(rc->rrdset)
  619. rrdcalc_unlink_from_rrdset(rc, having_ll_wrlock);
  620. dictionary_del_advanced(host->rrdcalc_root_index, string2str(rc->key), (ssize_t)string_strlen(rc->key) + 1);
  621. }
  622. // ----------------------------------------------------------------------------
  623. // RRDCALC cleanup API functions
  624. void rrdcalc_delete_alerts_not_matching_host_labels_from_this_host(RRDHOST *host) {
  625. RRDCALC *rc;
  626. foreach_rrdcalc_in_rrdhost_reentrant(host, rc) {
  627. if (!rc->host_labels)
  628. continue;
  629. if(!rrdlabels_match_simple_pattern_parsed(host->rrdlabels, rc->host_labels_pattern, '=', NULL)) {
  630. netdata_log_health("Health configuration for alarm '%s' cannot be applied, because the host %s does not have the label(s) '%s'",
  631. rrdcalc_name(rc),
  632. rrdhost_hostname(host),
  633. rrdcalc_host_labels(rc));
  634. rrdcalc_unlink_and_delete(host, rc, false);
  635. }
  636. }
  637. foreach_rrdcalc_in_rrdhost_done(rc);
  638. }
  639. void rrdcalc_delete_alerts_not_matching_host_labels_from_all_hosts() {
  640. RRDHOST *host;
  641. dfe_start_reentrant(rrdhost_root_index, host) {
  642. if (unlikely(!host->health.health_enabled))
  643. continue;
  644. if (host->rrdlabels)
  645. rrdcalc_delete_alerts_not_matching_host_labels_from_this_host(host);
  646. }
  647. dfe_done(host);
  648. }
  649. void rrdcalc_unlink_all_rrdset_alerts(RRDSET *st) {
  650. RRDCALC *rc, *last = NULL;
  651. rw_spinlock_write_lock(&st->alerts.spinlock);
  652. while((rc = st->alerts.base)) {
  653. if(last == rc) {
  654. netdata_log_error("RRDCALC: malformed list of alerts linked to chart - cannot cleanup - giving up.");
  655. break;
  656. }
  657. last = rc;
  658. if(rc->run_flags & RRDCALC_FLAG_FROM_TEMPLATE) {
  659. // if the alert comes from a template we can just delete it
  660. rrdcalc_unlink_and_delete(st->rrdhost, rc, true);
  661. }
  662. else {
  663. // this is a configuration for a specific chart
  664. // it should stay in the list
  665. rrdcalc_unlink_from_rrdset(rc, true);
  666. }
  667. }
  668. rw_spinlock_write_unlock(&st->alerts.spinlock);
  669. }
  670. void rrdcalc_delete_all(RRDHOST *host) {
  671. dictionary_flush(host->rrdcalc_root_index);
  672. }
  673. void rrdcalc_free_unused_rrdcalc_loaded_from_config(RRDCALC *rc) {
  674. if(rc->rrdset)
  675. rrdcalc_unlink_from_rrdset(rc, false);
  676. rrdcalc_free_internals(rc);
  677. freez(rc);
  678. }