sqlite_aclk_alert.c 40 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938939940941942943944945946947948949950951952953954955956957958959960961962963964965966967968969970971972973974975976977978979980981982983984985986987988989990991992993994995996997998999100010011002100310041005100610071008100910101011101210131014101510161017101810191020102110221023102410251026102710281029103010311032103310341035103610371038103910401041104210431044104510461047104810491050105110521053105410551056105710581059106010611062106310641065106610671068106910701071107210731074107510761077107810791080108110821083108410851086108710881089109010911092109310941095109610971098109911001101110211031104110511061107110811091110111111121113111411151116111711181119112011211122
  1. // SPDX-License-Identifier: GPL-3.0-or-later
  2. #include "sqlite_functions.h"
  3. #include "sqlite_aclk_alert.h"
  4. #ifdef ENABLE_NEW_CLOUD_PROTOCOL
  5. #include "../../aclk/aclk_alarm_api.h"
  6. #include "../../aclk/aclk.h"
  7. #endif
  8. time_t removed_when(uint32_t alarm_id, uint32_t before_unique_id, uint32_t after_unique_id, char *uuid_str) {
  9. sqlite3_stmt *res = NULL;
  10. int rc = 0;
  11. time_t when = 0;
  12. char sql[ACLK_SYNC_QUERY_SIZE];
  13. snprintfz(sql,ACLK_SYNC_QUERY_SIZE-1, "select when_key from health_log_%s where alarm_id = %u " \
  14. "and unique_id > %u and unique_id < %u " \
  15. "and new_status = -2;", uuid_str, alarm_id, after_unique_id, before_unique_id);
  16. rc = sqlite3_prepare_v2(db_meta, sql, -1, &res, 0);
  17. if (rc != SQLITE_OK) {
  18. error_report("Failed to prepare statement when trying to find removed gap.");
  19. return 0;
  20. }
  21. rc = sqlite3_step(res);
  22. if (likely(rc == SQLITE_ROW)) {
  23. when = (time_t) sqlite3_column_int64(res, 0);
  24. }
  25. rc = sqlite3_finalize(res);
  26. if (unlikely(rc != SQLITE_OK))
  27. error_report("Failed to finalize statement when trying to find removed gap, rc = %d", rc);
  28. return when;
  29. }
  30. #define MAX_REMOVED_PERIOD 900
  31. //decide if some events should be sent or not
  32. int should_send_to_cloud(RRDHOST *host, ALARM_ENTRY *ae)
  33. {
  34. sqlite3_stmt *res = NULL;
  35. char uuid_str[GUID_LEN + 1];
  36. uuid_unparse_lower_fix(&host->host_uuid, uuid_str);
  37. int send = 1, rc = 0;
  38. if (ae->new_status == RRDCALC_STATUS_REMOVED || ae->new_status == RRDCALC_STATUS_UNINITIALIZED) {
  39. return 0;
  40. }
  41. if (unlikely(uuid_is_null(ae->config_hash_id)))
  42. return 0;
  43. char sql[ACLK_SYNC_QUERY_SIZE];
  44. uuid_t config_hash_id;
  45. RRDCALC_STATUS status;
  46. uint32_t unique_id;
  47. //get the previous sent event of this alarm_id
  48. snprintfz(sql,ACLK_SYNC_QUERY_SIZE-1, "select hl.new_status, hl.config_hash_id, hl.unique_id from health_log_%s hl, aclk_alert_%s aa \
  49. where hl.unique_id = aa.alert_unique_id \
  50. and hl.alarm_id = %u and hl.unique_id <> %u \
  51. order by alarm_event_id desc LIMIT 1;", uuid_str, uuid_str, ae->alarm_id, ae->unique_id);
  52. rc = sqlite3_prepare_v2(db_meta, sql, -1, &res, 0);
  53. if (rc != SQLITE_OK) {
  54. error_report("Failed to prepare statement when trying to filter alert events.");
  55. send = 1;
  56. return send;
  57. }
  58. rc = sqlite3_step(res);
  59. if (likely(rc == SQLITE_ROW)) {
  60. status = (RRDCALC_STATUS) sqlite3_column_int(res, 0);
  61. if (sqlite3_column_type(res, 1) != SQLITE_NULL)
  62. uuid_copy(config_hash_id, *((uuid_t *) sqlite3_column_blob(res, 1)));
  63. unique_id = (uint32_t) sqlite3_column_int64(res, 2);
  64. } else {
  65. send = 1;
  66. goto done;
  67. }
  68. if (ae->new_status != (RRDCALC_STATUS)status) {
  69. send = 1;
  70. goto done;
  71. }
  72. if (uuid_compare(ae->config_hash_id, config_hash_id)) {
  73. send = 1;
  74. goto done;
  75. }
  76. //same status, same config
  77. if (ae->new_status == RRDCALC_STATUS_CLEAR) {
  78. send = 0;
  79. goto done;
  80. }
  81. //detect a long off period of the agent, TODO make global
  82. if (ae->new_status == RRDCALC_STATUS_WARNING || ae->new_status == RRDCALC_STATUS_CRITICAL) {
  83. time_t when = removed_when(ae->alarm_id, ae->unique_id, unique_id, uuid_str);
  84. if (when && (when + (time_t)MAX_REMOVED_PERIOD) < ae->when) {
  85. send = 1;
  86. goto done;
  87. } else {
  88. send = 0;
  89. goto done;
  90. }
  91. }
  92. done:
  93. rc = sqlite3_finalize(res);
  94. if (unlikely(rc != SQLITE_OK))
  95. error_report("Failed to finalize statement when trying to filter alert events, rc = %d", rc);
  96. return send;
  97. }
  98. // will replace call to aclk_update_alarm in health/health_log.c
  99. // and handle both cases
  100. int sql_queue_alarm_to_aclk(RRDHOST *host, ALARM_ENTRY *ae, int skip_filter)
  101. {
  102. //check aclk architecture and handle old json alarm update to cloud
  103. //include also the valid statuses for this case
  104. #ifdef ENABLE_ACLK
  105. #ifdef ENABLE_NEW_CLOUD_PROTOCOL
  106. if (!aclk_use_new_cloud_arch && aclk_connected) {
  107. #endif
  108. if ((ae->new_status == RRDCALC_STATUS_WARNING || ae->new_status == RRDCALC_STATUS_CRITICAL) ||
  109. ((ae->old_status == RRDCALC_STATUS_WARNING || ae->old_status == RRDCALC_STATUS_CRITICAL))) {
  110. aclk_update_alarm(host, ae);
  111. }
  112. #endif
  113. #ifdef ENABLE_NEW_CLOUD_PROTOCOL
  114. }
  115. if (!claimed())
  116. return 0;
  117. if (ae->flags & HEALTH_ENTRY_FLAG_ACLK_QUEUED) {
  118. return 0;
  119. }
  120. if (!skip_filter) {
  121. if (!should_send_to_cloud(host, ae)) {
  122. return 0;
  123. }
  124. }
  125. int rc = 0;
  126. CHECK_SQLITE_CONNECTION(db_meta);
  127. sqlite3_stmt *res_alert = NULL;
  128. char uuid_str[GUID_LEN + 1];
  129. uuid_unparse_lower_fix(&host->host_uuid, uuid_str);
  130. BUFFER *sql = buffer_create(1024);
  131. buffer_sprintf(
  132. sql,
  133. "INSERT INTO aclk_alert_%s (alert_unique_id, date_created) "
  134. "VALUES (@alert_unique_id, strftime('%%s')) on conflict (alert_unique_id) do nothing; ",
  135. uuid_str);
  136. rc = sqlite3_prepare_v2(db_meta, buffer_tostring(sql), -1, &res_alert, 0);
  137. if (unlikely(rc != SQLITE_OK)) {
  138. error_report("Failed to prepare statement to store alert event");
  139. buffer_free(sql);
  140. return 1;
  141. }
  142. rc = sqlite3_bind_int(res_alert, 1, ae->unique_id);
  143. if (unlikely(rc != SQLITE_OK))
  144. goto bind_fail;
  145. rc = execute_insert(res_alert);
  146. if (unlikely(rc != SQLITE_DONE)) {
  147. error_report("Failed to store alert event %u, rc = %d", ae->unique_id, rc);
  148. goto bind_fail;
  149. }
  150. ae->flags |= HEALTH_ENTRY_FLAG_ACLK_QUEUED;
  151. struct aclk_database_worker_config *wc = (struct aclk_database_worker_config *)host->dbsync_worker;
  152. if (wc) {
  153. wc->pause_alert_updates = 0;
  154. }
  155. bind_fail:
  156. if (unlikely(sqlite3_finalize(res_alert) != SQLITE_OK))
  157. error_report("Failed to reset statement in store alert event, rc = %d", rc);
  158. buffer_free(sql);
  159. return 0;
  160. #else
  161. UNUSED(host);
  162. UNUSED(ae);
  163. UNUSED(skip_filter);
  164. #endif
  165. return 0;
  166. }
  167. int rrdcalc_status_to_proto_enum(RRDCALC_STATUS status)
  168. {
  169. #ifdef ENABLE_NEW_CLOUD_PROTOCOL
  170. switch(status) {
  171. case RRDCALC_STATUS_REMOVED:
  172. return ALARM_STATUS_REMOVED;
  173. case RRDCALC_STATUS_UNDEFINED:
  174. return ALARM_STATUS_NOT_A_NUMBER;
  175. case RRDCALC_STATUS_CLEAR:
  176. return ALARM_STATUS_CLEAR;
  177. case RRDCALC_STATUS_WARNING:
  178. return ALARM_STATUS_WARNING;
  179. case RRDCALC_STATUS_CRITICAL:
  180. return ALARM_STATUS_CRITICAL;
  181. default:
  182. return ALARM_STATUS_UNKNOWN;
  183. }
  184. #else
  185. UNUSED(status);
  186. return 1;
  187. #endif
  188. }
  189. void aclk_push_alert_event(struct aclk_database_worker_config *wc, struct aclk_database_cmd cmd)
  190. {
  191. #ifndef ENABLE_NEW_CLOUD_PROTOCOL
  192. UNUSED(wc);
  193. UNUSED(cmd);
  194. #else
  195. int rc;
  196. if (unlikely(!wc->alert_updates)) {
  197. log_access("ACLK STA [%s (%s)]: Ignoring alert push event, updates have been turned off for this node.", wc->node_id, wc->host ? wc->host->hostname : "N/A");
  198. return;
  199. }
  200. char *claim_id = is_agent_claimed();
  201. if (unlikely(!claim_id))
  202. return;
  203. if (unlikely(!wc->host)) {
  204. freez(claim_id);
  205. return;
  206. }
  207. BUFFER *sql = buffer_create(1024);
  208. if (wc->alerts_start_seq_id != 0) {
  209. buffer_sprintf(
  210. sql,
  211. "UPDATE aclk_alert_%s SET date_submitted = NULL, date_cloud_ack = NULL WHERE sequence_id >= %"PRIu64
  212. "; UPDATE aclk_alert_%s SET date_cloud_ack = strftime('%%s','now') WHERE sequence_id < %"PRIu64
  213. " and date_cloud_ack is null "
  214. "; UPDATE aclk_alert_%s SET date_submitted = strftime('%%s','now') WHERE sequence_id < %"PRIu64
  215. " and date_submitted is null",
  216. wc->uuid_str,
  217. wc->alerts_start_seq_id,
  218. wc->uuid_str,
  219. wc->alerts_start_seq_id,
  220. wc->uuid_str,
  221. wc->alerts_start_seq_id);
  222. db_execute(buffer_tostring(sql));
  223. buffer_reset(sql);
  224. wc->alerts_start_seq_id = 0;
  225. }
  226. int limit = cmd.count > 0 ? cmd.count : 1;
  227. sqlite3_stmt *res = NULL;
  228. buffer_sprintf(sql, "select aa.sequence_id, hl.unique_id, hl.alarm_id, hl.config_hash_id, hl.updated_by_id, hl.when_key, \
  229. hl.duration, hl.non_clear_duration, hl.flags, hl.exec_run_timestamp, hl.delay_up_to_timestamp, hl.name, \
  230. hl.chart, hl.family, hl.exec, hl.recipient, hl.source, hl.units, hl.info, hl.exec_code, hl.new_status, \
  231. hl.old_status, hl.delay, hl.new_value, hl.old_value, hl.last_repeat \
  232. from health_log_%s hl, aclk_alert_%s aa \
  233. where hl.unique_id = aa.alert_unique_id and aa.date_submitted is null \
  234. order by aa.sequence_id asc limit %d;", wc->uuid_str, wc->uuid_str, limit);
  235. rc = sqlite3_prepare_v2(db_meta, buffer_tostring(sql), -1, &res, 0);
  236. if (rc != SQLITE_OK) {
  237. error_report("Failed to prepare statement when trying to send an alert update via ACLK");
  238. buffer_free(sql);
  239. freez(claim_id);
  240. return;
  241. }
  242. char uuid_str[GUID_LEN + 1];
  243. uint64_t first_sequence_id = 0;
  244. uint64_t last_sequence_id = 0;
  245. static __thread uint64_t log_first_sequence_id = 0;
  246. static __thread uint64_t log_last_sequence_id = 0;
  247. while (sqlite3_step(res) == SQLITE_ROW) {
  248. struct alarm_log_entry alarm_log;
  249. char old_value_string[100 + 1];
  250. char new_value_string[100 + 1];
  251. alarm_log.node_id = wc->node_id;
  252. alarm_log.claim_id = claim_id;
  253. alarm_log.chart = strdupz((char *)sqlite3_column_text(res, 12));
  254. alarm_log.name = strdupz((char *)sqlite3_column_text(res, 11));
  255. alarm_log.family = sqlite3_column_bytes(res, 13) > 0 ? strdupz((char *)sqlite3_column_text(res, 13)) : NULL;
  256. alarm_log.batch_id = wc->alerts_batch_id;
  257. alarm_log.sequence_id = (uint64_t) sqlite3_column_int64(res, 0);
  258. alarm_log.when = (time_t) sqlite3_column_int64(res, 5);
  259. uuid_unparse_lower(*((uuid_t *) sqlite3_column_blob(res, 3)), uuid_str);
  260. alarm_log.config_hash = strdupz((char *)uuid_str);
  261. alarm_log.utc_offset = wc->host->utc_offset;
  262. alarm_log.timezone = strdupz((char *)wc->host->abbrev_timezone);
  263. alarm_log.exec_path = sqlite3_column_bytes(res, 14) > 0 ? strdupz((char *)sqlite3_column_text(res, 14)) :
  264. strdupz((char *)wc->host->health_default_exec);
  265. alarm_log.conf_source = strdupz((char *)sqlite3_column_text(res, 16));
  266. char *edit_command = sqlite3_column_bytes(res, 16) > 0 ?
  267. health_edit_command_from_source((char *)sqlite3_column_text(res, 16)) :
  268. strdupz("UNKNOWN=0=UNKNOWN");
  269. alarm_log.command = strdupz(edit_command);
  270. alarm_log.duration = (time_t) sqlite3_column_int64(res, 6);
  271. alarm_log.non_clear_duration = (time_t) sqlite3_column_int64(res, 7);
  272. alarm_log.status = rrdcalc_status_to_proto_enum((RRDCALC_STATUS) sqlite3_column_int(res, 20));
  273. alarm_log.old_status = rrdcalc_status_to_proto_enum((RRDCALC_STATUS) sqlite3_column_int(res, 21));
  274. alarm_log.delay = (int) sqlite3_column_int(res, 22);
  275. alarm_log.delay_up_to_timestamp = (time_t) sqlite3_column_int64(res, 10);
  276. alarm_log.last_repeat = (time_t) sqlite3_column_int64(res, 25);
  277. alarm_log.silenced = ((sqlite3_column_int64(res, 8) & HEALTH_ENTRY_FLAG_SILENCED) ||
  278. (sqlite3_column_type(res, 15) != SQLITE_NULL &&
  279. !strncmp((char *)sqlite3_column_text(res, 15), "silent", 6))) ?
  280. 1 :
  281. 0;
  282. alarm_log.value_string =
  283. sqlite3_column_type(res, 23) == SQLITE_NULL ?
  284. strdupz((char *)"-") :
  285. strdupz((char *)format_value_and_unit(
  286. new_value_string, 100, sqlite3_column_double(res, 23), (char *)sqlite3_column_text(res, 17), -1));
  287. alarm_log.old_value_string =
  288. sqlite3_column_type(res, 24) == SQLITE_NULL ?
  289. strdupz((char *)"-") :
  290. strdupz((char *)format_value_and_unit(
  291. old_value_string, 100, sqlite3_column_double(res, 24), (char *)sqlite3_column_text(res, 17), -1));
  292. alarm_log.value = (calculated_number) sqlite3_column_double(res, 23);
  293. alarm_log.old_value = (calculated_number) sqlite3_column_double(res, 24);
  294. alarm_log.updated = (sqlite3_column_int64(res, 8) & HEALTH_ENTRY_FLAG_UPDATED) ? 1 : 0;
  295. alarm_log.rendered_info = sqlite3_column_type(res, 18) == SQLITE_NULL ?
  296. strdupz((char *)"") :
  297. strdupz((char *)sqlite3_column_text(res, 18));
  298. aclk_send_alarm_log_entry(&alarm_log);
  299. if (first_sequence_id == 0)
  300. first_sequence_id = (uint64_t) sqlite3_column_int64(res, 0);
  301. if (log_first_sequence_id == 0)
  302. log_first_sequence_id = (uint64_t) sqlite3_column_int64(res, 0);
  303. last_sequence_id = (uint64_t) sqlite3_column_int64(res, 0);
  304. log_last_sequence_id = (uint64_t) sqlite3_column_int64(res, 0);
  305. destroy_alarm_log_entry(&alarm_log);
  306. freez(edit_command);
  307. }
  308. if (first_sequence_id) {
  309. buffer_flush(sql);
  310. buffer_sprintf(sql, "UPDATE aclk_alert_%s SET date_submitted=strftime('%%s') "
  311. "WHERE date_submitted IS NULL AND sequence_id BETWEEN %" PRIu64 " AND %" PRIu64 ";",
  312. wc->uuid_str, first_sequence_id, last_sequence_id);
  313. db_execute(buffer_tostring(sql));
  314. } else {
  315. if (log_first_sequence_id)
  316. log_access(
  317. "ACLK RES [%s (%s)]: ALERTS SENT from %" PRIu64 " to %" PRIu64 " batch=%" PRIu64,
  318. wc->node_id,
  319. wc->host ? wc->host->hostname : "N/A",
  320. log_first_sequence_id,
  321. log_last_sequence_id,
  322. wc->alerts_batch_id);
  323. log_first_sequence_id = 0;
  324. log_last_sequence_id = 0;
  325. wc->pause_alert_updates = 1;
  326. }
  327. rc = sqlite3_finalize(res);
  328. if (unlikely(rc != SQLITE_OK))
  329. error_report("Failed to finalize statement to send alert entries from the database, rc = %d", rc);
  330. freez(claim_id);
  331. buffer_free(sql);
  332. #endif
  333. return;
  334. }
  335. void sql_queue_existing_alerts_to_aclk(RRDHOST *host)
  336. {
  337. char uuid_str[GUID_LEN + 1];
  338. uuid_unparse_lower_fix(&host->host_uuid, uuid_str);
  339. BUFFER *sql = buffer_create(1024);
  340. buffer_sprintf(sql,"insert into aclk_alert_%s (alert_unique_id, date_created) " \
  341. "select unique_id alert_unique_id, strftime('%%s') date_created from health_log_%s " \
  342. "where new_status <> 0 and new_status <> -2 and config_hash_id is not null and updated_by_id = 0 " \
  343. "order by unique_id asc on conflict (alert_unique_id) do nothing;", uuid_str, uuid_str);
  344. db_execute(buffer_tostring(sql));
  345. buffer_free(sql);
  346. struct aclk_database_worker_config *wc = (struct aclk_database_worker_config *)host->dbsync_worker;
  347. if (wc) {
  348. wc->pause_alert_updates = 0;
  349. }
  350. }
  351. void aclk_send_alarm_health_log(char *node_id)
  352. {
  353. if (unlikely(!node_id))
  354. return;
  355. char *hostname= NULL;
  356. struct aclk_database_worker_config *wc = NULL;
  357. struct aclk_database_cmd cmd;
  358. memset(&cmd, 0, sizeof(cmd));
  359. cmd.opcode = ACLK_DATABASE_ALARM_HEALTH_LOG;
  360. rrd_rdlock();
  361. RRDHOST *host = find_host_by_node_id(node_id);
  362. if (likely(host)) {
  363. wc = (struct aclk_database_worker_config *)host->dbsync_worker;
  364. hostname = host->hostname;
  365. }
  366. else
  367. hostname = get_hostname_by_node_id(node_id);
  368. rrd_unlock();
  369. log_access("ACLK REQ [%s (%s)]: HEALTH LOG request received", node_id, hostname ? hostname : "N/A");
  370. if (unlikely(!host))
  371. freez(hostname);
  372. if (wc)
  373. aclk_database_enq_cmd(wc, &cmd);
  374. else {
  375. if (aclk_worker_enq_cmd(node_id, &cmd))
  376. log_access("ACLK STA [%s (N/A)]: ACLK synchronization thread is not active.", node_id);
  377. }
  378. return;
  379. }
  380. void aclk_push_alarm_health_log(struct aclk_database_worker_config *wc, struct aclk_database_cmd cmd)
  381. {
  382. UNUSED(cmd);
  383. #ifndef ENABLE_NEW_CLOUD_PROTOCOL
  384. UNUSED(wc);
  385. #else
  386. int rc;
  387. char *claim_id = is_agent_claimed();
  388. if (unlikely(!claim_id))
  389. return;
  390. RRDHOST *host = wc->host;
  391. if (unlikely(!host)) {
  392. rrd_rdlock();
  393. host = find_host_by_node_id(wc->node_id);
  394. rrd_unlock();
  395. if (unlikely(!host)) {
  396. log_access(
  397. "AC [%s (N/A)]: ACLK synchronization thread for %s is not yet linked to HOST.",
  398. wc->node_id,
  399. wc->host_guid);
  400. freez(claim_id);
  401. return;
  402. }
  403. }
  404. uint64_t first_sequence = 0;
  405. uint64_t last_sequence = 0;
  406. struct timeval first_timestamp;
  407. struct timeval last_timestamp;
  408. BUFFER *sql = buffer_create(1024);
  409. sqlite3_stmt *res = NULL;
  410. //TODO: make this better: include info from health log too
  411. buffer_sprintf(sql, "SELECT MIN(sequence_id), MIN(date_created), MAX(sequence_id), MAX(date_created) " \
  412. "FROM aclk_alert_%s;", wc->uuid_str);
  413. rc = sqlite3_prepare_v2(db_meta, buffer_tostring(sql), -1, &res, 0);
  414. if (rc != SQLITE_OK) {
  415. error_report("Failed to prepare statement to get health log statistics from the database");
  416. buffer_free(sql);
  417. freez(claim_id);
  418. return;
  419. }
  420. first_timestamp.tv_sec = 0;
  421. first_timestamp.tv_usec = 0;
  422. last_timestamp.tv_sec = 0;
  423. last_timestamp.tv_usec = 0;
  424. while (sqlite3_step(res) == SQLITE_ROW) {
  425. first_sequence = sqlite3_column_bytes(res, 0) > 0 ? (uint64_t) sqlite3_column_int64(res, 0) : 0;
  426. if (sqlite3_column_bytes(res, 1) > 0) {
  427. first_timestamp.tv_sec = sqlite3_column_int64(res, 1);
  428. }
  429. last_sequence = sqlite3_column_bytes(res, 2) > 0 ? (uint64_t) sqlite3_column_int64(res, 2) : 0;
  430. if (sqlite3_column_bytes(res, 3) > 0) {
  431. last_timestamp.tv_sec = sqlite3_column_int64(res, 3);
  432. }
  433. }
  434. struct alarm_log_entries log_entries;
  435. log_entries.first_seq_id = first_sequence;
  436. log_entries.first_when = first_timestamp;
  437. log_entries.last_seq_id = last_sequence;
  438. log_entries.last_when = last_timestamp;
  439. struct alarm_log_health alarm_log;
  440. alarm_log.claim_id = claim_id;
  441. alarm_log.node_id = wc->node_id;
  442. alarm_log.log_entries = log_entries;
  443. alarm_log.status = wc->alert_updates == 0 ? 2 : 1;
  444. alarm_log.enabled = (int)host->health_enabled;
  445. wc->alert_sequence_id = last_sequence;
  446. aclk_send_alarm_log_health(&alarm_log);
  447. log_access("ACLK RES [%s (%s)]: HEALTH LOG SENT from %"PRIu64" to %"PRIu64, wc->node_id, wc->host ? wc->host->hostname : "N/A", first_sequence, last_sequence);
  448. rc = sqlite3_finalize(res);
  449. if (unlikely(rc != SQLITE_OK))
  450. error_report("Failed to reset statement to get health log statistics from the database, rc = %d", rc);
  451. freez(claim_id);
  452. buffer_free(sql);
  453. aclk_alert_reloaded = 1;
  454. #endif
  455. return;
  456. }
  457. void aclk_send_alarm_configuration(char *config_hash)
  458. {
  459. if (unlikely(!config_hash))
  460. return;
  461. struct aclk_database_worker_config *wc = (struct aclk_database_worker_config *) localhost->dbsync_worker;
  462. if (unlikely(!wc)) {
  463. return;
  464. }
  465. log_access("ACLK REQ [%s (%s)]: Request to send alert config %s.", wc->node_id, wc->host ? wc->host->hostname : "N/A", config_hash);
  466. struct aclk_database_cmd cmd;
  467. memset(&cmd, 0, sizeof(cmd));
  468. cmd.opcode = ACLK_DATABASE_PUSH_ALERT_CONFIG;
  469. cmd.data_param = (void *) strdupz(config_hash);
  470. cmd.completion = NULL;
  471. aclk_database_enq_cmd(wc, &cmd);
  472. return;
  473. }
  474. #define SQL_SELECT_ALERT_CONFIG "SELECT alarm, template, on_key, class, type, component, os, hosts, plugin," \
  475. "module, charts, families, lookup, every, units, green, red, calc, warn, crit, to_key, exec, delay, repeat, info," \
  476. "options, host_labels, p_db_lookup_dimensions, p_db_lookup_method, p_db_lookup_options, p_db_lookup_after," \
  477. "p_db_lookup_before, p_update_every FROM alert_hash WHERE hash_id = @hash_id;"
  478. int aclk_push_alert_config_event(struct aclk_database_worker_config *wc, struct aclk_database_cmd cmd)
  479. {
  480. UNUSED(wc);
  481. #ifndef ENABLE_NEW_CLOUD_PROTOCOL
  482. UNUSED(cmd);
  483. #else
  484. int rc = 0;
  485. CHECK_SQLITE_CONNECTION(db_meta);
  486. sqlite3_stmt *res = NULL;
  487. char *config_hash = (char *) cmd.data_param;
  488. rc = sqlite3_prepare_v2(db_meta, SQL_SELECT_ALERT_CONFIG, -1, &res, 0);
  489. if (rc != SQLITE_OK) {
  490. error_report("Failed to prepare statement when trying to fetch an alarm hash configuration");
  491. return 1;
  492. }
  493. uuid_t hash_uuid;
  494. if (uuid_parse(config_hash, hash_uuid))
  495. return 1;
  496. rc = sqlite3_bind_blob(res, 1, &hash_uuid , sizeof(hash_uuid), SQLITE_STATIC);
  497. if (unlikely(rc != SQLITE_OK))
  498. goto bind_fail;
  499. struct aclk_alarm_configuration alarm_config;
  500. struct provide_alarm_configuration p_alarm_config;
  501. p_alarm_config.cfg_hash = NULL;
  502. if (sqlite3_step(res) == SQLITE_ROW) {
  503. alarm_config.alarm = sqlite3_column_bytes(res, 0) > 0 ? strdupz((char *)sqlite3_column_text(res, 0)) : NULL;
  504. alarm_config.tmpl = sqlite3_column_bytes(res, 1) > 0 ? strdupz((char *)sqlite3_column_text(res, 1)) : NULL;
  505. alarm_config.on_chart = sqlite3_column_bytes(res, 2) > 0 ? strdupz((char *)sqlite3_column_text(res, 2)) : NULL;
  506. alarm_config.classification = sqlite3_column_bytes(res, 3) > 0 ? strdupz((char *)sqlite3_column_text(res, 3)) : NULL;
  507. alarm_config.type = sqlite3_column_bytes(res, 4) > 0 ? strdupz((char *)sqlite3_column_text(res, 4)) : NULL;
  508. alarm_config.component = sqlite3_column_bytes(res, 5) > 0 ? strdupz((char *)sqlite3_column_text(res, 5)) : NULL;
  509. alarm_config.os = sqlite3_column_bytes(res, 6) > 0 ? strdupz((char *)sqlite3_column_text(res, 6)) : NULL;
  510. alarm_config.hosts = sqlite3_column_bytes(res, 7) > 0 ? strdupz((char *)sqlite3_column_text(res, 7)) : NULL;
  511. alarm_config.plugin = sqlite3_column_bytes(res, 8) > 0 ? strdupz((char *)sqlite3_column_text(res, 8)) : NULL;
  512. alarm_config.module = sqlite3_column_bytes(res, 9) > 0 ? strdupz((char *)sqlite3_column_text(res, 9)) : NULL;
  513. alarm_config.charts = sqlite3_column_bytes(res, 10) > 0 ? strdupz((char *)sqlite3_column_text(res, 10)) : NULL;
  514. alarm_config.families = sqlite3_column_bytes(res, 11) > 0 ? strdupz((char *)sqlite3_column_text(res, 11)) : NULL;
  515. alarm_config.lookup = sqlite3_column_bytes(res, 12) > 0 ? strdupz((char *)sqlite3_column_text(res, 12)) : NULL;
  516. alarm_config.every = sqlite3_column_bytes(res, 13) > 0 ? strdupz((char *)sqlite3_column_text(res, 13)) : NULL;
  517. alarm_config.units = sqlite3_column_bytes(res, 14) > 0 ? strdupz((char *)sqlite3_column_text(res, 14)) : NULL;
  518. alarm_config.green = sqlite3_column_bytes(res, 15) > 0 ? strdupz((char *)sqlite3_column_text(res, 15)) : NULL;
  519. alarm_config.red = sqlite3_column_bytes(res, 16) > 0 ? strdupz((char *)sqlite3_column_text(res, 16)) : NULL;
  520. alarm_config.calculation_expr = sqlite3_column_bytes(res, 17) > 0 ? strdupz((char *)sqlite3_column_text(res, 17)) : NULL;
  521. alarm_config.warning_expr = sqlite3_column_bytes(res, 18) > 0 ? strdupz((char *)sqlite3_column_text(res, 18)) : NULL;
  522. alarm_config.critical_expr = sqlite3_column_bytes(res, 19) > 0 ? strdupz((char *)sqlite3_column_text(res, 19)) : NULL;
  523. alarm_config.recipient = sqlite3_column_bytes(res, 20) > 0 ? strdupz((char *)sqlite3_column_text(res, 20)) : NULL;
  524. alarm_config.exec = sqlite3_column_bytes(res, 21) > 0 ? strdupz((char *)sqlite3_column_text(res, 21)) : NULL;
  525. alarm_config.delay = sqlite3_column_bytes(res, 22) > 0 ? strdupz((char *)sqlite3_column_text(res, 22)) : NULL;
  526. alarm_config.repeat = sqlite3_column_bytes(res, 23) > 0 ? strdupz((char *)sqlite3_column_text(res, 23)) : NULL;
  527. alarm_config.info = sqlite3_column_bytes(res, 24) > 0 ? strdupz((char *)sqlite3_column_text(res, 24)) : NULL;
  528. alarm_config.options = sqlite3_column_bytes(res, 25) > 0 ? strdupz((char *)sqlite3_column_text(res, 25)) : NULL;
  529. alarm_config.host_labels = sqlite3_column_bytes(res, 26) > 0 ? strdupz((char *)sqlite3_column_text(res, 26)) : NULL;
  530. alarm_config.p_db_lookup_dimensions = NULL;
  531. alarm_config.p_db_lookup_method = NULL;
  532. alarm_config.p_db_lookup_options = NULL;
  533. alarm_config.p_db_lookup_after = 0;
  534. alarm_config.p_db_lookup_before = 0;
  535. if (sqlite3_column_bytes(res, 30) > 0) {
  536. alarm_config.p_db_lookup_dimensions = sqlite3_column_bytes(res, 27) > 0 ? strdupz((char *)sqlite3_column_text(res, 27)) : NULL;
  537. alarm_config.p_db_lookup_method = sqlite3_column_bytes(res, 28) > 0 ? strdupz((char *)sqlite3_column_text(res, 28)) : NULL;
  538. BUFFER *tmp_buf = buffer_create(1024);
  539. buffer_data_options2string(tmp_buf, sqlite3_column_int(res, 29));
  540. alarm_config.p_db_lookup_options = strdupz((char *)buffer_tostring(tmp_buf));
  541. buffer_free(tmp_buf);
  542. alarm_config.p_db_lookup_after = sqlite3_column_int(res, 30);
  543. alarm_config.p_db_lookup_before = sqlite3_column_int(res, 31);
  544. }
  545. alarm_config.p_update_every = sqlite3_column_int(res, 32);
  546. p_alarm_config.cfg_hash = strdupz((char *) config_hash);
  547. p_alarm_config.cfg = alarm_config;
  548. }
  549. if (likely(p_alarm_config.cfg_hash)) {
  550. log_access("ACLK RES [%s (%s)]: Sent alert config %s.", wc->node_id, wc->host ? wc->host->hostname : "N/A", config_hash);
  551. aclk_send_provide_alarm_cfg(&p_alarm_config);
  552. freez((char *) cmd.data_param);
  553. freez(p_alarm_config.cfg_hash);
  554. destroy_aclk_alarm_configuration(&alarm_config);
  555. }
  556. else
  557. log_access("ACLK STA [%s (%s)]: Alert config for %s not found.", wc->node_id, wc->host ? wc->host->hostname : "N/A", config_hash);
  558. bind_fail:
  559. rc = sqlite3_finalize(res);
  560. if (unlikely(rc != SQLITE_OK))
  561. error_report("Failed to reset statement when pushing alarm config hash, rc = %d", rc);
  562. return rc;
  563. #endif
  564. return 0;
  565. }
  566. // Start streaming alerts
  567. void aclk_start_alert_streaming(char *node_id, uint64_t batch_id, uint64_t start_seq_id)
  568. {
  569. #ifdef ENABLE_NEW_CLOUD_PROTOCOL
  570. if (unlikely(!node_id))
  571. return;
  572. //log_access("ACLK REQ [%s (N/A)]: ALERTS STREAM from %"PRIu64" batch=%"PRIu64".", node_id, start_seq_id, batch_id);
  573. uuid_t node_uuid;
  574. if (uuid_parse(node_id, node_uuid))
  575. return;
  576. struct aclk_database_worker_config *wc = NULL;
  577. rrd_rdlock();
  578. RRDHOST *host = find_host_by_node_id(node_id);
  579. rrd_unlock();
  580. if (likely(host)) {
  581. wc = (struct aclk_database_worker_config *)host->dbsync_worker ?
  582. (struct aclk_database_worker_config *)host->dbsync_worker :
  583. (struct aclk_database_worker_config *)find_inactive_wc_by_node_id(node_id);
  584. if (unlikely(!host->health_enabled)) {
  585. log_access("ACLK STA [%s (N/A)]: Ignoring request to stream alert state changes, health is disabled.", node_id);
  586. return;
  587. }
  588. if (unlikely(batch_id == 1) && unlikely(start_seq_id == 1))
  589. sql_queue_existing_alerts_to_aclk(host);
  590. } else
  591. wc = (struct aclk_database_worker_config *)find_inactive_wc_by_node_id(node_id);
  592. if (likely(wc)) {
  593. log_access("ACLK REQ [%s (%s)]: ALERTS STREAM from %"PRIu64" batch=%"PRIu64, node_id, wc->host ? wc->host->hostname : "N/A", start_seq_id, batch_id);
  594. __sync_synchronize();
  595. wc->alerts_batch_id = batch_id;
  596. wc->alerts_start_seq_id = start_seq_id;
  597. wc->alert_updates = 1;
  598. wc->pause_alert_updates = 0;
  599. __sync_synchronize();
  600. }
  601. else
  602. log_access("ACLK STA [%s (N/A)]: ACLK synchronization thread is not active.", node_id);
  603. #else
  604. UNUSED(node_id);
  605. UNUSED(start_seq_id);
  606. UNUSED(batch_id);
  607. #endif
  608. return;
  609. }
  610. void sql_process_queue_removed_alerts_to_aclk(struct aclk_database_worker_config *wc, struct aclk_database_cmd cmd)
  611. {
  612. UNUSED(cmd);
  613. #ifndef ENABLE_NEW_CLOUD_PROTOCOL
  614. UNUSED(wc);
  615. #else
  616. BUFFER *sql = buffer_create(1024);
  617. buffer_sprintf(sql,"insert into aclk_alert_%s (alert_unique_id, date_created) " \
  618. "select unique_id alert_unique_id, strftime('%%s') date_created from health_log_%s " \
  619. "where new_status = -2 and updated_by_id = 0 and unique_id not in " \
  620. "(select alert_unique_id from aclk_alert_%s) order by unique_id asc " \
  621. "on conflict (alert_unique_id) do nothing;", wc->uuid_str, wc->uuid_str, wc->uuid_str);
  622. db_execute(buffer_tostring(sql));
  623. log_access("ACLK STA [%s (%s)]: QUEUED REMOVED ALERTS", wc->node_id, wc->host ? wc->host->hostname : "N/A");
  624. buffer_free(sql);
  625. wc->pause_alert_updates = 0;
  626. #endif
  627. return;
  628. }
  629. void sql_queue_removed_alerts_to_aclk(RRDHOST *host)
  630. {
  631. #ifdef ENABLE_NEW_CLOUD_PROTOCOL
  632. if (unlikely(!host->dbsync_worker))
  633. return;
  634. if (!claimed())
  635. return;
  636. struct aclk_database_cmd cmd;
  637. memset(&cmd, 0, sizeof(cmd));
  638. cmd.opcode = ACLK_DATABASE_QUEUE_REMOVED_ALERTS;
  639. cmd.data = NULL;
  640. cmd.data_param = NULL;
  641. cmd.completion = NULL;
  642. aclk_database_enq_cmd((struct aclk_database_worker_config *) host->dbsync_worker, &cmd);
  643. #else
  644. UNUSED(host);
  645. #endif
  646. }
  647. void aclk_process_send_alarm_snapshot(char *node_id, char *claim_id, uint64_t snapshot_id, uint64_t sequence_id)
  648. {
  649. UNUSED(claim_id);
  650. #ifdef ENABLE_NEW_CLOUD_PROTOCOL
  651. if (unlikely(!node_id))
  652. return;
  653. uuid_t node_uuid;
  654. if (uuid_parse(node_id, node_uuid))
  655. return;
  656. struct aclk_database_worker_config *wc = NULL;
  657. rrd_rdlock();
  658. RRDHOST *host = find_host_by_node_id(node_id);
  659. if (likely(host))
  660. wc = (struct aclk_database_worker_config *)host->dbsync_worker;
  661. rrd_unlock();
  662. if (likely(wc)) {
  663. log_access(
  664. "IN [%s (%s)]: Request to send alerts snapshot, snapshot_id %" PRIu64 " and ack_sequence_id %" PRIu64,
  665. wc->node_id,
  666. wc->host ? wc->host->hostname : "N/A",
  667. snapshot_id,
  668. sequence_id);
  669. if (wc->alerts_snapshot_id == snapshot_id)
  670. return;
  671. __sync_synchronize();
  672. wc->alerts_snapshot_id = snapshot_id;
  673. wc->alerts_ack_sequence_id = sequence_id;
  674. __sync_synchronize();
  675. struct aclk_database_cmd cmd;
  676. memset(&cmd, 0, sizeof(cmd));
  677. cmd.opcode = ACLK_DATABASE_PUSH_ALERT_SNAPSHOT;
  678. cmd.data_param = NULL;
  679. cmd.completion = NULL;
  680. aclk_database_enq_cmd(wc, &cmd);
  681. } else
  682. log_access("ACLK STA [%s (N/A)]: ACLK synchronization thread is not active.", node_id);
  683. #else
  684. UNUSED(node_id);
  685. UNUSED(snapshot_id);
  686. UNUSED(sequence_id);
  687. #endif
  688. return;
  689. }
  690. void aclk_mark_alert_cloud_ack(char *uuid_str, uint64_t alerts_ack_sequence_id)
  691. {
  692. BUFFER *sql = buffer_create(1024);
  693. if (alerts_ack_sequence_id != 0) {
  694. buffer_sprintf(
  695. sql,
  696. "UPDATE aclk_alert_%s SET date_cloud_ack = strftime('%%s','now') WHERE sequence_id <= %" PRIu64 "",
  697. uuid_str,
  698. alerts_ack_sequence_id);
  699. db_execute(buffer_tostring(sql));
  700. }
  701. buffer_free(sql);
  702. }
  703. #ifdef ENABLE_NEW_CLOUD_PROTOCOL
  704. void health_alarm_entry2proto_nolock(struct alarm_log_entry *alarm_log, ALARM_ENTRY *ae, RRDHOST *host)
  705. {
  706. char *edit_command = ae->source ? health_edit_command_from_source(ae->source) : strdupz("UNKNOWN=0=UNKNOWN");
  707. char config_hash_id[GUID_LEN + 1];
  708. uuid_unparse_lower(ae->config_hash_id, config_hash_id);
  709. alarm_log->chart = strdupz((char *)ae->chart);
  710. alarm_log->name = strdupz((char *)ae->name);
  711. alarm_log->family = strdupz((char *)ae->family);
  712. alarm_log->batch_id = 0;
  713. alarm_log->sequence_id = 0;
  714. alarm_log->when = (time_t)ae->when;
  715. alarm_log->config_hash = strdupz((char *)config_hash_id);
  716. alarm_log->utc_offset = host->utc_offset;
  717. alarm_log->timezone = strdupz((char *)host->abbrev_timezone);
  718. alarm_log->exec_path = ae->exec ? strdupz((char *)ae->exec) : strdupz((char *)host->health_default_exec);
  719. alarm_log->conf_source = ae->source ? strdupz((char *)ae->source) : strdupz((char *)"");
  720. alarm_log->command = strdupz((char *)edit_command);
  721. alarm_log->duration = (time_t)ae->duration;
  722. alarm_log->non_clear_duration = (time_t)ae->non_clear_duration;
  723. alarm_log->status = rrdcalc_status_to_proto_enum((RRDCALC_STATUS)ae->new_status);
  724. alarm_log->old_status = rrdcalc_status_to_proto_enum((RRDCALC_STATUS)ae->old_status);
  725. alarm_log->delay = (int)ae->delay;
  726. alarm_log->delay_up_to_timestamp = (time_t)ae->delay_up_to_timestamp;
  727. alarm_log->last_repeat = (time_t)ae->last_repeat;
  728. alarm_log->silenced =
  729. ((ae->flags & HEALTH_ENTRY_FLAG_SILENCED) || (ae->recipient && !strncmp((char *)ae->recipient, "silent", 6))) ?
  730. 1 :
  731. 0;
  732. alarm_log->value_string = strdupz(ae->new_value_string);
  733. alarm_log->old_value_string = strdupz(ae->old_value_string);
  734. alarm_log->value = (!isnan(ae->new_value)) ? (calculated_number)ae->new_value : 0;
  735. alarm_log->old_value = (!isnan(ae->old_value)) ? (calculated_number)ae->old_value : 0;
  736. alarm_log->updated = (ae->flags & HEALTH_ENTRY_FLAG_UPDATED) ? 1 : 0;
  737. alarm_log->rendered_info = ae->info ? strdupz(ae->info) : strdupz((char *)"");
  738. freez(edit_command);
  739. }
  740. #endif
  741. #ifdef ENABLE_NEW_CLOUD_PROTOCOL
  742. static int have_recent_alarm(RRDHOST *host, uint32_t alarm_id, time_t mark)
  743. {
  744. ALARM_ENTRY *ae = host->health_log.alarms;
  745. while (ae) {
  746. if (ae->alarm_id == alarm_id && ae->unique_id > mark &&
  747. (ae->new_status != RRDCALC_STATUS_WARNING && ae->new_status != RRDCALC_STATUS_CRITICAL))
  748. return 1;
  749. ae = ae->next;
  750. }
  751. return 0;
  752. }
  753. #endif
  754. #define ALARM_EVENTS_PER_CHUNK 10
  755. void aclk_push_alert_snapshot_event(struct aclk_database_worker_config *wc, struct aclk_database_cmd cmd)
  756. {
  757. #ifndef ENABLE_NEW_CLOUD_PROTOCOL
  758. UNUSED(wc);
  759. UNUSED(cmd);
  760. #else
  761. UNUSED(cmd);
  762. // we perhaps we don't need this for snapshots
  763. if (unlikely(!wc->alert_updates)) {
  764. log_access("ACLK STA [%s (%s)]: Ignoring alert snapshot event, updates have been turned off for this node.", wc->node_id, wc->host ? wc->host->hostname : "N/A");
  765. return;
  766. }
  767. if (unlikely(!wc->host)) {
  768. error_report("ACLK synchronization thread for %s is not linked to HOST", wc->host_guid);
  769. return;
  770. }
  771. if (unlikely(!wc->alerts_snapshot_id))
  772. return;
  773. char *claim_id = is_agent_claimed();
  774. if (unlikely(!claim_id))
  775. return;
  776. log_access("ACLK REQ [%s (%s)]: Sending alerts snapshot, snapshot_id %" PRIu64, wc->node_id, wc->host ? wc->host->hostname : "N/A", wc->alerts_snapshot_id);
  777. aclk_mark_alert_cloud_ack(wc->uuid_str, wc->alerts_ack_sequence_id);
  778. RRDHOST *host = wc->host;
  779. uint32_t cnt = 0;
  780. netdata_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
  781. ALARM_ENTRY *ae = host->health_log.alarms;
  782. for (; ae; ae = ae->next) {
  783. if (likely(ae->updated_by_id))
  784. continue;
  785. if (unlikely(ae->new_status == RRDCALC_STATUS_UNINITIALIZED))
  786. continue;
  787. if (have_recent_alarm(host, ae->alarm_id, ae->unique_id))
  788. continue;
  789. cnt++;
  790. }
  791. if (cnt) {
  792. uint32_t chunk = 1, chunks = 0;
  793. chunks = (cnt / ALARM_EVENTS_PER_CHUNK) + (cnt % ALARM_EVENTS_PER_CHUNK != 0);
  794. ae = host->health_log.alarms;
  795. cnt = 0;
  796. struct alarm_snapshot alarm_snap;
  797. alarm_snap.node_id = wc->node_id;
  798. alarm_snap.claim_id = claim_id;
  799. alarm_snap.snapshot_id = wc->alerts_snapshot_id;
  800. alarm_snap.chunks = chunks;
  801. alarm_snap.chunk = chunk;
  802. alarm_snapshot_proto_ptr_t snapshot_proto = NULL;
  803. for (; ae; ae = ae->next) {
  804. if (likely(ae->updated_by_id))
  805. continue;
  806. if (unlikely(ae->new_status == RRDCALC_STATUS_UNINITIALIZED))
  807. continue;
  808. if (have_recent_alarm(host, ae->alarm_id, ae->unique_id))
  809. continue;
  810. cnt++;
  811. struct alarm_log_entry alarm_log;
  812. alarm_log.node_id = wc->node_id;
  813. alarm_log.claim_id = claim_id;
  814. if (!snapshot_proto)
  815. snapshot_proto = generate_alarm_snapshot_proto(&alarm_snap);
  816. health_alarm_entry2proto_nolock(&alarm_log, ae, host);
  817. add_alarm_log_entry2snapshot(snapshot_proto, &alarm_log);
  818. if (cnt == ALARM_EVENTS_PER_CHUNK) {
  819. aclk_send_alarm_snapshot(snapshot_proto);
  820. cnt = 0;
  821. if (chunk < chunks) {
  822. chunk++;
  823. struct alarm_snapshot alarm_snap;
  824. alarm_snap.node_id = wc->node_id;
  825. alarm_snap.claim_id = claim_id;
  826. alarm_snap.snapshot_id = wc->alerts_snapshot_id;
  827. alarm_snap.chunks = chunks;
  828. alarm_snap.chunk = chunk;
  829. snapshot_proto = generate_alarm_snapshot_proto(&alarm_snap);
  830. }
  831. }
  832. destroy_alarm_log_entry(&alarm_log);
  833. }
  834. if (cnt)
  835. aclk_send_alarm_snapshot(snapshot_proto);
  836. }
  837. netdata_rwlock_unlock(&host->health_log.alarm_log_rwlock);
  838. wc->alerts_snapshot_id = 0;
  839. freez(claim_id);
  840. #endif
  841. return;
  842. }
  843. void sql_aclk_alert_clean_dead_entries(RRDHOST *host)
  844. {
  845. #ifdef ENABLE_NEW_CLOUD_PROTOCOL
  846. if (!claimed())
  847. return;
  848. char uuid_str[GUID_LEN + 1];
  849. uuid_unparse_lower_fix(&host->host_uuid, uuid_str);
  850. BUFFER *sql = buffer_create(1024);
  851. buffer_sprintf(sql,"delete from aclk_alert_%s where alert_unique_id not in "
  852. " (select unique_id from health_log_%s); ", uuid_str, uuid_str);
  853. char *err_msg = NULL;
  854. int rc = sqlite3_exec(db_meta, buffer_tostring(sql), NULL, NULL, &err_msg);
  855. if (rc != SQLITE_OK) {
  856. error_report("Failed when trying to clean stale ACLK alert entries from aclk_alert_%s, error message \"%s""",
  857. uuid_str, err_msg);
  858. sqlite3_free(err_msg);
  859. }
  860. buffer_free(sql);
  861. #else
  862. UNUSED(host);
  863. #endif
  864. }
  865. int get_proto_alert_status(RRDHOST *host, struct proto_alert_status *proto_alert_status)
  866. {
  867. int rc;
  868. struct aclk_database_worker_config *wc = NULL;
  869. wc = (struct aclk_database_worker_config *)host->dbsync_worker;
  870. if (!wc)
  871. return 1;
  872. proto_alert_status->alert_updates = wc->alert_updates;
  873. proto_alert_status->alerts_batch_id = wc->alerts_batch_id;
  874. BUFFER *sql = buffer_create(1024);
  875. sqlite3_stmt *res = NULL;
  876. buffer_sprintf(sql, "SELECT MIN(sequence_id), MAX(sequence_id), " \
  877. "(select MAX(sequence_id) from aclk_alert_%s where date_cloud_ack is not NULL), " \
  878. "(select MAX(sequence_id) from aclk_alert_%s where date_submitted is not NULL) " \
  879. "FROM aclk_alert_%s where date_submitted is null;", wc->uuid_str, wc->uuid_str, wc->uuid_str);
  880. rc = sqlite3_prepare_v2(db_meta, buffer_tostring(sql), -1, &res, 0);
  881. if (rc != SQLITE_OK) {
  882. error_report("Failed to prepare statement to get alert log status from the database.");
  883. buffer_free(sql);
  884. return 1;
  885. }
  886. while (sqlite3_step(res) == SQLITE_ROW) {
  887. proto_alert_status->pending_min_sequence_id = sqlite3_column_bytes(res, 0) > 0 ? (uint64_t) sqlite3_column_int64(res, 0) : 0;
  888. proto_alert_status->pending_max_sequence_id = sqlite3_column_bytes(res, 1) > 0 ? (uint64_t) sqlite3_column_int64(res, 1) : 0;
  889. proto_alert_status->last_acked_sequence_id = sqlite3_column_bytes(res, 2) > 0 ? (uint64_t) sqlite3_column_int64(res, 2) : 0;
  890. proto_alert_status->last_submitted_sequence_id = sqlite3_column_bytes(res, 3) > 0 ? (uint64_t) sqlite3_column_int64(res, 3) : 0;
  891. }
  892. rc = sqlite3_finalize(res);
  893. if (unlikely(rc != SQLITE_OK))
  894. error_report("Failed to finalize statement to get alert log status from the database, rc = %d", rc);
  895. buffer_free(sql);
  896. return 0;
  897. }