sqlite_aclk.c 22 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678
  1. // SPDX-License-Identifier: GPL-3.0-or-later
  2. #include "sqlite_functions.h"
  3. #include "sqlite_aclk.h"
  4. #include "sqlite_aclk_node.h"
  5. struct aclk_sync_config_s {
  6. uv_thread_t thread;
  7. uv_loop_t loop;
  8. uv_timer_t timer_req;
  9. time_t cleanup_after; // Start a cleanup after this timestamp
  10. uv_async_t async;
  11. bool initialized;
  12. SPINLOCK cmd_queue_lock;
  13. struct aclk_database_cmd *cmd_base;
  14. } aclk_sync_config = { 0 };
  15. void sanity_check(void) {
  16. // make sure the compiler will stop on misconfigurations
  17. BUILD_BUG_ON(WORKER_UTILIZATION_MAX_JOB_TYPES < ACLK_MAX_ENUMERATIONS_DEFINED);
  18. }
  19. static struct aclk_database_cmd aclk_database_deq_cmd(void)
  20. {
  21. struct aclk_database_cmd ret;
  22. spinlock_lock(&aclk_sync_config.cmd_queue_lock);
  23. if(aclk_sync_config.cmd_base) {
  24. struct aclk_database_cmd *t = aclk_sync_config.cmd_base;
  25. DOUBLE_LINKED_LIST_REMOVE_ITEM_UNSAFE(aclk_sync_config.cmd_base, t, prev, next);
  26. ret = *t;
  27. freez(t);
  28. }
  29. else {
  30. ret.opcode = ACLK_DATABASE_NOOP;
  31. ret.completion = NULL;
  32. }
  33. spinlock_unlock(&aclk_sync_config.cmd_queue_lock);
  34. return ret;
  35. }
  36. static void aclk_database_enq_cmd(struct aclk_database_cmd *cmd)
  37. {
  38. struct aclk_database_cmd *t = mallocz(sizeof(*t));
  39. *t = *cmd;
  40. t->prev = t->next = NULL;
  41. spinlock_lock(&aclk_sync_config.cmd_queue_lock);
  42. DOUBLE_LINKED_LIST_APPEND_ITEM_UNSAFE(aclk_sync_config.cmd_base, t, prev, next);
  43. spinlock_unlock(&aclk_sync_config.cmd_queue_lock);
  44. (void) uv_async_send(&aclk_sync_config.async);
  45. }
  46. enum {
  47. IDX_HOST_ID,
  48. IDX_HOSTNAME,
  49. IDX_REGISTRY,
  50. IDX_UPDATE_EVERY,
  51. IDX_OS,
  52. IDX_TIMEZONE,
  53. IDX_TAGS,
  54. IDX_HOPS,
  55. IDX_MEMORY_MODE,
  56. IDX_ABBREV_TIMEZONE,
  57. IDX_UTC_OFFSET,
  58. IDX_PROGRAM_NAME,
  59. IDX_PROGRAM_VERSION,
  60. IDX_ENTRIES,
  61. IDX_HEALTH_ENABLED,
  62. IDX_LAST_CONNECTED,
  63. IDX_IS_EPHEMERAL,
  64. };
  65. static int create_host_callback(void *data, int argc, char **argv, char **column)
  66. {
  67. int *number_of_chidren = data;
  68. UNUSED(argc);
  69. UNUSED(column);
  70. time_t last_connected =
  71. (time_t)(argv[IDX_LAST_CONNECTED] ? str2uint64_t(argv[IDX_LAST_CONNECTED], NULL) : 0);
  72. if (!last_connected)
  73. last_connected = now_realtime_sec();
  74. time_t age = now_realtime_sec() - last_connected;
  75. int is_ephemeral = 0;
  76. if (argv[IDX_IS_EPHEMERAL])
  77. is_ephemeral = str2i(argv[IDX_IS_EPHEMERAL]);
  78. char guid[UUID_STR_LEN];
  79. uuid_unparse_lower(*(uuid_t *)argv[IDX_HOST_ID], guid);
  80. if (is_ephemeral && age > rrdhost_free_ephemeral_time_s) {
  81. netdata_log_info(
  82. "Skipping ephemeral hostname \"%s\" with GUID \"%s\", age = %ld seconds (limit %ld seconds)",
  83. (const char *)argv[IDX_HOSTNAME],
  84. guid,
  85. age,
  86. rrdhost_free_ephemeral_time_s);
  87. return 0;
  88. }
  89. struct rrdhost_system_info *system_info = callocz(1, sizeof(struct rrdhost_system_info));
  90. __atomic_sub_fetch(&netdata_buffers_statistics.rrdhost_allocations_size, sizeof(struct rrdhost_system_info), __ATOMIC_RELAXED);
  91. system_info->hops = str2i((const char *) argv[IDX_HOPS]);
  92. sql_build_host_system_info((uuid_t *)argv[IDX_HOST_ID], system_info);
  93. RRDHOST *host = rrdhost_find_or_create(
  94. (const char *)argv[IDX_HOSTNAME],
  95. (const char *)argv[IDX_REGISTRY],
  96. guid,
  97. (const char *)argv[IDX_OS],
  98. (const char *)argv[IDX_TIMEZONE],
  99. (const char *)argv[IDX_ABBREV_TIMEZONE],
  100. (int32_t)(argv[IDX_UTC_OFFSET] ? str2uint32_t(argv[IDX_UTC_OFFSET], NULL) : 0),
  101. (const char *)argv[IDX_TAGS],
  102. (const char *)(argv[IDX_PROGRAM_NAME] ? argv[IDX_PROGRAM_NAME] : "unknown"),
  103. (const char *)(argv[IDX_PROGRAM_VERSION] ? argv[IDX_PROGRAM_VERSION] : "unknown"),
  104. argv[IDX_UPDATE_EVERY] ? str2i(argv[IDX_UPDATE_EVERY]) : 1,
  105. argv[IDX_ENTRIES] ? str2i(argv[IDX_ENTRIES]) : 0,
  106. default_rrd_memory_mode,
  107. 0 // health
  108. ,
  109. 0 // rrdpush enabled
  110. ,
  111. NULL //destination
  112. ,
  113. NULL // api key
  114. ,
  115. NULL // send charts matching
  116. ,
  117. false // rrdpush_enable_replication
  118. ,
  119. 0 // rrdpush_seconds_to_replicate
  120. ,
  121. 0 // rrdpush_replication_step
  122. ,
  123. system_info,
  124. 1);
  125. if (likely(host)) {
  126. if (is_ephemeral)
  127. rrdhost_option_set(host, RRDHOST_OPTION_EPHEMERAL_HOST);
  128. if (is_ephemeral)
  129. host->child_disconnected_time = now_realtime_sec();
  130. host->rrdlabels = sql_load_host_labels((uuid_t *)argv[IDX_HOST_ID]);
  131. host->last_connected = last_connected;
  132. }
  133. (*number_of_chidren)++;
  134. #ifdef NETDATA_INTERNAL_CHECKS
  135. char node_str[UUID_STR_LEN] = "<none>";
  136. if (likely(host->node_id))
  137. uuid_unparse_lower(*host->node_id, node_str);
  138. internal_error(true, "Adding archived host \"%s\" with GUID \"%s\" node id = \"%s\" ephemeral=%d", rrdhost_hostname(host), host->machine_guid, node_str, is_ephemeral);
  139. #endif
  140. return 0;
  141. }
  142. #ifdef ENABLE_ACLK
  143. #define SQL_SELECT_HOST_BY_UUID "SELECT host_id FROM host WHERE host_id = @host_id"
  144. static int is_host_available(uuid_t *host_id)
  145. {
  146. sqlite3_stmt *res = NULL;
  147. int rc;
  148. if (unlikely(!db_meta)) {
  149. if (default_rrd_memory_mode == RRD_MEMORY_MODE_DBENGINE)
  150. error_report("Database has not been initialized");
  151. return 1;
  152. }
  153. rc = sqlite3_prepare_v2(db_meta, SQL_SELECT_HOST_BY_UUID, -1, &res, 0);
  154. if (unlikely(rc != SQLITE_OK)) {
  155. error_report("Failed to prepare statement to select node instance information for a node");
  156. return 1;
  157. }
  158. rc = sqlite3_bind_blob(res, 1, host_id, sizeof(*host_id), SQLITE_STATIC);
  159. if (unlikely(rc != SQLITE_OK)) {
  160. error_report("Failed to bind host_id parameter to check host existence");
  161. goto failed;
  162. }
  163. rc = sqlite3_step_monitored(res);
  164. failed:
  165. if (unlikely(sqlite3_finalize(res) != SQLITE_OK))
  166. error_report("Failed to finalize the prepared statement when checking host existence");
  167. return (rc == SQLITE_ROW);
  168. }
  169. // OPCODE: ACLK_DATABASE_DELETE_HOST
  170. static void sql_delete_aclk_table_list(char *host_guid)
  171. {
  172. char uuid_str[UUID_STR_LEN];
  173. char host_str[UUID_STR_LEN];
  174. int rc;
  175. uuid_t host_uuid;
  176. if (unlikely(!host_guid))
  177. return;
  178. rc = uuid_parse(host_guid, host_uuid);
  179. freez(host_guid);
  180. if (rc)
  181. return;
  182. uuid_unparse_lower(host_uuid, host_str);
  183. uuid_unparse_lower_fix(&host_uuid, uuid_str);
  184. if (is_host_available(&host_uuid))
  185. return;
  186. sqlite3_stmt *res = NULL;
  187. BUFFER *sql = buffer_create(ACLK_SYNC_QUERY_SIZE, &netdata_buffers_statistics.buffers_sqlite);
  188. buffer_sprintf(sql,"SELECT 'drop '||type||' IF EXISTS '||name||';' FROM sqlite_schema " \
  189. "WHERE name LIKE 'aclk_%%_%s' AND type IN ('table', 'trigger', 'index')", uuid_str);
  190. rc = sqlite3_prepare_v2(db_meta, buffer_tostring(sql), -1, &res, 0);
  191. if (rc != SQLITE_OK) {
  192. error_report("Failed to prepare statement to clean up aclk tables");
  193. goto fail;
  194. }
  195. buffer_flush(sql);
  196. while (sqlite3_step_monitored(res) == SQLITE_ROW)
  197. buffer_strcat(sql, (char *) sqlite3_column_text(res, 0));
  198. rc = sqlite3_finalize(res);
  199. if (unlikely(rc != SQLITE_OK))
  200. error_report("Failed to finalize statement to clean up aclk tables, rc = %d", rc);
  201. rc = db_execute(db_meta, buffer_tostring(sql));
  202. if (unlikely(rc))
  203. netdata_log_error("Failed to drop unused ACLK tables");
  204. fail:
  205. buffer_free(sql);
  206. }
  207. // OPCODE: ACLK_DATABASE_NODE_UNREGISTER
  208. static void sql_unregister_node(char *machine_guid)
  209. {
  210. int rc;
  211. uuid_t host_uuid;
  212. if (unlikely(!machine_guid))
  213. return;
  214. rc = uuid_parse(machine_guid, host_uuid);
  215. if (rc) {
  216. freez(machine_guid);
  217. return;
  218. }
  219. sqlite3_stmt *res = NULL;
  220. rc = sqlite3_prepare_v2(db_meta, "UPDATE node_instance SET node_id = NULL WHERE host_id = @host_id", -1, &res, 0);
  221. if (unlikely(rc != SQLITE_OK)) {
  222. error_report("Failed to prepare statement to remove the host node id");
  223. freez(machine_guid);
  224. return;
  225. }
  226. rc = sqlite3_bind_blob(res, 1, &host_uuid, sizeof(host_uuid), SQLITE_STATIC);
  227. if (unlikely(rc != SQLITE_OK)) {
  228. error_report("Failed to bind host_id parameter to remove host node id");
  229. goto skip;
  230. }
  231. rc = sqlite3_step_monitored(res);
  232. if (unlikely(rc != SQLITE_DONE)) {
  233. error_report("Failed to execute command to remove host node id");
  234. } else {
  235. // node: machine guid will be freed after processing
  236. metadata_delete_host_chart_labels(machine_guid);
  237. machine_guid = NULL;
  238. }
  239. skip:
  240. if (unlikely(sqlite3_finalize(res) != SQLITE_OK))
  241. error_report("Failed to finalize statement to remove host node id");
  242. freez(machine_guid);
  243. }
  244. static int sql_check_aclk_table(void *data __maybe_unused, int argc __maybe_unused, char **argv __maybe_unused, char **column __maybe_unused)
  245. {
  246. struct aclk_database_cmd cmd;
  247. memset(&cmd, 0, sizeof(cmd));
  248. cmd.opcode = ACLK_DATABASE_DELETE_HOST;
  249. cmd.param[0] = strdupz((char *) argv[0]);
  250. aclk_database_enq_cmd(&cmd);
  251. return 0;
  252. }
  253. #define SQL_SELECT_ACLK_ACTIVE_LIST "SELECT REPLACE(SUBSTR(name,19),'_','-') FROM sqlite_schema " \
  254. "WHERE name LIKE 'aclk_chart_latest_%' AND type IN ('table')"
  255. static void sql_check_aclk_table_list(void)
  256. {
  257. char *err_msg = NULL;
  258. int rc = sqlite3_exec_monitored(db_meta, SQL_SELECT_ACLK_ACTIVE_LIST, sql_check_aclk_table, NULL, &err_msg);
  259. if (rc != SQLITE_OK) {
  260. error_report("Query failed when trying to check for obsolete ACLK sync tables, %s", err_msg);
  261. sqlite3_free(err_msg);
  262. }
  263. }
  264. #define SQL_ALERT_CLEANUP "DELETE FROM aclk_alert_%s WHERE date_submitted IS NOT NULL AND CAST(date_cloud_ack AS INT) < unixepoch()-%d"
  265. static int sql_maint_aclk_sync_database(void *data __maybe_unused, int argc __maybe_unused, char **argv, char **column __maybe_unused)
  266. {
  267. char sql[ACLK_SYNC_QUERY_SIZE];
  268. snprintfz(sql,sizeof(sql) - 1, SQL_ALERT_CLEANUP, (char *) argv[0], ACLK_DELETE_ACK_ALERTS_INTERNAL);
  269. if (unlikely(db_execute(db_meta, sql)))
  270. error_report("Failed to clean stale ACLK alert entries");
  271. return 0;
  272. }
  273. #define SQL_SELECT_ACLK_ALERT_LIST "SELECT SUBSTR(name,12) FROM sqlite_schema WHERE name LIKE 'aclk_alert_%' AND type IN ('table')"
  274. static void sql_maint_aclk_sync_database_all(void)
  275. {
  276. char *err_msg = NULL;
  277. int rc = sqlite3_exec_monitored(db_meta, SQL_SELECT_ACLK_ALERT_LIST, sql_maint_aclk_sync_database, NULL, &err_msg);
  278. if (rc != SQLITE_OK) {
  279. error_report("Query failed when trying to check for obsolete ACLK sync tables, %s", err_msg);
  280. sqlite3_free(err_msg);
  281. }
  282. }
  283. static int aclk_config_parameters(void *data __maybe_unused, int argc __maybe_unused, char **argv, char **column __maybe_unused)
  284. {
  285. char uuid_str[UUID_STR_LEN];
  286. uuid_unparse_lower(*((uuid_t *) argv[0]), uuid_str);
  287. RRDHOST *host = rrdhost_find_by_guid(uuid_str);
  288. if (host == localhost)
  289. return 0;
  290. sql_create_aclk_table(host, (uuid_t *) argv[0], (uuid_t *) argv[1]);
  291. return 0;
  292. }
  293. static void async_cb(uv_async_t *handle)
  294. {
  295. uv_stop(handle->loop);
  296. uv_update_time(handle->loop);
  297. }
  298. #define TIMER_PERIOD_MS (1000)
  299. static void timer_cb(uv_timer_t *handle)
  300. {
  301. uv_stop(handle->loop);
  302. uv_update_time(handle->loop);
  303. struct aclk_sync_config_s *config = handle->data;
  304. struct aclk_database_cmd cmd;
  305. memset(&cmd, 0, sizeof(cmd));
  306. if (config->cleanup_after < now_realtime_sec()) {
  307. cmd.opcode = ACLK_DATABASE_CLEANUP;
  308. aclk_database_enq_cmd(&cmd);
  309. config->cleanup_after += ACLK_DATABASE_CLEANUP_INTERVAL;
  310. }
  311. if (aclk_connected) {
  312. cmd.opcode = ACLK_DATABASE_PUSH_ALERT;
  313. aclk_database_enq_cmd(&cmd);
  314. aclk_check_node_info_and_collectors();
  315. }
  316. }
  317. static void aclk_synchronization(void *arg __maybe_unused)
  318. {
  319. struct aclk_sync_config_s *config = arg;
  320. uv_thread_set_name_np(config->thread, "ACLKSYNC");
  321. worker_register("ACLKSYNC");
  322. service_register(SERVICE_THREAD_TYPE_EVENT_LOOP, NULL, NULL, NULL, true);
  323. worker_register_job_name(ACLK_DATABASE_NOOP, "noop");
  324. worker_register_job_name(ACLK_DATABASE_CLEANUP, "cleanup");
  325. worker_register_job_name(ACLK_DATABASE_DELETE_HOST, "node delete");
  326. worker_register_job_name(ACLK_DATABASE_NODE_STATE, "node state");
  327. worker_register_job_name(ACLK_DATABASE_PUSH_ALERT, "alert push");
  328. worker_register_job_name(ACLK_DATABASE_PUSH_ALERT_CONFIG, "alert conf push");
  329. worker_register_job_name(ACLK_DATABASE_PUSH_ALERT_CHECKPOINT,"alert checkpoint");
  330. worker_register_job_name(ACLK_DATABASE_PUSH_ALERT_SNAPSHOT, "alert snapshot");
  331. worker_register_job_name(ACLK_DATABASE_QUEUE_REMOVED_ALERTS, "alerts check");
  332. worker_register_job_name(ACLK_DATABASE_TIMER, "timer");
  333. uv_loop_t *loop = &config->loop;
  334. fatal_assert(0 == uv_loop_init(loop));
  335. fatal_assert(0 == uv_async_init(loop, &config->async, async_cb));
  336. fatal_assert(0 == uv_timer_init(loop, &config->timer_req));
  337. config->timer_req.data = config;
  338. fatal_assert(0 == uv_timer_start(&config->timer_req, timer_cb, TIMER_PERIOD_MS, TIMER_PERIOD_MS));
  339. netdata_log_info("Starting ACLK synchronization thread");
  340. config->cleanup_after = now_realtime_sec() + ACLK_DATABASE_CLEANUP_FIRST;
  341. config->initialized = true;
  342. while (likely(service_running(SERVICE_ACLKSYNC))) {
  343. enum aclk_database_opcode opcode;
  344. worker_is_idle();
  345. uv_run(loop, UV_RUN_DEFAULT);
  346. /* wait for commands */
  347. do {
  348. struct aclk_database_cmd cmd = aclk_database_deq_cmd();
  349. if (unlikely(!service_running(SERVICE_ACLKSYNC)))
  350. break;
  351. opcode = cmd.opcode;
  352. if(likely(opcode != ACLK_DATABASE_NOOP))
  353. worker_is_busy(opcode);
  354. switch (opcode) {
  355. case ACLK_DATABASE_NOOP:
  356. /* the command queue was empty, do nothing */
  357. break;
  358. // MAINTENANCE
  359. case ACLK_DATABASE_CLEANUP:
  360. // Scan all aclk_alert_ tables and cleanup as needed
  361. sql_maint_aclk_sync_database_all();
  362. sql_check_aclk_table_list();
  363. break;
  364. case ACLK_DATABASE_DELETE_HOST:
  365. sql_delete_aclk_table_list(cmd.param[0]);
  366. break;
  367. // NODE STATE
  368. case ACLK_DATABASE_NODE_STATE:;
  369. RRDHOST *host = cmd.param[0];
  370. int live = (host == localhost || host->receiver || !(rrdhost_flag_check(host, RRDHOST_FLAG_ORPHAN))) ? 1 : 0;
  371. struct aclk_sync_cfg_t *ahc = host->aclk_config;
  372. if (unlikely(!ahc))
  373. sql_create_aclk_table(host, &host->host_uuid, host->node_id);
  374. aclk_host_state_update(host, live, 1);
  375. break;
  376. case ACLK_DATABASE_NODE_UNREGISTER:
  377. sql_unregister_node(cmd.param[0]);
  378. break;
  379. // ALERTS
  380. case ACLK_DATABASE_PUSH_ALERT_CONFIG:
  381. aclk_push_alert_config_event(cmd.param[0], cmd.param[1]);
  382. break;
  383. case ACLK_DATABASE_PUSH_ALERT:
  384. aclk_push_alert_events_for_all_hosts();
  385. break;
  386. case ACLK_DATABASE_PUSH_ALERT_SNAPSHOT:;
  387. aclk_push_alert_snapshot_event(cmd.param[0]);
  388. break;
  389. case ACLK_DATABASE_QUEUE_REMOVED_ALERTS:
  390. sql_process_queue_removed_alerts_to_aclk(cmd.param[0]);
  391. break;
  392. default:
  393. break;
  394. }
  395. if (cmd.completion)
  396. completion_mark_complete(cmd.completion);
  397. } while (opcode != ACLK_DATABASE_NOOP);
  398. }
  399. if (!uv_timer_stop(&config->timer_req))
  400. uv_close((uv_handle_t *)&config->timer_req, NULL);
  401. uv_close((uv_handle_t *)&config->async, NULL);
  402. (void) uv_loop_close(loop);
  403. worker_unregister();
  404. service_exits();
  405. netdata_log_info("ACLK SYNC: Shutting down ACLK synchronization event loop");
  406. }
  407. static void aclk_synchronization_init(void)
  408. {
  409. memset(&aclk_sync_config, 0, sizeof(aclk_sync_config));
  410. fatal_assert(0 == uv_thread_create(&aclk_sync_config.thread, aclk_synchronization, &aclk_sync_config));
  411. }
  412. #endif
  413. // -------------------------------------------------------------
  414. void sql_create_aclk_table(RRDHOST *host __maybe_unused, uuid_t *host_uuid __maybe_unused, uuid_t *node_id __maybe_unused)
  415. {
  416. #ifdef ENABLE_ACLK
  417. char uuid_str[UUID_STR_LEN];
  418. char host_guid[UUID_STR_LEN];
  419. int rc;
  420. uuid_unparse_lower_fix(host_uuid, uuid_str);
  421. uuid_unparse_lower(*host_uuid, host_guid);
  422. char sql[ACLK_SYNC_QUERY_SIZE];
  423. snprintfz(sql, sizeof(sql) - 1, TABLE_ACLK_ALERT, uuid_str);
  424. rc = db_execute(db_meta, sql);
  425. if (unlikely(rc))
  426. error_report("Failed to create ACLK alert table for host %s", host ? rrdhost_hostname(host) : host_guid);
  427. else {
  428. snprintfz(sql, sizeof(sql) - 1, INDEX_ACLK_ALERT1, uuid_str, uuid_str);
  429. rc = db_execute(db_meta, sql);
  430. if (unlikely(rc))
  431. error_report(
  432. "Failed to create ACLK alert table index 1 for host %s", host ? string2str(host->hostname) : host_guid);
  433. snprintfz(sql, sizeof(sql) - 1, INDEX_ACLK_ALERT2, uuid_str, uuid_str);
  434. rc = db_execute(db_meta, sql);
  435. if (unlikely(rc))
  436. error_report(
  437. "Failed to create ACLK alert table index 2 for host %s", host ? string2str(host->hostname) : host_guid);
  438. }
  439. if (likely(host) && unlikely(host->aclk_config))
  440. return;
  441. if (unlikely(!host))
  442. return;
  443. struct aclk_sync_cfg_t *wc = callocz(1, sizeof(struct aclk_sync_cfg_t));
  444. if (node_id && !uuid_is_null(*node_id))
  445. uuid_unparse_lower(*node_id, wc->node_id);
  446. host->aclk_config = wc;
  447. if (node_id && !host->node_id) {
  448. host->node_id = mallocz(sizeof(*host->node_id));
  449. uuid_copy(*host->node_id, *node_id);
  450. }
  451. wc->host = host;
  452. strcpy(wc->uuid_str, uuid_str);
  453. wc->alert_updates = 0;
  454. time_t now = now_realtime_sec();
  455. wc->node_info_send_time = (host == localhost || NULL == localhost) ? now - 25 : now;
  456. #endif
  457. }
  458. #define SQL_FETCH_ALL_HOSTS \
  459. "SELECT host_id, hostname, registry_hostname, update_every, os, " \
  460. "timezone, tags, hops, memory_mode, abbrev_timezone, utc_offset, program_name, " \
  461. "program_version, entries, health_enabled, last_connected, " \
  462. "(SELECT CASE WHEN hl.label_value = 'true' THEN 1 ELSE 0 END FROM " \
  463. "host_label hl WHERE hl.host_id = h.host_id AND hl.label_key = '_is_ephemeral') " \
  464. "FROM host h WHERE hops > 0"
  465. #define SQL_FETCH_ALL_INSTANCES \
  466. "SELECT ni.host_id, ni.node_id FROM host h, node_instance ni " \
  467. "WHERE h.host_id = ni.host_id AND ni.node_id IS NOT NULL"
  468. void sql_aclk_sync_init(void)
  469. {
  470. char *err_msg = NULL;
  471. int rc;
  472. if (unlikely(!db_meta)) {
  473. if (default_rrd_memory_mode != RRD_MEMORY_MODE_DBENGINE) {
  474. return;
  475. }
  476. error_report("Database has not been initialized");
  477. return;
  478. }
  479. netdata_log_info("Creating archived hosts");
  480. int number_of_children = 0;
  481. rc = sqlite3_exec_monitored(db_meta, SQL_FETCH_ALL_HOSTS, create_host_callback, &number_of_children, &err_msg);
  482. if (rc != SQLITE_OK) {
  483. error_report("SQLite error when loading archived hosts, rc = %d (%s)", rc, err_msg);
  484. sqlite3_free(err_msg);
  485. }
  486. netdata_log_info("Created %d archived hosts", number_of_children);
  487. // Trigger host context load for hosts that have been created
  488. metadata_queue_load_host_context(NULL);
  489. #ifdef ENABLE_ACLK
  490. if (!number_of_children)
  491. aclk_queue_node_info(localhost, true);
  492. rc = sqlite3_exec_monitored(db_meta, SQL_FETCH_ALL_INSTANCES,aclk_config_parameters, NULL,&err_msg);
  493. if (rc != SQLITE_OK) {
  494. error_report("SQLite error when configuring host ACLK synchonization parameters, rc = %d (%s)", rc, err_msg);
  495. sqlite3_free(err_msg);
  496. }
  497. aclk_synchronization_init();
  498. netdata_log_info("ACLK sync initialization completed");
  499. #endif
  500. }
  501. // Public
  502. static inline void queue_aclk_sync_cmd(enum aclk_database_opcode opcode, const void *param0, const void *param1)
  503. {
  504. struct aclk_database_cmd cmd;
  505. cmd.opcode = opcode;
  506. cmd.param[0] = (void *) param0;
  507. cmd.param[1] = (void *) param1;
  508. cmd.completion = NULL;
  509. aclk_database_enq_cmd(&cmd);
  510. }
  511. // Public
  512. void aclk_push_alert_config(const char *node_id, const char *config_hash)
  513. {
  514. if (unlikely(!aclk_sync_config.initialized))
  515. return;
  516. queue_aclk_sync_cmd(ACLK_DATABASE_PUSH_ALERT_CONFIG, strdupz(node_id), strdupz(config_hash));
  517. }
  518. void aclk_push_node_alert_snapshot(const char *node_id)
  519. {
  520. if (unlikely(!aclk_sync_config.initialized))
  521. return;
  522. queue_aclk_sync_cmd(ACLK_DATABASE_PUSH_ALERT_SNAPSHOT, strdupz(node_id), NULL);
  523. }
  524. void aclk_push_node_removed_alerts(const char *node_id)
  525. {
  526. if (unlikely(!aclk_sync_config.initialized))
  527. return;
  528. queue_aclk_sync_cmd(ACLK_DATABASE_QUEUE_REMOVED_ALERTS, strdupz(node_id), NULL);
  529. }
  530. void schedule_node_info_update(RRDHOST *host __maybe_unused)
  531. {
  532. #ifdef ENABLE_ACLK
  533. if (unlikely(!host))
  534. return;
  535. struct aclk_database_cmd cmd;
  536. memset(&cmd, 0, sizeof(cmd));
  537. cmd.opcode = ACLK_DATABASE_NODE_STATE;
  538. cmd.param[0] = host;
  539. cmd.completion = NULL;
  540. aclk_database_enq_cmd(&cmd);
  541. #endif
  542. }
  543. #ifdef ENABLE_ACLK
  544. void unregister_node(const char *machine_guid)
  545. {
  546. if (unlikely(!machine_guid))
  547. return;
  548. struct aclk_database_cmd cmd;
  549. memset(&cmd, 0, sizeof(cmd));
  550. cmd.opcode = ACLK_DATABASE_NODE_UNREGISTER;
  551. cmd.param[0] = strdupz(machine_guid);
  552. cmd.completion = NULL;
  553. aclk_database_enq_cmd(&cmd);
  554. }
  555. #endif