service.c 13 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370
  1. // SPDX-License-Identifier: GPL-3.0-or-later
  2. #include "common.h"
  3. /* Run service jobs every X seconds */
  4. #define SERVICE_HEARTBEAT 10
  5. #define TIME_TO_RUN_OBSOLETIONS_ON_CHILD_CONNECT (3600 / 2)
  6. #define ITERATIONS_TO_RUN_OBSOLETIONS_ON_CHILD_CONNECT 60
  7. #define WORKER_JOB_CHILD_CHART_OBSOLETION_CHECK 1
  8. #define WORKER_JOB_CLEANUP_OBSOLETE_CHARTS 2
  9. #define WORKER_JOB_ARCHIVE_CHART 3
  10. #define WORKER_JOB_ARCHIVE_CHART_DIMENSIONS 4
  11. #define WORKER_JOB_ARCHIVE_DIMENSION 5
  12. #define WORKER_JOB_CLEANUP_ORPHAN_HOSTS 6
  13. #define WORKER_JOB_CLEANUP_OBSOLETE_CHARTS_ON_HOSTS 7
  14. #define WORKER_JOB_FREE_HOST 9
  15. #define WORKER_JOB_SAVE_HOST_CHARTS 10
  16. #define WORKER_JOB_DELETE_HOST_CHARTS 11
  17. #define WORKER_JOB_FREE_CHART 12
  18. #define WORKER_JOB_SAVE_CHART 13
  19. #define WORKER_JOB_DELETE_CHART 14
  20. #define WORKER_JOB_FREE_DIMENSION 15
  21. #define WORKER_JOB_PGC_MAIN_EVICT 16
  22. #define WORKER_JOB_PGC_MAIN_FLUSH 17
  23. #define WORKER_JOB_PGC_OPEN_EVICT 18
  24. #define WORKER_JOB_PGC_OPEN_FLUSH 19
  25. static void svc_rrddim_obsolete_to_archive(RRDDIM *rd) {
  26. RRDSET *st = rd->rrdset;
  27. if(rrddim_flag_check(rd, RRDDIM_FLAG_ARCHIVED) || !rrddim_flag_check(rd, RRDDIM_FLAG_OBSOLETE))
  28. return;
  29. worker_is_busy(WORKER_JOB_ARCHIVE_DIMENSION);
  30. rrddim_flag_set(rd, RRDDIM_FLAG_ARCHIVED);
  31. rrddim_flag_clear(rd, RRDDIM_FLAG_OBSOLETE);
  32. const char *cache_filename = rrddim_cache_filename(rd);
  33. if(cache_filename) {
  34. netdata_log_info("Deleting dimension file '%s'.", cache_filename);
  35. if (unlikely(unlink(cache_filename) == -1))
  36. netdata_log_error("Cannot delete dimension file '%s'", cache_filename);
  37. }
  38. if (rd->rrd_memory_mode == RRD_MEMORY_MODE_DBENGINE) {
  39. rrddimvar_delete_all(rd);
  40. /* only a collector can mark a chart as obsolete, so we must remove the reference */
  41. size_t tiers_available = 0, tiers_said_no_retention = 0;
  42. for(size_t tier = 0; tier < storage_tiers ;tier++) {
  43. if(rd->tiers[tier].db_collection_handle) {
  44. tiers_available++;
  45. if(storage_engine_store_finalize(rd->tiers[tier].db_collection_handle))
  46. tiers_said_no_retention++;
  47. rd->tiers[tier].db_collection_handle = NULL;
  48. }
  49. }
  50. if (tiers_available == tiers_said_no_retention && tiers_said_no_retention) {
  51. /* This metric has no data and no references */
  52. metaqueue_delete_dimension_uuid(&rd->metric_uuid);
  53. }
  54. else {
  55. /* Do not delete this dimension */
  56. return;
  57. }
  58. }
  59. worker_is_busy(WORKER_JOB_FREE_DIMENSION);
  60. rrddim_free(st, rd);
  61. }
  62. static inline bool svc_rrdset_archive_obsolete_dimensions(RRDSET *st, bool all_dimensions) {
  63. if(!all_dimensions && !rrdset_flag_check(st, RRDSET_FLAG_OBSOLETE_DIMENSIONS))
  64. return true;
  65. worker_is_busy(WORKER_JOB_ARCHIVE_CHART_DIMENSIONS);
  66. rrdset_flag_clear(st, RRDSET_FLAG_OBSOLETE_DIMENSIONS);
  67. RRDDIM *rd;
  68. time_t now = now_realtime_sec();
  69. size_t dim_candidates = 0;
  70. size_t dim_archives = 0;
  71. dfe_start_write(st->rrddim_root_index, rd) {
  72. bool candidate = (all_dimensions || rrddim_flag_check(rd, RRDDIM_FLAG_OBSOLETE));
  73. if(candidate) {
  74. dim_candidates++;
  75. if(rd->collector.last_collected_time.tv_sec + rrdset_free_obsolete_time_s < now) {
  76. size_t references = dictionary_acquired_item_references(rd_dfe.item);
  77. if(references == 1) {
  78. // netdata_log_info("Removing obsolete dimension 'host:%s/chart:%s/dim:%s'",
  79. // rrdhost_hostname(st->rrdhost), rrdset_id(st), rrddim_id(rd));
  80. svc_rrddim_obsolete_to_archive(rd);
  81. dim_archives++;
  82. }
  83. // else
  84. // netdata_log_info("Cannot remove obsolete dimension 'host:%s/chart:%s/dim:%s'",
  85. // rrdhost_hostname(st->rrdhost), rrdset_id(st), rrddim_id(rd));
  86. }
  87. }
  88. }
  89. dfe_done(rd);
  90. if(dim_archives != dim_candidates) {
  91. rrdset_flag_set(st, RRDSET_FLAG_OBSOLETE_DIMENSIONS);
  92. return false;
  93. }
  94. return true;
  95. }
  96. static void svc_rrdset_obsolete_to_free(RRDSET *st) {
  97. if(!svc_rrdset_archive_obsolete_dimensions(st, true))
  98. return;
  99. worker_is_busy(WORKER_JOB_FREE_CHART);
  100. rrdcalc_unlink_all_rrdset_alerts(st);
  101. rrdsetvar_release_and_delete_all(st);
  102. // has to be run after all dimensions are archived - or use-after-free will occur
  103. rrdvar_delete_all(st->rrdvars);
  104. if(st->rrd_memory_mode != RRD_MEMORY_MODE_DBENGINE) {
  105. if(rrdhost_option_check(st->rrdhost, RRDHOST_OPTION_DELETE_OBSOLETE_CHARTS)) {
  106. worker_is_busy(WORKER_JOB_DELETE_CHART);
  107. rrdset_delete_files(st);
  108. }
  109. else {
  110. worker_is_busy(WORKER_JOB_SAVE_CHART);
  111. rrdset_save(st);
  112. }
  113. }
  114. rrdset_free(st);
  115. }
  116. static inline void svc_rrdhost_cleanup_charts_marked_obsolete(RRDHOST *host) {
  117. if(!rrdhost_flag_check(host, RRDHOST_FLAG_PENDING_OBSOLETE_CHARTS|RRDHOST_FLAG_PENDING_OBSOLETE_DIMENSIONS))
  118. return;
  119. worker_is_busy(WORKER_JOB_CLEANUP_OBSOLETE_CHARTS);
  120. rrdhost_flag_clear(host, RRDHOST_FLAG_PENDING_OBSOLETE_CHARTS|RRDHOST_FLAG_PENDING_OBSOLETE_DIMENSIONS);
  121. size_t full_candidates = 0;
  122. size_t full_archives = 0;
  123. size_t partial_candidates = 0;
  124. size_t partial_archives = 0;
  125. time_t now = now_realtime_sec();
  126. RRDSET *st;
  127. rrdset_foreach_reentrant(st, host) {
  128. if(rrdset_is_replicating(st))
  129. continue;
  130. RRDSET_FLAGS flags = rrdset_flag_get(st);
  131. bool obsolete_chart = flags & RRDSET_FLAG_OBSOLETE;
  132. bool obsolete_dims = flags & RRDSET_FLAG_OBSOLETE_DIMENSIONS;
  133. if(obsolete_dims) {
  134. partial_candidates++;
  135. if(svc_rrdset_archive_obsolete_dimensions(st, false))
  136. partial_archives++;
  137. }
  138. if(obsolete_chart) {
  139. full_candidates++;
  140. if(unlikely( st->last_accessed_time_s + rrdset_free_obsolete_time_s < now
  141. && st->last_updated.tv_sec + rrdset_free_obsolete_time_s < now
  142. && st->last_collected_time.tv_sec + rrdset_free_obsolete_time_s < now
  143. )) {
  144. svc_rrdset_obsolete_to_free(st);
  145. full_archives++;
  146. }
  147. }
  148. }
  149. rrdset_foreach_done(st);
  150. if(partial_archives != partial_candidates)
  151. rrdhost_flag_set(host, RRDHOST_FLAG_PENDING_OBSOLETE_DIMENSIONS);
  152. if(full_archives != full_candidates)
  153. rrdhost_flag_set(host, RRDHOST_FLAG_PENDING_OBSOLETE_CHARTS);
  154. }
  155. static void svc_rrdhost_detect_obsolete_charts(RRDHOST *host) {
  156. worker_is_busy(WORKER_JOB_CHILD_CHART_OBSOLETION_CHECK);
  157. time_t now = now_realtime_sec();
  158. time_t last_entry_t;
  159. RRDSET *st;
  160. time_t child_connect_time = host->child_connect_time;
  161. rrdset_foreach_read(st, host) {
  162. if(rrdset_is_replicating(st))
  163. continue;
  164. last_entry_t = rrdset_last_entry_s(st);
  165. if (last_entry_t && last_entry_t < child_connect_time &&
  166. child_connect_time + TIME_TO_RUN_OBSOLETIONS_ON_CHILD_CONNECT +
  167. (ITERATIONS_TO_RUN_OBSOLETIONS_ON_CHILD_CONNECT * st->update_every) <
  168. now)
  169. rrdset_is_obsolete___safe_from_collector_thread(st);
  170. }
  171. rrdset_foreach_done(st);
  172. }
  173. static void svc_rrd_cleanup_obsolete_charts_from_all_hosts() {
  174. worker_is_busy(WORKER_JOB_CLEANUP_OBSOLETE_CHARTS_ON_HOSTS);
  175. rrd_rdlock();
  176. RRDHOST *host;
  177. rrdhost_foreach_read(host) {
  178. if(rrdhost_receiver_replicating_charts(host) || rrdhost_sender_replicating_charts(host))
  179. continue;
  180. svc_rrdhost_cleanup_charts_marked_obsolete(host);
  181. if (host == localhost)
  182. continue;
  183. netdata_mutex_lock(&host->receiver_lock);
  184. time_t now = now_realtime_sec();
  185. if (host->trigger_chart_obsoletion_check &&
  186. ((host->child_last_chart_command &&
  187. host->child_last_chart_command + host->health.health_delay_up_to < now) ||
  188. (host->child_connect_time + TIME_TO_RUN_OBSOLETIONS_ON_CHILD_CONNECT < now))) {
  189. svc_rrdhost_detect_obsolete_charts(host);
  190. host->trigger_chart_obsoletion_check = 0;
  191. }
  192. netdata_mutex_unlock(&host->receiver_lock);
  193. }
  194. rrd_unlock();
  195. }
  196. static void svc_rrdhost_cleanup_orphan_hosts(RRDHOST *protected_host) {
  197. worker_is_busy(WORKER_JOB_CLEANUP_ORPHAN_HOSTS);
  198. rrd_wrlock();
  199. time_t now = now_realtime_sec();
  200. RRDHOST *host;
  201. restart_after_removal:
  202. rrdhost_foreach_write(host) {
  203. if(!rrdhost_should_be_removed(host, protected_host, now))
  204. continue;
  205. bool is_archived = rrdhost_flag_check(host, RRDHOST_FLAG_ARCHIVED);
  206. if (!is_archived) {
  207. netdata_log_info("Host '%s' with machine guid '%s' is obsolete - cleaning up.", rrdhost_hostname(host), host->machine_guid);
  208. if (rrdhost_option_check(host, RRDHOST_OPTION_DELETE_ORPHAN_HOST)
  209. /* don't delete multi-host DB host files */
  210. && !(host->rrd_memory_mode == RRD_MEMORY_MODE_DBENGINE && is_storage_engine_shared(host->db[0].instance))
  211. ) {
  212. worker_is_busy(WORKER_JOB_DELETE_HOST_CHARTS);
  213. rrdhost_delete_charts(host);
  214. }
  215. else {
  216. worker_is_busy(WORKER_JOB_SAVE_HOST_CHARTS);
  217. rrdhost_save_charts(host);
  218. }
  219. }
  220. bool force = false;
  221. if (rrdhost_option_check(host, RRDHOST_OPTION_EPHEMERAL_HOST) && now - host->last_connected > rrdhost_free_ephemeral_time_s)
  222. force = true;
  223. if (!force && is_archived)
  224. continue;
  225. if (force) {
  226. netdata_log_info("Host '%s' with machine guid '%s' is archived, ephemeral clean up.", rrdhost_hostname(host), host->machine_guid);
  227. }
  228. worker_is_busy(WORKER_JOB_FREE_HOST);
  229. #ifdef ENABLE_ACLK
  230. // in case we have cloud connection we inform cloud
  231. // a child disconnected
  232. if (netdata_cloud_enabled && force) {
  233. aclk_host_state_update(host, 0, 0);
  234. unregister_node(host->machine_guid);
  235. }
  236. #endif
  237. rrdhost_free___while_having_rrd_wrlock(host, force);
  238. goto restart_after_removal;
  239. }
  240. rrd_unlock();
  241. }
  242. static void service_main_cleanup(void *ptr)
  243. {
  244. struct netdata_static_thread *static_thread = (struct netdata_static_thread *)ptr;
  245. static_thread->enabled = NETDATA_MAIN_THREAD_EXITING;
  246. netdata_log_debug(D_SYSTEM, "Cleaning up...");
  247. worker_unregister();
  248. static_thread->enabled = NETDATA_MAIN_THREAD_EXITED;
  249. }
  250. /*
  251. * The service thread.
  252. */
  253. void *service_main(void *ptr)
  254. {
  255. worker_register("SERVICE");
  256. worker_register_job_name(WORKER_JOB_CHILD_CHART_OBSOLETION_CHECK, "child chart obsoletion check");
  257. worker_register_job_name(WORKER_JOB_CLEANUP_OBSOLETE_CHARTS, "cleanup obsolete charts");
  258. worker_register_job_name(WORKER_JOB_ARCHIVE_CHART, "archive chart");
  259. worker_register_job_name(WORKER_JOB_ARCHIVE_CHART_DIMENSIONS, "archive chart dimensions");
  260. worker_register_job_name(WORKER_JOB_ARCHIVE_DIMENSION, "archive dimension");
  261. worker_register_job_name(WORKER_JOB_CLEANUP_ORPHAN_HOSTS, "cleanup orphan hosts");
  262. worker_register_job_name(WORKER_JOB_CLEANUP_OBSOLETE_CHARTS_ON_HOSTS, "cleanup obsolete charts on all hosts");
  263. worker_register_job_name(WORKER_JOB_FREE_HOST, "free host");
  264. worker_register_job_name(WORKER_JOB_SAVE_HOST_CHARTS, "save host charts");
  265. worker_register_job_name(WORKER_JOB_DELETE_HOST_CHARTS, "delete host charts");
  266. worker_register_job_name(WORKER_JOB_FREE_CHART, "free chart");
  267. worker_register_job_name(WORKER_JOB_SAVE_CHART, "save chart");
  268. worker_register_job_name(WORKER_JOB_DELETE_CHART, "delete chart");
  269. worker_register_job_name(WORKER_JOB_FREE_DIMENSION, "free dimension");
  270. worker_register_job_name(WORKER_JOB_PGC_MAIN_EVICT, "main cache evictions");
  271. worker_register_job_name(WORKER_JOB_PGC_MAIN_FLUSH, "main cache flushes");
  272. worker_register_job_name(WORKER_JOB_PGC_OPEN_EVICT, "open cache evictions");
  273. worker_register_job_name(WORKER_JOB_PGC_OPEN_FLUSH, "open cache flushes");
  274. netdata_thread_cleanup_push(service_main_cleanup, ptr);
  275. heartbeat_t hb;
  276. heartbeat_init(&hb);
  277. usec_t step = USEC_PER_SEC * SERVICE_HEARTBEAT;
  278. netdata_log_debug(D_SYSTEM, "Service thread starts");
  279. while (service_running(SERVICE_MAINTENANCE)) {
  280. worker_is_idle();
  281. heartbeat_next(&hb, step);
  282. svc_rrd_cleanup_obsolete_charts_from_all_hosts();
  283. svc_rrdhost_cleanup_orphan_hosts(localhost);
  284. }
  285. netdata_thread_cleanup_pop(1);
  286. return NULL;
  287. }