service.c 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327
  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_FREE_CHART 12
  16. #define WORKER_JOB_FREE_DIMENSION 15
  17. #define WORKER_JOB_PGC_MAIN_EVICT 16
  18. #define WORKER_JOB_PGC_MAIN_FLUSH 17
  19. #define WORKER_JOB_PGC_OPEN_EVICT 18
  20. #define WORKER_JOB_PGC_OPEN_FLUSH 19
  21. static void svc_rrddim_obsolete_to_archive(RRDDIM *rd) {
  22. RRDSET *st = rd->rrdset;
  23. if(rrddim_flag_check(rd, RRDDIM_FLAG_ARCHIVED) || !rrddim_flag_check(rd, RRDDIM_FLAG_OBSOLETE))
  24. return;
  25. worker_is_busy(WORKER_JOB_ARCHIVE_DIMENSION);
  26. rrddim_flag_set(rd, RRDDIM_FLAG_ARCHIVED);
  27. rrddim_flag_clear(rd, RRDDIM_FLAG_OBSOLETE);
  28. if (rd->rrd_memory_mode == RRD_MEMORY_MODE_DBENGINE) {
  29. rrddimvar_delete_all(rd);
  30. /* only a collector can mark a chart as obsolete, so we must remove the reference */
  31. size_t tiers_available = 0, tiers_said_no_retention = 0;
  32. for(size_t tier = 0; tier < storage_tiers ;tier++) {
  33. if(rd->tiers[tier].sch) {
  34. tiers_available++;
  35. if(storage_engine_store_finalize(rd->tiers[tier].sch))
  36. tiers_said_no_retention++;
  37. rd->tiers[tier].sch = NULL;
  38. }
  39. }
  40. if (tiers_available == tiers_said_no_retention && tiers_said_no_retention) {
  41. /* This metric has no data and no references */
  42. metaqueue_delete_dimension_uuid(&rd->metric_uuid);
  43. }
  44. else {
  45. /* Do not delete this dimension */
  46. return;
  47. }
  48. }
  49. worker_is_busy(WORKER_JOB_FREE_DIMENSION);
  50. rrddim_free(st, rd);
  51. }
  52. static inline bool svc_rrdset_archive_obsolete_dimensions(RRDSET *st, bool all_dimensions) {
  53. if(!all_dimensions && !rrdset_flag_check(st, RRDSET_FLAG_OBSOLETE_DIMENSIONS))
  54. return true;
  55. worker_is_busy(WORKER_JOB_ARCHIVE_CHART_DIMENSIONS);
  56. rrdset_flag_clear(st, RRDSET_FLAG_OBSOLETE_DIMENSIONS);
  57. RRDDIM *rd;
  58. time_t now = now_realtime_sec();
  59. size_t dim_candidates = 0;
  60. size_t dim_archives = 0;
  61. dfe_start_write(st->rrddim_root_index, rd) {
  62. bool candidate = (all_dimensions || rrddim_flag_check(rd, RRDDIM_FLAG_OBSOLETE));
  63. if(candidate) {
  64. dim_candidates++;
  65. if(rd->collector.last_collected_time.tv_sec + rrdset_free_obsolete_time_s < now) {
  66. size_t references = dictionary_acquired_item_references(rd_dfe.item);
  67. if(references == 1) {
  68. // netdata_log_info("Removing obsolete dimension 'host:%s/chart:%s/dim:%s'",
  69. // rrdhost_hostname(st->rrdhost), rrdset_id(st), rrddim_id(rd));
  70. svc_rrddim_obsolete_to_archive(rd);
  71. dim_archives++;
  72. }
  73. // else
  74. // netdata_log_info("Cannot remove obsolete dimension 'host:%s/chart:%s/dim:%s'",
  75. // rrdhost_hostname(st->rrdhost), rrdset_id(st), rrddim_id(rd));
  76. }
  77. }
  78. }
  79. dfe_done(rd);
  80. if(dim_archives != dim_candidates) {
  81. rrdset_flag_set(st, RRDSET_FLAG_OBSOLETE_DIMENSIONS);
  82. return false;
  83. }
  84. return true;
  85. }
  86. static void svc_rrdset_obsolete_to_free(RRDSET *st) {
  87. if(!svc_rrdset_archive_obsolete_dimensions(st, true))
  88. return;
  89. worker_is_busy(WORKER_JOB_FREE_CHART);
  90. rrdcalc_unlink_all_rrdset_alerts(st);
  91. rrdsetvar_release_and_delete_all(st);
  92. // has to be run after all dimensions are archived - or use-after-free will occur
  93. rrdvar_delete_all(st->rrdvars);
  94. rrdset_free(st);
  95. }
  96. static inline void svc_rrdhost_cleanup_charts_marked_obsolete(RRDHOST *host) {
  97. if(!rrdhost_flag_check(host, RRDHOST_FLAG_PENDING_OBSOLETE_CHARTS|RRDHOST_FLAG_PENDING_OBSOLETE_DIMENSIONS))
  98. return;
  99. worker_is_busy(WORKER_JOB_CLEANUP_OBSOLETE_CHARTS);
  100. rrdhost_flag_clear(host, RRDHOST_FLAG_PENDING_OBSOLETE_CHARTS|RRDHOST_FLAG_PENDING_OBSOLETE_DIMENSIONS);
  101. size_t full_candidates = 0;
  102. size_t full_archives = 0;
  103. size_t partial_candidates = 0;
  104. size_t partial_archives = 0;
  105. time_t now = now_realtime_sec();
  106. RRDSET *st;
  107. rrdset_foreach_reentrant(st, host) {
  108. if(rrdset_is_replicating(st))
  109. continue;
  110. RRDSET_FLAGS flags = rrdset_flag_get(st);
  111. bool obsolete_chart = flags & RRDSET_FLAG_OBSOLETE;
  112. bool obsolete_dims = flags & RRDSET_FLAG_OBSOLETE_DIMENSIONS;
  113. if(obsolete_dims) {
  114. partial_candidates++;
  115. if(svc_rrdset_archive_obsolete_dimensions(st, false))
  116. partial_archives++;
  117. }
  118. if(obsolete_chart) {
  119. full_candidates++;
  120. if(unlikely( st->last_accessed_time_s + rrdset_free_obsolete_time_s < now
  121. && st->last_updated.tv_sec + rrdset_free_obsolete_time_s < now
  122. && st->last_collected_time.tv_sec + rrdset_free_obsolete_time_s < now
  123. )) {
  124. svc_rrdset_obsolete_to_free(st);
  125. full_archives++;
  126. }
  127. }
  128. }
  129. rrdset_foreach_done(st);
  130. if(partial_archives != partial_candidates)
  131. rrdhost_flag_set(host, RRDHOST_FLAG_PENDING_OBSOLETE_DIMENSIONS);
  132. if(full_archives != full_candidates)
  133. rrdhost_flag_set(host, RRDHOST_FLAG_PENDING_OBSOLETE_CHARTS);
  134. }
  135. static void svc_rrdhost_detect_obsolete_charts(RRDHOST *host) {
  136. worker_is_busy(WORKER_JOB_CHILD_CHART_OBSOLETION_CHECK);
  137. time_t now = now_realtime_sec();
  138. time_t last_entry_t;
  139. RRDSET *st;
  140. time_t child_connect_time = host->child_connect_time;
  141. rrdset_foreach_read(st, host) {
  142. if(rrdset_is_replicating(st))
  143. continue;
  144. last_entry_t = rrdset_last_entry_s(st);
  145. if (last_entry_t && last_entry_t < child_connect_time &&
  146. child_connect_time + TIME_TO_RUN_OBSOLETIONS_ON_CHILD_CONNECT +
  147. (ITERATIONS_TO_RUN_OBSOLETIONS_ON_CHILD_CONNECT * st->update_every) <
  148. now)
  149. rrdset_is_obsolete___safe_from_collector_thread(st);
  150. }
  151. rrdset_foreach_done(st);
  152. }
  153. static void svc_rrd_cleanup_obsolete_charts_from_all_hosts() {
  154. worker_is_busy(WORKER_JOB_CLEANUP_OBSOLETE_CHARTS_ON_HOSTS);
  155. rrd_rdlock();
  156. RRDHOST *host;
  157. rrdhost_foreach_read(host) {
  158. if(rrdhost_receiver_replicating_charts(host) || rrdhost_sender_replicating_charts(host))
  159. continue;
  160. svc_rrdhost_cleanup_charts_marked_obsolete(host);
  161. if (host == localhost)
  162. continue;
  163. netdata_mutex_lock(&host->receiver_lock);
  164. time_t now = now_realtime_sec();
  165. if (host->trigger_chart_obsoletion_check &&
  166. ((host->child_last_chart_command &&
  167. host->child_last_chart_command + host->health.health_delay_up_to < now) ||
  168. (host->child_connect_time + TIME_TO_RUN_OBSOLETIONS_ON_CHILD_CONNECT < now))) {
  169. svc_rrdhost_detect_obsolete_charts(host);
  170. host->trigger_chart_obsoletion_check = 0;
  171. }
  172. netdata_mutex_unlock(&host->receiver_lock);
  173. }
  174. rrd_unlock();
  175. }
  176. static void svc_rrdhost_cleanup_orphan_hosts(RRDHOST *protected_host) {
  177. worker_is_busy(WORKER_JOB_CLEANUP_ORPHAN_HOSTS);
  178. rrd_wrlock();
  179. time_t now = now_realtime_sec();
  180. RRDHOST *host;
  181. restart_after_removal:
  182. rrdhost_foreach_write(host) {
  183. if(!rrdhost_should_be_removed(host, protected_host, now))
  184. continue;
  185. bool force = false;
  186. if (rrdhost_option_check(host, RRDHOST_OPTION_EPHEMERAL_HOST) && now - host->last_connected > rrdhost_free_ephemeral_time_s)
  187. force = true;
  188. bool is_archived = rrdhost_flag_check(host, RRDHOST_FLAG_ARCHIVED);
  189. if (!force && is_archived)
  190. continue;
  191. if (force) {
  192. netdata_log_info("Host '%s' with machine guid '%s' is archived, ephemeral clean up.", rrdhost_hostname(host), host->machine_guid);
  193. }
  194. worker_is_busy(WORKER_JOB_FREE_HOST);
  195. #ifdef ENABLE_ACLK
  196. // in case we have cloud connection we inform cloud
  197. // a child disconnected
  198. if (netdata_cloud_enabled && force) {
  199. aclk_host_state_update(host, 0, 0);
  200. unregister_node(host->machine_guid);
  201. }
  202. #endif
  203. rrdhost_free___while_having_rrd_wrlock(host, force);
  204. goto restart_after_removal;
  205. }
  206. rrd_unlock();
  207. }
  208. static void service_main_cleanup(void *ptr)
  209. {
  210. struct netdata_static_thread *static_thread = (struct netdata_static_thread *)ptr;
  211. static_thread->enabled = NETDATA_MAIN_THREAD_EXITING;
  212. netdata_log_debug(D_SYSTEM, "Cleaning up...");
  213. worker_unregister();
  214. static_thread->enabled = NETDATA_MAIN_THREAD_EXITED;
  215. }
  216. /*
  217. * The service thread.
  218. */
  219. void *service_main(void *ptr)
  220. {
  221. worker_register("SERVICE");
  222. worker_register_job_name(WORKER_JOB_CHILD_CHART_OBSOLETION_CHECK, "child chart obsoletion check");
  223. worker_register_job_name(WORKER_JOB_CLEANUP_OBSOLETE_CHARTS, "cleanup obsolete charts");
  224. worker_register_job_name(WORKER_JOB_ARCHIVE_CHART, "archive chart");
  225. worker_register_job_name(WORKER_JOB_ARCHIVE_CHART_DIMENSIONS, "archive chart dimensions");
  226. worker_register_job_name(WORKER_JOB_ARCHIVE_DIMENSION, "archive dimension");
  227. worker_register_job_name(WORKER_JOB_CLEANUP_ORPHAN_HOSTS, "cleanup orphan hosts");
  228. worker_register_job_name(WORKER_JOB_CLEANUP_OBSOLETE_CHARTS_ON_HOSTS, "cleanup obsolete charts on all hosts");
  229. worker_register_job_name(WORKER_JOB_FREE_HOST, "free host");
  230. worker_register_job_name(WORKER_JOB_FREE_CHART, "free chart");
  231. worker_register_job_name(WORKER_JOB_FREE_DIMENSION, "free dimension");
  232. worker_register_job_name(WORKER_JOB_PGC_MAIN_EVICT, "main cache evictions");
  233. worker_register_job_name(WORKER_JOB_PGC_MAIN_FLUSH, "main cache flushes");
  234. worker_register_job_name(WORKER_JOB_PGC_OPEN_EVICT, "open cache evictions");
  235. worker_register_job_name(WORKER_JOB_PGC_OPEN_FLUSH, "open cache flushes");
  236. netdata_thread_cleanup_push(service_main_cleanup, ptr);
  237. heartbeat_t hb;
  238. heartbeat_init(&hb);
  239. usec_t step = USEC_PER_SEC * SERVICE_HEARTBEAT;
  240. netdata_log_debug(D_SYSTEM, "Service thread starts");
  241. while (service_running(SERVICE_MAINTENANCE)) {
  242. worker_is_idle();
  243. heartbeat_next(&hb, step);
  244. svc_rrd_cleanup_obsolete_charts_from_all_hosts();
  245. svc_rrdhost_cleanup_orphan_hosts(localhost);
  246. }
  247. netdata_thread_cleanup_pop(1);
  248. return NULL;
  249. }