service.c 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304
  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 bool svc_rrdset_archive_obsolete_dimensions(RRDSET *st, bool all_dimensions) {
  63. worker_is_busy(WORKER_JOB_ARCHIVE_CHART_DIMENSIONS);
  64. RRDDIM *rd;
  65. time_t now = now_realtime_sec();
  66. bool done_all_dimensions = true;
  67. dfe_start_write(st->rrddim_root_index, rd) {
  68. if(unlikely(
  69. all_dimensions ||
  70. (rrddim_flag_check(rd, RRDDIM_FLAG_OBSOLETE) && (rd->collector.last_collected_time.tv_sec + rrdset_free_obsolete_time_s < now))
  71. )) {
  72. if(dictionary_acquired_item_references(rd_dfe.item) == 1) {
  73. netdata_log_info("Removing obsolete dimension '%s' (%s) of '%s' (%s).", rrddim_name(rd), rrddim_id(rd), rrdset_name(st), rrdset_id(st));
  74. svc_rrddim_obsolete_to_archive(rd);
  75. }
  76. else
  77. done_all_dimensions = false;
  78. }
  79. else
  80. done_all_dimensions = false;
  81. }
  82. dfe_done(rd);
  83. return done_all_dimensions;
  84. }
  85. static void svc_rrdset_obsolete_to_free(RRDSET *st) {
  86. if(!svc_rrdset_archive_obsolete_dimensions(st, true))
  87. return;
  88. worker_is_busy(WORKER_JOB_FREE_CHART);
  89. rrdcalc_unlink_all_rrdset_alerts(st);
  90. rrdsetvar_release_and_delete_all(st);
  91. // has to be run after all dimensions are archived - or use-after-free will occur
  92. rrdvar_delete_all(st->rrdvars);
  93. if(st->rrd_memory_mode != RRD_MEMORY_MODE_DBENGINE) {
  94. if(rrdhost_option_check(st->rrdhost, RRDHOST_OPTION_DELETE_OBSOLETE_CHARTS)) {
  95. worker_is_busy(WORKER_JOB_DELETE_CHART);
  96. rrdset_delete_files(st);
  97. }
  98. else {
  99. worker_is_busy(WORKER_JOB_SAVE_CHART);
  100. rrdset_save(st);
  101. }
  102. }
  103. rrdset_free(st);
  104. }
  105. static void svc_rrdhost_cleanup_obsolete_charts(RRDHOST *host) {
  106. worker_is_busy(WORKER_JOB_CLEANUP_OBSOLETE_CHARTS);
  107. time_t now = now_realtime_sec();
  108. RRDSET *st;
  109. rrdset_foreach_reentrant(st, host) {
  110. if(rrdset_is_replicating(st))
  111. continue;
  112. if(unlikely(rrdset_flag_check(st, RRDSET_FLAG_OBSOLETE)
  113. && st->last_accessed_time_s + rrdset_free_obsolete_time_s < now
  114. && st->last_updated.tv_sec + rrdset_free_obsolete_time_s < now
  115. && st->last_collected_time.tv_sec + rrdset_free_obsolete_time_s < now
  116. )) {
  117. svc_rrdset_obsolete_to_free(st);
  118. }
  119. else if(rrdset_flag_check(st, RRDSET_FLAG_OBSOLETE_DIMENSIONS)) {
  120. rrdset_flag_clear(st, RRDSET_FLAG_OBSOLETE_DIMENSIONS);
  121. svc_rrdset_archive_obsolete_dimensions(st, false);
  122. }
  123. else if (unlikely(rrdset_flag_check(st, RRDSET_FLAG_OBSOLETE))) {
  124. rrdhost_flag_set(host, RRDHOST_FLAG_PENDING_OBSOLETE_CHARTS);
  125. }
  126. }
  127. rrdset_foreach_done(st);
  128. }
  129. static void svc_rrdset_check_obsoletion(RRDHOST *host) {
  130. worker_is_busy(WORKER_JOB_CHILD_CHART_OBSOLETION_CHECK);
  131. time_t now = now_realtime_sec();
  132. time_t last_entry_t;
  133. RRDSET *st;
  134. rrdset_foreach_read(st, host) {
  135. if(rrdset_is_replicating(st))
  136. continue;
  137. last_entry_t = rrdset_last_entry_s(st);
  138. if(last_entry_t && last_entry_t < host->child_connect_time &&
  139. host->child_connect_time + TIME_TO_RUN_OBSOLETIONS_ON_CHILD_CONNECT + ITERATIONS_TO_RUN_OBSOLETIONS_ON_CHILD_CONNECT * st->update_every
  140. < now)
  141. rrdset_is_obsolete(st);
  142. }
  143. rrdset_foreach_done(st);
  144. }
  145. static void svc_rrd_cleanup_obsolete_charts_from_all_hosts() {
  146. worker_is_busy(WORKER_JOB_CLEANUP_OBSOLETE_CHARTS_ON_HOSTS);
  147. rrd_rdlock();
  148. RRDHOST *host;
  149. rrdhost_foreach_read(host) {
  150. if(rrdhost_receiver_replicating_charts(host) || rrdhost_sender_replicating_charts(host))
  151. continue;
  152. if(rrdhost_flag_check(host, RRDHOST_FLAG_PENDING_OBSOLETE_CHARTS|RRDHOST_FLAG_PENDING_OBSOLETE_DIMENSIONS)) {
  153. rrdhost_flag_clear(host, RRDHOST_FLAG_PENDING_OBSOLETE_CHARTS|RRDHOST_FLAG_PENDING_OBSOLETE_DIMENSIONS);
  154. svc_rrdhost_cleanup_obsolete_charts(host);
  155. }
  156. if(host != localhost
  157. && host->trigger_chart_obsoletion_check
  158. && (
  159. (
  160. host->child_last_chart_command
  161. && host->child_last_chart_command + host->health.health_delay_up_to < now_realtime_sec()
  162. )
  163. || (host->child_connect_time + TIME_TO_RUN_OBSOLETIONS_ON_CHILD_CONNECT < now_realtime_sec())
  164. )
  165. ) {
  166. svc_rrdset_check_obsoletion(host);
  167. host->trigger_chart_obsoletion_check = 0;
  168. }
  169. }
  170. rrd_unlock();
  171. }
  172. static void svc_rrdhost_cleanup_orphan_hosts(RRDHOST *protected_host) {
  173. worker_is_busy(WORKER_JOB_CLEANUP_ORPHAN_HOSTS);
  174. rrd_wrlock();
  175. time_t now = now_realtime_sec();
  176. RRDHOST *host;
  177. restart_after_removal:
  178. rrdhost_foreach_write(host) {
  179. if(!rrdhost_should_be_removed(host, protected_host, now))
  180. continue;
  181. netdata_log_info("Host '%s' with machine guid '%s' is obsolete - cleaning up.", rrdhost_hostname(host), host->machine_guid);
  182. if (rrdhost_option_check(host, RRDHOST_OPTION_DELETE_ORPHAN_HOST)
  183. /* don't delete multi-host DB host files */
  184. && !(host->rrd_memory_mode == RRD_MEMORY_MODE_DBENGINE && is_storage_engine_shared(host->db[0].instance))
  185. ) {
  186. worker_is_busy(WORKER_JOB_DELETE_HOST_CHARTS);
  187. rrdhost_delete_charts(host);
  188. }
  189. else {
  190. worker_is_busy(WORKER_JOB_SAVE_HOST_CHARTS);
  191. rrdhost_save_charts(host);
  192. }
  193. worker_is_busy(WORKER_JOB_FREE_HOST);
  194. rrdhost_free___while_having_rrd_wrlock(host, false);
  195. goto restart_after_removal;
  196. }
  197. rrd_unlock();
  198. }
  199. static void service_main_cleanup(void *ptr)
  200. {
  201. struct netdata_static_thread *static_thread = (struct netdata_static_thread *)ptr;
  202. static_thread->enabled = NETDATA_MAIN_THREAD_EXITING;
  203. netdata_log_debug(D_SYSTEM, "Cleaning up...");
  204. worker_unregister();
  205. static_thread->enabled = NETDATA_MAIN_THREAD_EXITED;
  206. }
  207. /*
  208. * The service thread.
  209. */
  210. void *service_main(void *ptr)
  211. {
  212. worker_register("SERVICE");
  213. worker_register_job_name(WORKER_JOB_CHILD_CHART_OBSOLETION_CHECK, "child chart obsoletion check");
  214. worker_register_job_name(WORKER_JOB_CLEANUP_OBSOLETE_CHARTS, "cleanup obsolete charts");
  215. worker_register_job_name(WORKER_JOB_ARCHIVE_CHART, "archive chart");
  216. worker_register_job_name(WORKER_JOB_ARCHIVE_CHART_DIMENSIONS, "archive chart dimensions");
  217. worker_register_job_name(WORKER_JOB_ARCHIVE_DIMENSION, "archive dimension");
  218. worker_register_job_name(WORKER_JOB_CLEANUP_ORPHAN_HOSTS, "cleanup orphan hosts");
  219. worker_register_job_name(WORKER_JOB_CLEANUP_OBSOLETE_CHARTS_ON_HOSTS, "cleanup obsolete charts on all hosts");
  220. worker_register_job_name(WORKER_JOB_FREE_HOST, "free host");
  221. worker_register_job_name(WORKER_JOB_SAVE_HOST_CHARTS, "save host charts");
  222. worker_register_job_name(WORKER_JOB_DELETE_HOST_CHARTS, "delete host charts");
  223. worker_register_job_name(WORKER_JOB_FREE_CHART, "free chart");
  224. worker_register_job_name(WORKER_JOB_SAVE_CHART, "save chart");
  225. worker_register_job_name(WORKER_JOB_DELETE_CHART, "delete chart");
  226. worker_register_job_name(WORKER_JOB_FREE_DIMENSION, "free dimension");
  227. worker_register_job_name(WORKER_JOB_PGC_MAIN_EVICT, "main cache evictions");
  228. worker_register_job_name(WORKER_JOB_PGC_MAIN_FLUSH, "main cache flushes");
  229. worker_register_job_name(WORKER_JOB_PGC_OPEN_EVICT, "open cache evictions");
  230. worker_register_job_name(WORKER_JOB_PGC_OPEN_FLUSH, "open cache flushes");
  231. netdata_thread_cleanup_push(service_main_cleanup, ptr);
  232. heartbeat_t hb;
  233. heartbeat_init(&hb);
  234. usec_t step = USEC_PER_SEC * SERVICE_HEARTBEAT;
  235. netdata_log_debug(D_SYSTEM, "Service thread starts");
  236. while (service_running(SERVICE_MAINTENANCE)) {
  237. worker_is_idle();
  238. heartbeat_next(&hb, step);
  239. svc_rrd_cleanup_obsolete_charts_from_all_hosts();
  240. svc_rrdhost_cleanup_orphan_hosts(localhost);
  241. }
  242. netdata_thread_cleanup_pop(1);
  243. return NULL;
  244. }