rrdhost.c 32 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537538539540541542543544545546547548549550551552553554555556557558559560561562563564565566567568569570571572573574575576577578579580581582583584585586587588589590591592593594595596597598599600601602603604605606607608609610611612613614615616617618619620621622623624625626627628629630631632633634635636637638639640641642643644645646647648649650651652653654655656657658659660661662663664665666667668669670671672673674675676677678679680681682683684685686687688689690691692693694695696697698699700701702703704705706707708709710711712713714715716717718719720721722723724725726727728729730731732733734735736737738739740741742743744745746747748749750751752753754755756757758759760761762763764765766767768769770771772773774775776777778779780781782783784785786787788789790791792793794795796797798799800801802803804805806807808809810811812813814815816817818819820821822823824825826827828829830831832833834835836837838839840841842843844845846847848849850851852853854855856857858859860861862863864865866867868869870871872873874875876877878879880881882883884885886887888889890891892893894895896897898899900901902903904905906907908909910911912913914915916917918919920921922923924925926927928929930931932933934935936937938
  1. // SPDX-License-Identifier: GPL-3.0-or-later
  2. #define NETDATA_RRD_INTERNALS
  3. #include "rrd.h"
  4. RRDHOST *localhost = NULL;
  5. size_t rrd_hosts_available = 0;
  6. netdata_rwlock_t rrd_rwlock = NETDATA_RWLOCK_INITIALIZER;
  7. time_t rrdset_free_obsolete_time = 3600;
  8. time_t rrdhost_free_orphan_time = 3600;
  9. // ----------------------------------------------------------------------------
  10. // RRDHOST index
  11. int rrdhost_compare(void* a, void* b) {
  12. if(((RRDHOST *)a)->hash_machine_guid < ((RRDHOST *)b)->hash_machine_guid) return -1;
  13. else if(((RRDHOST *)a)->hash_machine_guid > ((RRDHOST *)b)->hash_machine_guid) return 1;
  14. else return strcmp(((RRDHOST *)a)->machine_guid, ((RRDHOST *)b)->machine_guid);
  15. }
  16. avl_tree_lock rrdhost_root_index = {
  17. .avl_tree = { NULL, rrdhost_compare },
  18. .rwlock = AVL_LOCK_INITIALIZER
  19. };
  20. RRDHOST *rrdhost_find_by_guid(const char *guid, uint32_t hash) {
  21. debug(D_RRDHOST, "Searching in index for host with guid '%s'", guid);
  22. RRDHOST tmp;
  23. strncpyz(tmp.machine_guid, guid, GUID_LEN);
  24. tmp.hash_machine_guid = (hash)?hash:simple_hash(tmp.machine_guid);
  25. return (RRDHOST *)avl_search_lock(&(rrdhost_root_index), (avl *) &tmp);
  26. }
  27. RRDHOST *rrdhost_find_by_hostname(const char *hostname, uint32_t hash) {
  28. if(unlikely(!strcmp(hostname, "localhost")))
  29. return localhost;
  30. if(unlikely(!hash)) hash = simple_hash(hostname);
  31. rrd_rdlock();
  32. RRDHOST *host;
  33. rrdhost_foreach_read(host) {
  34. if(unlikely((hash == host->hash_hostname && !strcmp(hostname, host->hostname)))) {
  35. rrd_unlock();
  36. return host;
  37. }
  38. }
  39. rrd_unlock();
  40. return NULL;
  41. }
  42. #define rrdhost_index_add(rrdhost) (RRDHOST *)avl_insert_lock(&(rrdhost_root_index), (avl *)(rrdhost))
  43. #define rrdhost_index_del(rrdhost) (RRDHOST *)avl_remove_lock(&(rrdhost_root_index), (avl *)(rrdhost))
  44. // ----------------------------------------------------------------------------
  45. // RRDHOST - internal helpers
  46. static inline void rrdhost_init_tags(RRDHOST *host, const char *tags) {
  47. if(host->tags && tags && !strcmp(host->tags, tags))
  48. return;
  49. void *old = (void *)host->tags;
  50. host->tags = (tags && *tags)?strdupz(tags):NULL;
  51. freez(old);
  52. }
  53. static inline void rrdhost_init_hostname(RRDHOST *host, const char *hostname) {
  54. if(host->hostname && hostname && !strcmp(host->hostname, hostname))
  55. return;
  56. void *old = host->hostname;
  57. host->hostname = strdupz(hostname?hostname:"localhost");
  58. host->hash_hostname = simple_hash(host->hostname);
  59. freez(old);
  60. }
  61. static inline void rrdhost_init_os(RRDHOST *host, const char *os) {
  62. if(host->os && os && !strcmp(host->os, os))
  63. return;
  64. void *old = (void *)host->os;
  65. host->os = strdupz(os?os:"unknown");
  66. freez(old);
  67. }
  68. static inline void rrdhost_init_timezone(RRDHOST *host, const char *timezone) {
  69. if(host->timezone && timezone && !strcmp(host->timezone, timezone))
  70. return;
  71. void *old = (void *)host->timezone;
  72. host->timezone = strdupz((timezone && *timezone)?timezone:"unknown");
  73. freez(old);
  74. }
  75. static inline void rrdhost_init_machine_guid(RRDHOST *host, const char *machine_guid) {
  76. strncpy(host->machine_guid, machine_guid, GUID_LEN);
  77. host->machine_guid[GUID_LEN] = '\0';
  78. host->hash_machine_guid = simple_hash(host->machine_guid);
  79. }
  80. // ----------------------------------------------------------------------------
  81. // RRDHOST - add a host
  82. RRDHOST *rrdhost_create(const char *hostname,
  83. const char *registry_hostname,
  84. const char *guid,
  85. const char *os,
  86. const char *timezone,
  87. const char *tags,
  88. const char *program_name,
  89. const char *program_version,
  90. int update_every,
  91. long entries,
  92. RRD_MEMORY_MODE memory_mode,
  93. unsigned int health_enabled,
  94. unsigned int rrdpush_enabled,
  95. char *rrdpush_destination,
  96. char *rrdpush_api_key,
  97. char *rrdpush_send_charts_matching,
  98. struct rrdhost_system_info *system_info,
  99. int is_localhost
  100. ) {
  101. debug(D_RRDHOST, "Host '%s': adding with guid '%s'", hostname, guid);
  102. rrd_check_wrlock();
  103. RRDHOST *host = callocz(1, sizeof(RRDHOST));
  104. host->rrd_update_every = (update_every > 0)?update_every:1;
  105. host->rrd_history_entries = align_entries_to_pagesize(memory_mode, entries);
  106. host->rrd_memory_mode = memory_mode;
  107. #ifdef ENABLE_DBENGINE
  108. host->page_cache_mb = default_rrdeng_page_cache_mb;
  109. host->disk_space_mb = default_rrdeng_disk_quota_mb;
  110. #endif
  111. host->health_enabled = (memory_mode == RRD_MEMORY_MODE_NONE)? 0 : health_enabled;
  112. host->rrdpush_send_enabled = (rrdpush_enabled && rrdpush_destination && *rrdpush_destination && rrdpush_api_key && *rrdpush_api_key) ? 1 : 0;
  113. host->rrdpush_send_destination = (host->rrdpush_send_enabled)?strdupz(rrdpush_destination):NULL;
  114. host->rrdpush_send_api_key = (host->rrdpush_send_enabled)?strdupz(rrdpush_api_key):NULL;
  115. host->rrdpush_send_charts_matching = simple_pattern_create(rrdpush_send_charts_matching, NULL, SIMPLE_PATTERN_EXACT);
  116. host->rrdpush_sender_pipe[0] = -1;
  117. host->rrdpush_sender_pipe[1] = -1;
  118. host->rrdpush_sender_socket = -1;
  119. #ifdef ENABLE_HTTPS
  120. host->ssl.conn = NULL;
  121. host->ssl.flags = NETDATA_SSL_START;
  122. host->stream_ssl.conn = NULL;
  123. host->stream_ssl.flags = NETDATA_SSL_START;
  124. #endif
  125. netdata_mutex_init(&host->rrdpush_sender_buffer_mutex);
  126. netdata_rwlock_init(&host->rrdhost_rwlock);
  127. rrdhost_init_hostname(host, hostname);
  128. rrdhost_init_machine_guid(host, guid);
  129. rrdhost_init_os(host, os);
  130. rrdhost_init_timezone(host, timezone);
  131. rrdhost_init_tags(host, tags);
  132. host->program_name = strdupz((program_name && *program_name)?program_name:"unknown");
  133. host->program_version = strdupz((program_version && *program_version)?program_version:"unknown");
  134. host->registry_hostname = strdupz((registry_hostname && *registry_hostname)?registry_hostname:hostname);
  135. host->system_info = system_info;
  136. avl_init_lock(&(host->rrdset_root_index), rrdset_compare);
  137. avl_init_lock(&(host->rrdset_root_index_name), rrdset_compare_name);
  138. avl_init_lock(&(host->rrdfamily_root_index), rrdfamily_compare);
  139. avl_init_lock(&(host->rrdvar_root_index), rrdvar_compare);
  140. if(config_get_boolean(CONFIG_SECTION_GLOBAL, "delete obsolete charts files", 1))
  141. rrdhost_flag_set(host, RRDHOST_FLAG_DELETE_OBSOLETE_CHARTS);
  142. if(config_get_boolean(CONFIG_SECTION_GLOBAL, "delete orphan hosts files", 1) && !is_localhost)
  143. rrdhost_flag_set(host, RRDHOST_FLAG_DELETE_ORPHAN_HOST);
  144. host->health_default_warn_repeat_every = config_get_duration(CONFIG_SECTION_HEALTH, "default repeat warning", "never");
  145. host->health_default_crit_repeat_every = config_get_duration(CONFIG_SECTION_HEALTH, "default repeat critical", "never");
  146. avl_init_lock(&(host->alarms_idx_health_log), alarm_compare_id);
  147. avl_init_lock(&(host->alarms_idx_name), alarm_compare_name);
  148. // ------------------------------------------------------------------------
  149. // initialize health variables
  150. host->health_log.next_log_id = 1;
  151. host->health_log.next_alarm_id = 1;
  152. host->health_log.max = 1000;
  153. host->health_log.next_log_id =
  154. host->health_log.next_alarm_id = (uint32_t)now_realtime_sec();
  155. long n = config_get_number(CONFIG_SECTION_HEALTH, "in memory max health log entries", host->health_log.max);
  156. if(n < 10) {
  157. error("Host '%s': health configuration has invalid max log entries %ld. Using default %u", host->hostname, n, host->health_log.max);
  158. config_set_number(CONFIG_SECTION_HEALTH, "in memory max health log entries", (long)host->health_log.max);
  159. }
  160. else
  161. host->health_log.max = (unsigned int)n;
  162. netdata_rwlock_init(&host->health_log.alarm_log_rwlock);
  163. char filename[FILENAME_MAX + 1];
  164. if(is_localhost) {
  165. host->cache_dir = strdupz(netdata_configured_cache_dir);
  166. host->varlib_dir = strdupz(netdata_configured_varlib_dir);
  167. }
  168. else {
  169. // this is not localhost - append our GUID to localhost path
  170. snprintfz(filename, FILENAME_MAX, "%s/%s", netdata_configured_cache_dir, host->machine_guid);
  171. host->cache_dir = strdupz(filename);
  172. if(host->rrd_memory_mode == RRD_MEMORY_MODE_MAP || host->rrd_memory_mode == RRD_MEMORY_MODE_SAVE ||
  173. host->rrd_memory_mode == RRD_MEMORY_MODE_DBENGINE) {
  174. int r = mkdir(host->cache_dir, 0775);
  175. if(r != 0 && errno != EEXIST)
  176. error("Host '%s': cannot create directory '%s'", host->hostname, host->cache_dir);
  177. }
  178. snprintfz(filename, FILENAME_MAX, "%s/%s", netdata_configured_varlib_dir, host->machine_guid);
  179. host->varlib_dir = strdupz(filename);
  180. if(host->health_enabled) {
  181. int r = mkdir(host->varlib_dir, 0775);
  182. if(r != 0 && errno != EEXIST)
  183. error("Host '%s': cannot create directory '%s'", host->hostname, host->varlib_dir);
  184. }
  185. }
  186. if (host->rrd_memory_mode == RRD_MEMORY_MODE_DBENGINE) {
  187. #ifdef ENABLE_DBENGINE
  188. char dbenginepath[FILENAME_MAX + 1];
  189. int ret;
  190. snprintfz(dbenginepath, FILENAME_MAX, "%s/dbengine", host->cache_dir);
  191. ret = mkdir(dbenginepath, 0775);
  192. if(ret != 0 && errno != EEXIST)
  193. error("Host '%s': cannot create directory '%s'", host->hostname, dbenginepath);
  194. else
  195. ret = rrdeng_init(&host->rrdeng_ctx, dbenginepath, host->page_cache_mb, host->disk_space_mb);
  196. if(ret) {
  197. error("Host '%s': cannot initialize host with machine guid '%s'. Failed to initialize DB engine at '%s'.",
  198. host->hostname, host->machine_guid, host->cache_dir);
  199. rrdhost_free(host);
  200. host = NULL;
  201. //rrd_hosts_available++; //TODO: maybe we want this?
  202. return host;
  203. }
  204. #else
  205. fatal("RRD_MEMORY_MODE_DBENGINE is not supported in this platform.");
  206. #endif
  207. }
  208. if(host->health_enabled) {
  209. snprintfz(filename, FILENAME_MAX, "%s/health", host->varlib_dir);
  210. int r = mkdir(filename, 0775);
  211. if(r != 0 && errno != EEXIST)
  212. error("Host '%s': cannot create directory '%s'", host->hostname, filename);
  213. }
  214. snprintfz(filename, FILENAME_MAX, "%s/health/health-log.db", host->varlib_dir);
  215. host->health_log_filename = strdupz(filename);
  216. snprintfz(filename, FILENAME_MAX, "%s/alarm-notify.sh", netdata_configured_primary_plugins_dir);
  217. host->health_default_exec = strdupz(config_get(CONFIG_SECTION_HEALTH, "script to execute on alarm", filename));
  218. host->health_default_recipient = strdupz("root");
  219. // ------------------------------------------------------------------------
  220. // load health configuration
  221. if(host->health_enabled) {
  222. rrdhost_wrlock(host);
  223. health_readdir(host, health_user_config_dir(), health_stock_config_dir(), NULL);
  224. rrdhost_unlock(host);
  225. health_alarm_log_load(host);
  226. health_alarm_log_open(host);
  227. }
  228. // ------------------------------------------------------------------------
  229. // link it and add it to the index
  230. if(is_localhost) {
  231. host->next = localhost;
  232. localhost = host;
  233. }
  234. else {
  235. if(localhost) {
  236. host->next = localhost->next;
  237. localhost->next = host;
  238. }
  239. else localhost = host;
  240. }
  241. RRDHOST *t = rrdhost_index_add(host);
  242. if(t != host) {
  243. error("Host '%s': cannot add host with machine guid '%s' to index. It already exists as host '%s' with machine guid '%s'.", host->hostname, host->machine_guid, t->hostname, t->machine_guid);
  244. rrdhost_free(host);
  245. host = NULL;
  246. }
  247. else {
  248. info("Host '%s' (at registry as '%s') with guid '%s' initialized"
  249. ", os '%s'"
  250. ", timezone '%s'"
  251. ", tags '%s'"
  252. ", program_name '%s'"
  253. ", program_version '%s'"
  254. ", update every %d"
  255. ", memory mode %s"
  256. ", history entries %ld"
  257. ", streaming %s"
  258. " (to '%s' with api key '%s')"
  259. ", health %s"
  260. ", cache_dir '%s'"
  261. ", varlib_dir '%s'"
  262. ", health_log '%s'"
  263. ", alarms default handler '%s'"
  264. ", alarms default recipient '%s'"
  265. , host->hostname
  266. , host->registry_hostname
  267. , host->machine_guid
  268. , host->os
  269. , host->timezone
  270. , (host->tags)?host->tags:""
  271. , host->program_name
  272. , host->program_version
  273. , host->rrd_update_every
  274. , rrd_memory_mode_name(host->rrd_memory_mode)
  275. , host->rrd_history_entries
  276. , host->rrdpush_send_enabled?"enabled":"disabled"
  277. , host->rrdpush_send_destination?host->rrdpush_send_destination:""
  278. , host->rrdpush_send_api_key?host->rrdpush_send_api_key:""
  279. , host->health_enabled?"enabled":"disabled"
  280. , host->cache_dir
  281. , host->varlib_dir
  282. , host->health_log_filename
  283. , host->health_default_exec
  284. , host->health_default_recipient
  285. );
  286. }
  287. rrd_hosts_available++;
  288. return host;
  289. }
  290. RRDHOST *rrdhost_find_or_create(
  291. const char *hostname
  292. , const char *registry_hostname
  293. , const char *guid
  294. , const char *os
  295. , const char *timezone
  296. , const char *tags
  297. , const char *program_name
  298. , const char *program_version
  299. , int update_every
  300. , long history
  301. , RRD_MEMORY_MODE mode
  302. , unsigned int health_enabled
  303. , unsigned int rrdpush_enabled
  304. , char *rrdpush_destination
  305. , char *rrdpush_api_key
  306. , char *rrdpush_send_charts_matching
  307. , struct rrdhost_system_info *system_info
  308. ) {
  309. debug(D_RRDHOST, "Searching for host '%s' with guid '%s'", hostname, guid);
  310. rrd_wrlock();
  311. RRDHOST *host = rrdhost_find_by_guid(guid, 0);
  312. if(!host) {
  313. host = rrdhost_create(
  314. hostname
  315. , registry_hostname
  316. , guid
  317. , os
  318. , timezone
  319. , tags
  320. , program_name
  321. , program_version
  322. , update_every
  323. , history
  324. , mode
  325. , health_enabled
  326. , rrdpush_enabled
  327. , rrdpush_destination
  328. , rrdpush_api_key
  329. , rrdpush_send_charts_matching
  330. , system_info
  331. , 0
  332. );
  333. }
  334. else {
  335. host->health_enabled = health_enabled;
  336. if(strcmp(host->hostname, hostname) != 0) {
  337. info("Host '%s' has been renamed to '%s'. If this is not intentional it may mean multiple hosts are using the same machine_guid.", host->hostname, hostname);
  338. char *t = host->hostname;
  339. host->hostname = strdupz(hostname);
  340. host->hash_hostname = simple_hash(host->hostname);
  341. freez(t);
  342. }
  343. if(strcmp(host->program_name, program_name) != 0) {
  344. info("Host '%s' switched program name from '%s' to '%s'", host->hostname, host->program_name, program_name);
  345. char *t = host->program_name;
  346. host->program_name = strdupz(program_name);
  347. freez(t);
  348. }
  349. if(strcmp(host->program_version, program_version) != 0) {
  350. info("Host '%s' switched program version from '%s' to '%s'", host->hostname, host->program_version, program_version);
  351. char *t = host->program_version;
  352. host->program_version = strdupz(program_version);
  353. freez(t);
  354. }
  355. if(host->rrd_update_every != update_every)
  356. error("Host '%s' has an update frequency of %d seconds, but the wanted one is %d seconds. Restart netdata here to apply the new settings.", host->hostname, host->rrd_update_every, update_every);
  357. if(host->rrd_history_entries < history)
  358. error("Host '%s' has history of %ld entries, but the wanted one is %ld entries. Restart netdata here to apply the new settings.", host->hostname, host->rrd_history_entries, history);
  359. if(host->rrd_memory_mode != mode)
  360. error("Host '%s' has memory mode '%s', but the wanted one is '%s'. Restart netdata here to apply the new settings.", host->hostname, rrd_memory_mode_name(host->rrd_memory_mode), rrd_memory_mode_name(mode));
  361. // update host tags
  362. rrdhost_init_tags(host, tags);
  363. }
  364. rrdhost_cleanup_orphan_hosts_nolock(host);
  365. rrd_unlock();
  366. return host;
  367. }
  368. inline int rrdhost_should_be_removed(RRDHOST *host, RRDHOST *protected, time_t now) {
  369. if(host != protected
  370. && host != localhost
  371. && rrdhost_flag_check(host, RRDHOST_FLAG_ORPHAN)
  372. && !host->connected_senders
  373. && host->senders_disconnected_time
  374. && host->senders_disconnected_time + rrdhost_free_orphan_time < now)
  375. return 1;
  376. return 0;
  377. }
  378. void rrdhost_cleanup_orphan_hosts_nolock(RRDHOST *protected) {
  379. time_t now = now_realtime_sec();
  380. RRDHOST *host;
  381. restart_after_removal:
  382. rrdhost_foreach_write(host) {
  383. if(rrdhost_should_be_removed(host, protected, now)) {
  384. info("Host '%s' with machine guid '%s' is obsolete - cleaning up.", host->hostname, host->machine_guid);
  385. if(rrdhost_flag_check(host, RRDHOST_FLAG_DELETE_ORPHAN_HOST))
  386. rrdhost_delete_charts(host);
  387. else
  388. rrdhost_save_charts(host);
  389. rrdhost_free(host);
  390. goto restart_after_removal;
  391. }
  392. }
  393. }
  394. // ----------------------------------------------------------------------------
  395. // RRDHOST global / startup initialization
  396. void rrd_init(char *hostname, struct rrdhost_system_info *system_info) {
  397. rrdset_free_obsolete_time = config_get_number(CONFIG_SECTION_GLOBAL, "cleanup obsolete charts after seconds", rrdset_free_obsolete_time);
  398. gap_when_lost_iterations_above = (int)config_get_number(CONFIG_SECTION_GLOBAL, "gap when lost iterations above", gap_when_lost_iterations_above);
  399. if (gap_when_lost_iterations_above < 1)
  400. gap_when_lost_iterations_above = 1;
  401. health_init();
  402. registry_init();
  403. rrdpush_init();
  404. debug(D_RRDHOST, "Initializing localhost with hostname '%s'", hostname);
  405. rrd_wrlock();
  406. localhost = rrdhost_create(
  407. hostname
  408. , registry_get_this_machine_hostname()
  409. , registry_get_this_machine_guid()
  410. , os_type
  411. , netdata_configured_timezone
  412. , config_get(CONFIG_SECTION_BACKEND, "host tags", "")
  413. , program_name
  414. , program_version
  415. , default_rrd_update_every
  416. , default_rrd_history_entries
  417. , default_rrd_memory_mode
  418. , default_health_enabled
  419. , default_rrdpush_enabled
  420. , default_rrdpush_destination
  421. , default_rrdpush_api_key
  422. , default_rrdpush_send_charts_matching
  423. , system_info
  424. , 1
  425. );
  426. rrd_unlock();
  427. web_client_api_v1_management_init();
  428. }
  429. // ----------------------------------------------------------------------------
  430. // RRDHOST - lock validations
  431. // there are only used when NETDATA_INTERNAL_CHECKS is set
  432. void __rrdhost_check_rdlock(RRDHOST *host, const char *file, const char *function, const unsigned long line) {
  433. debug(D_RRDHOST, "Checking read lock on host '%s'", host->hostname);
  434. int ret = netdata_rwlock_trywrlock(&host->rrdhost_rwlock);
  435. if(ret == 0)
  436. fatal("RRDHOST '%s' should be read-locked, but it is not, at function %s() at line %lu of file '%s'", host->hostname, function, line, file);
  437. }
  438. void __rrdhost_check_wrlock(RRDHOST *host, const char *file, const char *function, const unsigned long line) {
  439. debug(D_RRDHOST, "Checking write lock on host '%s'", host->hostname);
  440. int ret = netdata_rwlock_tryrdlock(&host->rrdhost_rwlock);
  441. if(ret == 0)
  442. fatal("RRDHOST '%s' should be write-locked, but it is not, at function %s() at line %lu of file '%s'", host->hostname, function, line, file);
  443. }
  444. void __rrd_check_rdlock(const char *file, const char *function, const unsigned long line) {
  445. debug(D_RRDHOST, "Checking read lock on all RRDs");
  446. int ret = netdata_rwlock_trywrlock(&rrd_rwlock);
  447. if(ret == 0)
  448. fatal("RRDs should be read-locked, but it are not, at function %s() at line %lu of file '%s'", function, line, file);
  449. }
  450. void __rrd_check_wrlock(const char *file, const char *function, const unsigned long line) {
  451. debug(D_RRDHOST, "Checking write lock on all RRDs");
  452. int ret = netdata_rwlock_tryrdlock(&rrd_rwlock);
  453. if(ret == 0)
  454. fatal("RRDs should be write-locked, but it are not, at function %s() at line %lu of file '%s'", function, line, file);
  455. }
  456. // ----------------------------------------------------------------------------
  457. // RRDHOST - free
  458. void rrdhost_system_info_free(struct rrdhost_system_info *system_info) {
  459. info("SYSTEM_INFO: free %p", system_info);
  460. if(likely(system_info)) {
  461. freez(system_info->os_name);
  462. freez(system_info->os_id);
  463. freez(system_info->os_id_like);
  464. freez(system_info->os_version);
  465. freez(system_info->os_version_id);
  466. freez(system_info->os_detection);
  467. freez(system_info->kernel_name);
  468. freez(system_info->kernel_version);
  469. freez(system_info->architecture);
  470. freez(system_info->virtualization);
  471. freez(system_info->virt_detection);
  472. freez(system_info->container);
  473. freez(system_info->container_detection);
  474. freez(system_info);
  475. }
  476. }
  477. void rrdhost_free(RRDHOST *host) {
  478. if(!host) return;
  479. info("Freeing all memory for host '%s'...", host->hostname);
  480. rrd_check_wrlock(); // make sure the RRDs are write locked
  481. // stop a possibly running thread
  482. rrdpush_sender_thread_stop(host);
  483. rrdhost_wrlock(host); // lock this RRDHOST
  484. // ------------------------------------------------------------------------
  485. // release its children resources
  486. while(host->rrdset_root)
  487. rrdset_free(host->rrdset_root);
  488. while(host->alarms)
  489. rrdcalc_unlink_and_free(host, host->alarms);
  490. RRDCALC *rc,*nc;
  491. for(rc = host->alarms_with_foreach; rc ; rc = nc) {
  492. nc = rc->next;
  493. rrdcalc_free(rc);
  494. }
  495. host->alarms_with_foreach = NULL;
  496. while(host->templates)
  497. rrdcalctemplate_unlink_and_free(host, host->templates);
  498. RRDCALCTEMPLATE *rt,*next;
  499. for(rt = host->alarms_template_with_foreach; rt ; rt = next) {
  500. next = rt->next;
  501. rrdcalctemplate_free(rt);
  502. }
  503. host->alarms_template_with_foreach = NULL;
  504. debug(D_RRD_CALLS, "RRDHOST: Cleaning up remaining host variables for host '%s'", host->hostname);
  505. rrdvar_free_remaining_variables(host, &host->rrdvar_root_index);
  506. health_alarm_log_free(host);
  507. if (host->rrd_memory_mode == RRD_MEMORY_MODE_DBENGINE) {
  508. #ifdef ENABLE_DBENGINE
  509. rrdeng_exit(host->rrdeng_ctx);
  510. #endif
  511. }
  512. // ------------------------------------------------------------------------
  513. // remove it from the indexes
  514. if(rrdhost_index_del(host) != host)
  515. error("RRDHOST '%s' removed from index, deleted the wrong entry.", host->hostname);
  516. // ------------------------------------------------------------------------
  517. // unlink it from the host
  518. if(host == localhost) {
  519. localhost = host->next;
  520. }
  521. else {
  522. // find the previous one
  523. RRDHOST *h;
  524. for(h = localhost; h && h->next != host ; h = h->next) ;
  525. // bypass it
  526. if(h) h->next = host->next;
  527. else error("Request to free RRDHOST '%s': cannot find it", host->hostname);
  528. }
  529. // ------------------------------------------------------------------------
  530. // free it
  531. freez((void *)host->tags);
  532. freez((void *)host->os);
  533. freez((void *)host->timezone);
  534. freez(host->program_version);
  535. freez(host->program_name);
  536. rrdhost_system_info_free(host->system_info);
  537. freez(host->cache_dir);
  538. freez(host->varlib_dir);
  539. freez(host->rrdpush_send_api_key);
  540. freez(host->rrdpush_send_destination);
  541. freez(host->health_default_exec);
  542. freez(host->health_default_recipient);
  543. freez(host->health_log_filename);
  544. freez(host->hostname);
  545. freez(host->registry_hostname);
  546. simple_pattern_free(host->rrdpush_send_charts_matching);
  547. rrdhost_unlock(host);
  548. netdata_rwlock_destroy(&host->health_log.alarm_log_rwlock);
  549. netdata_rwlock_destroy(&host->rrdhost_rwlock);
  550. freez(host);
  551. rrd_hosts_available--;
  552. }
  553. void rrdhost_free_all(void) {
  554. rrd_wrlock();
  555. while(localhost) rrdhost_free(localhost);
  556. rrd_unlock();
  557. }
  558. // ----------------------------------------------------------------------------
  559. // RRDHOST - save host files
  560. void rrdhost_save_charts(RRDHOST *host) {
  561. if(!host) return;
  562. info("Saving/Closing database of host '%s'...", host->hostname);
  563. RRDSET *st;
  564. // we get a write lock
  565. // to ensure only one thread is saving the database
  566. rrdhost_wrlock(host);
  567. rrdset_foreach_write(st, host) {
  568. rrdset_rdlock(st);
  569. rrdset_save(st);
  570. rrdset_unlock(st);
  571. }
  572. rrdhost_unlock(host);
  573. }
  574. // ----------------------------------------------------------------------------
  575. // RRDHOST - delete host files
  576. void rrdhost_delete_charts(RRDHOST *host) {
  577. if(!host) return;
  578. info("Deleting database of host '%s'...", host->hostname);
  579. RRDSET *st;
  580. // we get a write lock
  581. // to ensure only one thread is saving the database
  582. rrdhost_wrlock(host);
  583. rrdset_foreach_write(st, host) {
  584. rrdset_rdlock(st);
  585. rrdset_delete(st);
  586. rrdset_unlock(st);
  587. }
  588. recursively_delete_dir(host->cache_dir, "left over host");
  589. rrdhost_unlock(host);
  590. }
  591. // ----------------------------------------------------------------------------
  592. // RRDHOST - cleanup host files
  593. void rrdhost_cleanup_charts(RRDHOST *host) {
  594. if(!host) return;
  595. info("Cleaning up database of host '%s'...", host->hostname);
  596. RRDSET *st;
  597. uint32_t rrdhost_delete_obsolete_charts = rrdhost_flag_check(host, RRDHOST_FLAG_DELETE_OBSOLETE_CHARTS);
  598. // we get a write lock
  599. // to ensure only one thread is saving the database
  600. rrdhost_wrlock(host);
  601. rrdset_foreach_write(st, host) {
  602. rrdset_rdlock(st);
  603. if(rrdhost_delete_obsolete_charts && rrdset_flag_check(st, RRDSET_FLAG_OBSOLETE))
  604. rrdset_delete(st);
  605. else if(rrdhost_delete_obsolete_charts && rrdset_flag_check(st, RRDSET_FLAG_OBSOLETE_DIMENSIONS))
  606. rrdset_delete_obsolete_dimensions(st);
  607. else
  608. rrdset_save(st);
  609. rrdset_unlock(st);
  610. }
  611. rrdhost_unlock(host);
  612. }
  613. // ----------------------------------------------------------------------------
  614. // RRDHOST - save all hosts to disk
  615. void rrdhost_save_all(void) {
  616. info("Saving database [%zu hosts(s)]...", rrd_hosts_available);
  617. rrd_rdlock();
  618. RRDHOST *host;
  619. rrdhost_foreach_read(host)
  620. rrdhost_save_charts(host);
  621. rrd_unlock();
  622. }
  623. // ----------------------------------------------------------------------------
  624. // RRDHOST - save or delete all hosts from disk
  625. void rrdhost_cleanup_all(void) {
  626. info("Cleaning up database [%zu hosts(s)]...", rrd_hosts_available);
  627. rrd_rdlock();
  628. RRDHOST *host;
  629. rrdhost_foreach_read(host) {
  630. if(host != localhost && rrdhost_flag_check(host, RRDHOST_FLAG_DELETE_OBSOLETE_CHARTS) && !host->connected_senders)
  631. rrdhost_delete_charts(host);
  632. else
  633. rrdhost_cleanup_charts(host);
  634. }
  635. rrd_unlock();
  636. }
  637. // ----------------------------------------------------------------------------
  638. // RRDHOST - save or delete all the host charts from disk
  639. void rrdhost_cleanup_obsolete_charts(RRDHOST *host) {
  640. time_t now = now_realtime_sec();
  641. RRDSET *st;
  642. uint32_t rrdhost_delete_obsolete_charts = rrdhost_flag_check(host, RRDHOST_FLAG_DELETE_OBSOLETE_CHARTS);
  643. restart_after_removal:
  644. rrdset_foreach_write(st, host) {
  645. if(unlikely(rrdset_flag_check(st, RRDSET_FLAG_OBSOLETE)
  646. && st->last_accessed_time + rrdset_free_obsolete_time < now
  647. && st->last_updated.tv_sec + rrdset_free_obsolete_time < now
  648. && st->last_collected_time.tv_sec + rrdset_free_obsolete_time < now
  649. )) {
  650. rrdset_rdlock(st);
  651. if(rrdhost_delete_obsolete_charts)
  652. rrdset_delete(st);
  653. else
  654. rrdset_save(st);
  655. rrdset_unlock(st);
  656. rrdset_free(st);
  657. goto restart_after_removal;
  658. }
  659. }
  660. }
  661. // ----------------------------------------------------------------------------
  662. // RRDHOST - set system info from environment variables
  663. int rrdhost_set_system_info_variable(struct rrdhost_system_info *system_info, char *name, char *value) {
  664. int res = 0;
  665. if(!strcmp(name, "NETDATA_SYSTEM_OS_NAME")){
  666. freez(system_info->os_name);
  667. system_info->os_name = strdupz(value);
  668. }
  669. else if(!strcmp(name, "NETDATA_SYSTEM_OS_ID")){
  670. freez(system_info->os_id);
  671. system_info->os_id = strdupz(value);
  672. }
  673. else if(!strcmp(name, "NETDATA_SYSTEM_OS_ID_LIKE")){
  674. freez(system_info->os_id_like);
  675. system_info->os_id_like = strdupz(value);
  676. }
  677. else if(!strcmp(name, "NETDATA_SYSTEM_OS_VERSION")){
  678. freez(system_info->os_version);
  679. system_info->os_version = strdupz(value);
  680. }
  681. else if(!strcmp(name, "NETDATA_SYSTEM_OS_VERSION_ID")){
  682. freez(system_info->os_version_id);
  683. system_info->os_version_id = strdupz(value);
  684. }
  685. else if(!strcmp(name, "NETDATA_SYSTEM_OS_DETECTION")){
  686. freez(system_info->os_detection);
  687. system_info->os_detection = strdupz(value);
  688. }
  689. else if(!strcmp(name, "NETDATA_SYSTEM_KERNEL_NAME")){
  690. freez(system_info->kernel_name);
  691. system_info->kernel_name = strdupz(value);
  692. }
  693. else if(!strcmp(name, "NETDATA_SYSTEM_KERNEL_VERSION")){
  694. freez(system_info->kernel_version);
  695. system_info->kernel_version = strdupz(value);
  696. }
  697. else if(!strcmp(name, "NETDATA_SYSTEM_ARCHITECTURE")){
  698. freez(system_info->architecture);
  699. system_info->architecture = strdupz(value);
  700. }
  701. else if(!strcmp(name, "NETDATA_SYSTEM_VIRTUALIZATION")){
  702. freez(system_info->virtualization);
  703. system_info->virtualization = strdupz(value);
  704. }
  705. else if(!strcmp(name, "NETDATA_SYSTEM_VIRT_DETECTION")){
  706. freez(system_info->virt_detection);
  707. system_info->virt_detection = strdupz(value);
  708. }
  709. else if(!strcmp(name, "NETDATA_SYSTEM_CONTAINER")){
  710. freez(system_info->container);
  711. system_info->container = strdupz(value);
  712. }
  713. else if(!strcmp(name, "NETDATA_SYSTEM_CONTAINER_DETECTION")){
  714. freez(system_info->container_detection);
  715. system_info->container_detection = strdupz(value);
  716. }
  717. else {
  718. res = 1;
  719. }
  720. return res;
  721. }
  722. /**
  723. * Alarm Compare ID
  724. *
  725. * Callback function used with the binary trees to compare the id of RRDCALC
  726. *
  727. * @param a a pointer to the RRDCAL item to insert,compare or update the binary tree
  728. * @param b the pointer to the binary tree.
  729. *
  730. * @return It returns 0 case the values are equal, 1 case a is bigger than b and -1 case a is smaller than b.
  731. */
  732. int alarm_compare_id(void *a, void *b) {
  733. register uint32_t hash1 = ((RRDCALC *)a)->id;
  734. register uint32_t hash2 = ((RRDCALC *)b)->id;
  735. if(hash1 < hash2) return -1;
  736. else if(hash1 > hash2) return 1;
  737. return 0;
  738. }
  739. /**
  740. * Alarm Compare NAME
  741. *
  742. * Callback function used with the binary trees to compare the name of RRDCALC
  743. *
  744. * @param a a pointer to the RRDCAL item to insert,compare or update the binary tree
  745. * @param b the pointer to the binary tree.
  746. *
  747. * @return It returns 0 case the values are equal, 1 case a is bigger than b and -1 case a is smaller than b.
  748. */
  749. int alarm_compare_name(void *a, void *b) {
  750. RRDCALC *in1 = (RRDCALC *)a;
  751. RRDCALC *in2 = (RRDCALC *)b;
  752. if(in1->hash < in2->hash) return -1;
  753. else if(in1->hash > in2->hash) return 1;
  754. return strcmp(in1->name,in2->name);
  755. }