src/rrdhost.c

   1 #define NETDATA_RRD_INTERNALS 1
   2 #include "common.h"
   3
   4 RRDHOST *localhost = NULL;
   5
   6 pthread_rwlock_t rrd_rwlock = PTHREAD_RWLOCK_INITIALIZER;
   7
   8 time_t rrdhost_free_orphan_time = 3600;
   9
  10 // ----------------------------------------------------------------------------
  11 // RRDHOST index
  12
  13 int rrdhost_compare(void* a, void* b) {
  14     if(((RRDHOST *)a)->hash_machine_guid < ((RRDHOST *)b)->hash_machine_guid) return -1;
  15     else if(((RRDHOST *)a)->hash_machine_guid > ((RRDHOST *)b)->hash_machine_guid) return 1;
  16     else return strcmp(((RRDHOST *)a)->machine_guid, ((RRDHOST *)b)->machine_guid);
  17 }
  18
  19 avl_tree_lock rrdhost_root_index = {
  20         .avl_tree = { NULL, rrdhost_compare },
  21         .rwlock = AVL_LOCK_INITIALIZER
  22 };
  23
  24 RRDHOST *rrdhost_find_guid(const char *guid, uint32_t hash) {
  25     debug(D_RRDHOST, "Searching in index for host with guid '%s'", guid);
  26
  27     RRDHOST tmp;
  28     strncpyz(tmp.machine_guid, guid, GUID_LEN);
  29     tmp.hash_machine_guid = (hash)?hash:simple_hash(tmp.machine_guid);
  30
  31     return (RRDHOST *)avl_search_lock(&(rrdhost_root_index), (avl *) &tmp);
  32 }
  33
  34 #define rrdhost_index_add(rrdhost) (RRDHOST *)avl_insert_lock(&(rrdhost_root_index), (avl *)(rrdhost))
  35 #define rrdhost_index_del(rrdhost) (RRDHOST *)avl_remove_lock(&(rrdhost_root_index), (avl *)(rrdhost))
  36
  37
  38 // ----------------------------------------------------------------------------
  39 // RRDHOST - internal helpers
  40
  41 static inline void rrdhost_init_hostname(RRDHOST *host, const char *hostname) {
  42     freez(host->hostname);
  43     host->hostname = strdupz(hostname);
  44     host->hash_hostname = simple_hash(host->hostname);
  45 }
  46
  47 static inline void rrdhost_init_os(RRDHOST *host, const char *os) {
  48     freez(host->os);
  49     host->os = strdupz(os?os:"unknown");
  50 }
  51
  52 static inline void rrdhost_init_machine_guid(RRDHOST *host, const char *machine_guid) {
  53     strncpy(host->machine_guid, machine_guid, GUID_LEN);
  54     host->machine_guid[GUID_LEN] = '\0';
  55     host->hash_machine_guid = simple_hash(host->machine_guid);
  56 }
  57
  58
  59 // ----------------------------------------------------------------------------
  60 // RRDHOST - add a host
  61
  62 RRDHOST *rrdhost_create(const char *hostname,
  63         const char *guid,
  64         const char *os,
  65         int update_every,
  66         int entries,
  67         RRD_MEMORY_MODE memory_mode,
  68         int health_enabled,
  69         int rrdpush_enabled,
  70         char *rrdpush_destination,
  71         char *rrdpush_api_key,
  72         int is_localhost
  73 ) {
  74
  75     debug(D_RRDHOST, "Host '%s': adding with guid '%s'", hostname, guid);
  76
  77     RRDHOST *host = callocz(1, sizeof(RRDHOST));
  78
  79     host->rrd_update_every    = update_every;
  80     host->rrd_history_entries = entries;
  81     host->rrd_memory_mode     = memory_mode;
  82     host->health_enabled      = (memory_mode == RRD_MEMORY_MODE_NONE)? 0 : health_enabled;
  83     host->rrdpush_enabled     = (rrdpush_enabled && rrdpush_destination && *rrdpush_destination && rrdpush_api_key && *rrdpush_api_key);
  84     host->rrdpush_destination = (host->rrdpush_enabled)?strdupz(rrdpush_destination):NULL;
  85     host->rrdpush_api_key     = (host->rrdpush_enabled)?strdupz(rrdpush_api_key):NULL;
  86
  87     host->rrdpush_pipe[0] = -1;
  88     host->rrdpush_pipe[1] = -1;
  89     host->rrdpush_socket  = -1;
  90
  91     pthread_mutex_init(&host->rrdpush_mutex, NULL);
  92     pthread_rwlock_init(&host->rrdhost_rwlock, NULL);
  93
  94     rrdhost_init_hostname(host, hostname);
  95     rrdhost_init_machine_guid(host, guid);
  96     rrdhost_init_os(host, os);
  97
  98     avl_init_lock(&(host->rrdset_root_index),      rrdset_compare);
  99     avl_init_lock(&(host->rrdset_root_index_name), rrdset_compare_name);
 100     avl_init_lock(&(host->rrdfamily_root_index),   rrdfamily_compare);
 101     avl_init_lock(&(host->variables_root_index),   rrdvar_compare);
 102
 103     // ------------------------------------------------------------------------
 104     // initialize health variables
 105
 106     host->health_log.next_log_id = 1;
 107     host->health_log.next_alarm_id = 1;
 108     host->health_log.max = 1000;
 109     host->health_log.next_log_id =
 110     host->health_log.next_alarm_id = (uint32_t)now_realtime_sec();
 111
 112     long n = config_get_number(CONFIG_SECTION_HEALTH, "in memory max health log entries", host->health_log.max);
 113     if(n < 10) {
 114         error("Host '%s': health configuration has invalid max log entries %ld. Using default %u", host->hostname, n, host->health_log.max);
 115         config_set_number(CONFIG_SECTION_HEALTH, "in memory max health log entries", (long)host->health_log.max);
 116     }
 117     else
 118         host->health_log.max = (unsigned int)n;
 119
 120     pthread_rwlock_init(&(host->health_log.alarm_log_rwlock), NULL);
 121
 122     char filename[FILENAME_MAX + 1];
 123
 124     if(is_localhost) {
 125
 126         host->cache_dir  = strdupz(netdata_configured_cache_dir);
 127         host->varlib_dir = strdupz(netdata_configured_varlib_dir);
 128
 129     }
 130     else {
 131         // this is not localhost - append our GUID to localhost path
 132
 133         snprintfz(filename, FILENAME_MAX, "%s/%s", netdata_configured_cache_dir, host->machine_guid);
 134         host->cache_dir = strdupz(filename);
 135
 136         if(host->rrd_memory_mode == RRD_MEMORY_MODE_MAP || host->rrd_memory_mode == RRD_MEMORY_MODE_SAVE) {
 137             int r = mkdir(host->cache_dir, 0775);
 138             if(r != 0 && errno != EEXIST)
 139                 error("Host '%s': cannot create directory '%s'", host->hostname, host->cache_dir);
 140         }
 141
 142         snprintfz(filename, FILENAME_MAX, "%s/%s", netdata_configured_varlib_dir, host->machine_guid);
 143         host->varlib_dir = strdupz(filename);
 144
 145         if(host->health_enabled) {
 146             int r = mkdir(host->varlib_dir, 0775);
 147             if(r != 0 && errno != EEXIST)
 148                 error("Host '%s': cannot create directory '%s'", host->hostname, host->varlib_dir);
 149
 150             snprintfz(filename, FILENAME_MAX, "%s/health", host->varlib_dir);
 151             r = mkdir(filename, 0775);
 152             if(r != 0 && errno != EEXIST)
 153                 error("Host '%s': cannot create directory '%s'", host->hostname, filename);
 154         }
 155
 156     }
 157
 158     snprintfz(filename, FILENAME_MAX, "%s/health/health-log.db", host->varlib_dir);
 159     host->health_log_filename = strdupz(filename);
 160
 161     snprintfz(filename, FILENAME_MAX, "%s/alarm-notify.sh", netdata_configured_plugins_dir);
 162     host->health_default_exec = strdupz(config_get(CONFIG_SECTION_HEALTH, "script to execute on alarm", filename));
 163     host->health_default_recipient = strdup("root");
 164
 165
 166     // ------------------------------------------------------------------------
 167     // load health configuration
 168
 169     if(host->health_enabled) {
 170         health_alarm_log_load(host);
 171         health_alarm_log_open(host);
 172
 173         rrdhost_wrlock(host);
 174         health_readdir(host, health_config_dir());
 175         rrdhost_unlock(host);
 176     }
 177
 178
 179     // ------------------------------------------------------------------------
 180     // link it and add it to the index
 181
 182     rrd_wrlock();
 183
 184     if(is_localhost) {
 185         host->next = localhost;
 186         localhost = host;
 187     }
 188     else {
 189         if(localhost) {
 190             host->next = localhost->next;
 191             localhost->next = host;
 192         }
 193         else localhost = host;
 194     }
 195
 196     RRDHOST *t = rrdhost_index_add(host);
 197
 198     if(t != host) {
 199         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);
 200         rrdhost_free(host);
 201         host = NULL;
 202     }
 203     else {
 204         info("Host '%s' with guid '%s' initialized"
 205                      ", os: %s"
 206                      ", update every: %d"
 207                      ", memory mode: %s"
 208                      ", history entries: %d"
 209                      ", streaming: %s"
 210                      " to: '%s' (api key: '%s')"
 211                      ", health: %s"
 212                      ", cache_dir: '%s'"
 213                      ", varlib_dir: '%s'"
 214                      ", health_log: '%s'"
 215                      ", alarms default handler: '%s'"
 216                      ", alarms default recipient: '%s'"
 217              , host->hostname
 218              , host->machine_guid
 219              , host->os
 220              , host->rrd_update_every
 221              , rrd_memory_mode_name(host->rrd_memory_mode)
 222              , host->rrd_history_entries
 223              , host->rrdpush_enabled?"enabled":"disabled"
 224              , host->rrdpush_destination
 225              , host->rrdpush_api_key
 226              , host->health_enabled?"enabled":"disabled"
 227              , host->cache_dir
 228              , host->varlib_dir
 229              , host->health_log_filename
 230              , host->health_default_exec
 231              , host->health_default_recipient
 232         );
 233     }
 234
 235     rrd_unlock();
 236
 237     return host;
 238 }
 239
 240 RRDHOST *rrdhost_find_or_create(
 241           const char *hostname
 242         , const char *guid
 243         , const char *os
 244         , int update_every
 245         , int history
 246         , RRD_MEMORY_MODE mode
 247         , int health_enabled
 248         , int rrdpush_enabled
 249         , char *rrdpush_destination
 250         , char *rrdpush_api_key
 251 ) {
 252     debug(D_RRDHOST, "Searching for host '%s' with guid '%s'", hostname, guid);
 253
 254     RRDHOST *host = rrdhost_find_guid(guid, 0);
 255     if(!host) {
 256         host = rrdhost_create(
 257                 hostname
 258                 , guid
 259                 , os
 260                 , update_every
 261                 , history
 262                 , mode
 263                 , health_enabled
 264                 , rrdpush_enabled
 265                 , rrdpush_destination
 266                 , rrdpush_api_key
 267                 , 0
 268         );
 269     }
 270     else {
 271         host->health_enabled = health_enabled;
 272
 273         if(strcmp(host->hostname, hostname)) {
 274             char *t = host->hostname;
 275             char *n = strdupz(hostname);
 276             host->hostname = n;
 277             freez(t);
 278         }
 279
 280         if(host->rrd_update_every != update_every)
 281             error("Host '%s' has an update frequency of %d seconds, but the wanted one is %d seconds.", host->hostname, host->rrd_update_every, update_every);
 282
 283         if(host->rrd_history_entries != history)
 284             error("Host '%s' has history of %d entries, but the wanted one is %d entries.", host->hostname, host->rrd_history_entries, history);
 285
 286         if(host->rrd_memory_mode != mode)
 287             error("Host '%s' has memory mode '%s', but the wanted one is '%s'.", host->hostname, rrd_memory_mode_name(host->rrd_memory_mode), rrd_memory_mode_name(mode));
 288     }
 289
 290     rrdhost_cleanup_remote_stale(host);
 291
 292     return host;
 293 }
 294
 295 void rrdhost_cleanup_remote_stale(RRDHOST *protected) {
 296     rrd_wrlock();
 297
 298     RRDHOST *h;
 299     rrdhost_foreach_write(h) {
 300         if(h != protected
 301            && h != localhost
 302            && !h->connected_senders
 303            && h->senders_disconnected_time + rrdhost_free_orphan_time > now_realtime_sec()) {
 304             info("Host '%s' with machine guid '%s' is obsolete - cleaning up.", h->hostname, h->machine_guid);
 305             rrdhost_save(h);
 306             rrdhost_free(h);
 307             break;
 308         }
 309     }
 310
 311     rrd_unlock();
 312 }
 313
 314 // ----------------------------------------------------------------------------
 315 // RRDHOST global / startup initialization
 316
 317 void rrd_init(char *hostname) {
 318     health_init();
 319     registry_init();
 320     rrdpush_init();
 321
 322     debug(D_RRDHOST, "Initializing localhost with hostname '%s'", hostname);
 323     localhost = rrdhost_create(
 324             hostname
 325             , registry_get_this_machine_guid()
 326             , os_type
 327             , default_rrd_update_every
 328             , default_rrd_history_entries
 329             , default_rrd_memory_mode
 330             , default_health_enabled
 331             , default_rrdpush_enabled
 332             , default_rrdpush_destination
 333             , default_rrdpush_api_key
 334             , 1
 335     );
 336 }
 337
 338 // ----------------------------------------------------------------------------
 339 // RRDHOST - lock validations
 340 // there are only used when NETDATA_INTERNAL_CHECKS is set
 341
 342 void rrdhost_check_rdlock_int(RRDHOST *host, const char *file, const char *function, const unsigned long line) {
 343     debug(D_RRDHOST, "Checking read lock on host '%s'", host->hostname);
 344
 345     int ret = pthread_rwlock_trywrlock(&host->rrdhost_rwlock);
 346     if(ret == 0)
 347         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);
 348 }
 349
 350 void rrdhost_check_wrlock_int(RRDHOST *host, const char *file, const char *function, const unsigned long line) {
 351     debug(D_RRDHOST, "Checking write lock on host '%s'", host->hostname);
 352
 353     int ret = pthread_rwlock_tryrdlock(&host->rrdhost_rwlock);
 354     if(ret == 0)
 355         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);
 356 }
 357
 358 void rrd_check_rdlock_int(const char *file, const char *function, const unsigned long line) {
 359     debug(D_RRDHOST, "Checking read lock on all RRDs");
 360
 361     int ret = pthread_rwlock_trywrlock(&rrd_rwlock);
 362     if(ret == 0)
 363         fatal("RRDs should be read-locked, but it are not, at function %s() at line %lu of file '%s'", function, line, file);
 364 }
 365
 366 void rrd_check_wrlock_int(const char *file, const char *function, const unsigned long line) {
 367     debug(D_RRDHOST, "Checking write lock on all RRDs");
 368
 369     int ret = pthread_rwlock_tryrdlock(&rrd_rwlock);
 370     if(ret == 0)
 371         fatal("RRDs should be write-locked, but it are not, at function %s() at line %lu of file '%s'", function, line, file);
 372 }
 373
 374 // ----------------------------------------------------------------------------
 375 // RRDHOST - free
 376
 377 void rrdhost_free(RRDHOST *host) {
 378     if(!host) return;
 379
 380     info("Freeing all memory for host '%s'...", host->hostname);
 381
 382     rrd_check_wrlock();     // make sure the RRDs are write locked
 383     rrdhost_wrlock(host);   // lock this RRDHOST
 384
 385     // ------------------------------------------------------------------------
 386     // release its children resources
 387
 388     while(host->rrdset_root) rrdset_free(host->rrdset_root);
 389
 390     while(host->alarms) rrdcalc_free(host, host->alarms);
 391     while(host->templates) rrdcalctemplate_free(host, host->templates);
 392     health_alarm_log_free(host);
 393
 394
 395     // ------------------------------------------------------------------------
 396     // remove it from the indexes
 397
 398     if(rrdhost_index_del(host) != host)
 399         error("RRDHOST '%s' removed from index, deleted the wrong entry.", host->hostname);
 400
 401
 402     // ------------------------------------------------------------------------
 403     // unlink it from the host
 404
 405     if(host == localhost) {
 406         localhost = host->next;
 407     }
 408     else {
 409         // find the previous one
 410         RRDHOST *h;
 411         for(h = localhost; h && h->next != host ; h = h->next) ;
 412
 413         // bypass it
 414         if(h) h->next = host->next;
 415         else error("Request to free RRDHOST '%s': cannot find it", host->hostname);
 416     }
 417
 418     // ------------------------------------------------------------------------
 419     // free it
 420
 421     rrdpush_sender_thread_stop(host);
 422
 423     freez(host->os);
 424     freez(host->cache_dir);
 425     freez(host->varlib_dir);
 426     freez(host->rrdpush_api_key);
 427     freez(host->rrdpush_destination);
 428     freez(host->health_default_exec);
 429     freez(host->health_default_recipient);
 430     freez(host->health_log_filename);
 431     freez(host->hostname);
 432     rrdhost_unlock(host);
 433     freez(host);
 434
 435     info("Host memory cleanup completed...");
 436 }
 437
 438 void rrdhost_free_all(void) {
 439     rrd_wrlock();
 440     while(localhost) rrdhost_free(localhost);
 441     rrd_unlock();
 442 }
 443
 444 // ----------------------------------------------------------------------------
 445 // RRDHOST - save
 446
 447 void rrdhost_save(RRDHOST *host) {
 448     if(!host) return;
 449
 450     info("Saving host '%s' database...", host->hostname);
 451
 452     RRDSET *st;
 453     RRDDIM *rd;
 454
 455     // we get a write lock
 456     // to ensure only one thread is saving the database
 457     rrdhost_wrlock(host);
 458
 459     rrdset_foreach_write(st, host) {
 460         rrdset_rdlock(st);
 461
 462         if(st->rrd_memory_mode == RRD_MEMORY_MODE_SAVE) {
 463             debug(D_RRD_STATS, "Saving stats '%s' to '%s'.", st->name, st->cache_filename);
 464             savememory(st->cache_filename, st, st->memsize);
 465         }
 466
 467         rrddim_foreach_read(rd, st) {
 468             if(likely(rd->rrd_memory_mode == RRD_MEMORY_MODE_SAVE)) {
 469                 debug(D_RRD_STATS, "Saving dimension '%s' to '%s'.", rd->name, rd->cache_filename);
 470                 savememory(rd->cache_filename, rd, rd->memsize);
 471             }
 472         }
 473
 474         rrdset_unlock(st);
 475     }
 476
 477     rrdhost_unlock(host);
 478 }
 479
 480 void rrdhost_save_all(void) {
 481     info("Saving database...");
 482
 483     rrd_rdlock();
 484
 485     RRDHOST *host;
 486     rrdhost_foreach_read(host)
 487         rrdhost_save(host);
 488
 489     rrd_unlock();
 490 }