src/rrdhost.c

   1 #define NETDATA_RRD_INTERNALS 1
   2 #include "common.h"
   3
   4 RRDHOST *localhost = NULL;
   5 size_t rrd_hosts_available = 0;
   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_by_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 RRDHOST *rrdhost_find_by_hostname(const char *hostname, uint32_t hash) {
  35     if(unlikely(!strcmp(hostname, "localhost")))
  36         return localhost;
  37
  38     if(unlikely(!hash)) hash = simple_hash(hostname);
  39
  40     rrd_rdlock();
  41     RRDHOST *host;
  42     rrdhost_foreach_read(host) {
  43         if(unlikely((hash == host->hash_hostname && !strcmp(hostname, host->hostname)))) {
  44             rrd_unlock();
  45             return host;
  46         }
  47     }
  48     rrd_unlock();
  49
  50     return NULL;
  51 }
  52
  53 #define rrdhost_index_add(rrdhost) (RRDHOST *)avl_insert_lock(&(rrdhost_root_index), (avl *)(rrdhost))
  54 #define rrdhost_index_del(rrdhost) (RRDHOST *)avl_remove_lock(&(rrdhost_root_index), (avl *)(rrdhost))
  55
  56
  57 // ----------------------------------------------------------------------------
  58 // RRDHOST - internal helpers
  59
  60 static inline void rrdhost_init_hostname(RRDHOST *host, const char *hostname) {
  61     freez(host->hostname);
  62     host->hostname = strdupz(hostname);
  63     host->hash_hostname = simple_hash(host->hostname);
  64 }
  65
  66 static inline void rrdhost_init_os(RRDHOST *host, const char *os) {
  67     freez(host->os);
  68     host->os = strdupz(os?os:"unknown");
  69 }
  70
  71 static inline void rrdhost_init_machine_guid(RRDHOST *host, const char *machine_guid) {
  72     strncpy(host->machine_guid, machine_guid, GUID_LEN);
  73     host->machine_guid[GUID_LEN] = '\0';
  74     host->hash_machine_guid = simple_hash(host->machine_guid);
  75 }
  76
  77
  78 // ----------------------------------------------------------------------------
  79 // RRDHOST - add a host
  80
  81 RRDHOST *rrdhost_create(const char *hostname,
  82         const char *guid,
  83         const char *os,
  84         int update_every,
  85         int entries,
  86         RRD_MEMORY_MODE memory_mode,
  87         int health_enabled,
  88         int rrdpush_enabled,
  89         char *rrdpush_destination,
  90         char *rrdpush_api_key,
  91         int is_localhost
  92 ) {
  93
  94     debug(D_RRDHOST, "Host '%s': adding with guid '%s'", hostname, guid);
  95
  96     RRDHOST *host = callocz(1, sizeof(RRDHOST));
  97
  98     host->rrd_update_every    = update_every;
  99     host->rrd_history_entries = entries;
 100     host->rrd_memory_mode     = memory_mode;
 101     host->health_enabled      = (memory_mode == RRD_MEMORY_MODE_NONE)? 0 : health_enabled;
 102     host->rrdpush_enabled     = (rrdpush_enabled && rrdpush_destination && *rrdpush_destination && rrdpush_api_key && *rrdpush_api_key);
 103     host->rrdpush_destination = (host->rrdpush_enabled)?strdupz(rrdpush_destination):NULL;
 104     host->rrdpush_api_key     = (host->rrdpush_enabled)?strdupz(rrdpush_api_key):NULL;
 105
 106     host->rrdpush_pipe[0] = -1;
 107     host->rrdpush_pipe[1] = -1;
 108     host->rrdpush_socket  = -1;
 109
 110     pthread_mutex_init(&host->rrdpush_mutex, NULL);
 111     pthread_rwlock_init(&host->rrdhost_rwlock, NULL);
 112
 113     rrdhost_init_hostname(host, hostname);
 114     rrdhost_init_machine_guid(host, guid);
 115     rrdhost_init_os(host, os);
 116
 117     avl_init_lock(&(host->rrdset_root_index),      rrdset_compare);
 118     avl_init_lock(&(host->rrdset_root_index_name), rrdset_compare_name);
 119     avl_init_lock(&(host->rrdfamily_root_index),   rrdfamily_compare);
 120     avl_init_lock(&(host->variables_root_index),   rrdvar_compare);
 121
 122     // ------------------------------------------------------------------------
 123     // initialize health variables
 124
 125     host->health_log.next_log_id = 1;
 126     host->health_log.next_alarm_id = 1;
 127     host->health_log.max = 1000;
 128     host->health_log.next_log_id =
 129     host->health_log.next_alarm_id = (uint32_t)now_realtime_sec();
 130
 131     long n = config_get_number(CONFIG_SECTION_HEALTH, "in memory max health log entries", host->health_log.max);
 132     if(n < 10) {
 133         error("Host '%s': health configuration has invalid max log entries %ld. Using default %u", host->hostname, n, host->health_log.max);
 134         config_set_number(CONFIG_SECTION_HEALTH, "in memory max health log entries", (long)host->health_log.max);
 135     }
 136     else
 137         host->health_log.max = (unsigned int)n;
 138
 139     pthread_rwlock_init(&(host->health_log.alarm_log_rwlock), NULL);
 140
 141     char filename[FILENAME_MAX + 1];
 142
 143     if(is_localhost) {
 144
 145         host->cache_dir  = strdupz(netdata_configured_cache_dir);
 146         host->varlib_dir = strdupz(netdata_configured_varlib_dir);
 147
 148     }
 149     else {
 150         // this is not localhost - append our GUID to localhost path
 151
 152         snprintfz(filename, FILENAME_MAX, "%s/%s", netdata_configured_cache_dir, host->machine_guid);
 153         host->cache_dir = strdupz(filename);
 154
 155         if(host->rrd_memory_mode == RRD_MEMORY_MODE_MAP || host->rrd_memory_mode == RRD_MEMORY_MODE_SAVE) {
 156             int r = mkdir(host->cache_dir, 0775);
 157             if(r != 0 && errno != EEXIST)
 158                 error("Host '%s': cannot create directory '%s'", host->hostname, host->cache_dir);
 159         }
 160
 161         snprintfz(filename, FILENAME_MAX, "%s/%s", netdata_configured_varlib_dir, host->machine_guid);
 162         host->varlib_dir = strdupz(filename);
 163
 164         if(host->health_enabled) {
 165             int r = mkdir(host->varlib_dir, 0775);
 166             if(r != 0 && errno != EEXIST)
 167                 error("Host '%s': cannot create directory '%s'", host->hostname, host->varlib_dir);
 168
 169             snprintfz(filename, FILENAME_MAX, "%s/health", host->varlib_dir);
 170             r = mkdir(filename, 0775);
 171             if(r != 0 && errno != EEXIST)
 172                 error("Host '%s': cannot create directory '%s'", host->hostname, filename);
 173         }
 174
 175     }
 176
 177     snprintfz(filename, FILENAME_MAX, "%s/health/health-log.db", host->varlib_dir);
 178     host->health_log_filename = strdupz(filename);
 179
 180     snprintfz(filename, FILENAME_MAX, "%s/alarm-notify.sh", netdata_configured_plugins_dir);
 181     host->health_default_exec = strdupz(config_get(CONFIG_SECTION_HEALTH, "script to execute on alarm", filename));
 182     host->health_default_recipient = strdup("root");
 183
 184
 185     // ------------------------------------------------------------------------
 186     // load health configuration
 187
 188     if(host->health_enabled) {
 189         health_alarm_log_load(host);
 190         health_alarm_log_open(host);
 191
 192         rrdhost_wrlock(host);
 193         health_readdir(host, health_config_dir());
 194         rrdhost_unlock(host);
 195     }
 196
 197
 198     // ------------------------------------------------------------------------
 199     // link it and add it to the index
 200
 201     rrd_wrlock();
 202
 203     if(is_localhost) {
 204         host->next = localhost;
 205         localhost = host;
 206     }
 207     else {
 208         if(localhost) {
 209             host->next = localhost->next;
 210             localhost->next = host;
 211         }
 212         else localhost = host;
 213     }
 214
 215     RRDHOST *t = rrdhost_index_add(host);
 216
 217     if(t != host) {
 218         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);
 219         rrdhost_free(host);
 220         host = NULL;
 221     }
 222     else {
 223         info("Host '%s' with guid '%s' initialized"
 224                      ", os %s"
 225                      ", update every %d"
 226                      ", memory mode %s"
 227                      ", history entries %d"
 228                      ", streaming %s"
 229                      " (to '%s' with api key '%s')"
 230                      ", health %s"
 231                      ", cache_dir '%s'"
 232                      ", varlib_dir '%s'"
 233                      ", health_log '%s'"
 234                      ", alarms default handler '%s'"
 235                      ", alarms default recipient '%s'"
 236              , host->hostname
 237              , host->machine_guid
 238              , host->os
 239              , host->rrd_update_every
 240              , rrd_memory_mode_name(host->rrd_memory_mode)
 241              , host->rrd_history_entries
 242              , host->rrdpush_enabled?"enabled":"disabled"
 243              , host->rrdpush_destination?host->rrdpush_destination:""
 244              , host->rrdpush_api_key?host->rrdpush_api_key:""
 245              , host->health_enabled?"enabled":"disabled"
 246              , host->cache_dir
 247              , host->varlib_dir
 248              , host->health_log_filename
 249              , host->health_default_exec
 250              , host->health_default_recipient
 251         );
 252     }
 253
 254     rrd_hosts_available++;
 255     rrd_unlock();
 256
 257     return host;
 258 }
 259
 260 RRDHOST *rrdhost_find_or_create(
 261           const char *hostname
 262         , const char *guid
 263         , const char *os
 264         , int update_every
 265         , int history
 266         , RRD_MEMORY_MODE mode
 267         , int health_enabled
 268         , int rrdpush_enabled
 269         , char *rrdpush_destination
 270         , char *rrdpush_api_key
 271 ) {
 272     debug(D_RRDHOST, "Searching for host '%s' with guid '%s'", hostname, guid);
 273
 274     RRDHOST *host = rrdhost_find_by_guid(guid, 0);
 275     if(!host) {
 276         host = rrdhost_create(
 277                 hostname
 278                 , guid
 279                 , os
 280                 , update_every
 281                 , history
 282                 , mode
 283                 , health_enabled
 284                 , rrdpush_enabled
 285                 , rrdpush_destination
 286                 , rrdpush_api_key
 287                 , 0
 288         );
 289     }
 290     else {
 291         host->health_enabled = health_enabled;
 292
 293         if(strcmp(host->hostname, hostname)) {
 294             char *t = host->hostname;
 295             char *n = strdupz(hostname);
 296             host->hostname = n;
 297             freez(t);
 298         }
 299
 300         if(host->rrd_update_every != update_every)
 301             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);
 302
 303         if(host->rrd_history_entries != history)
 304             error("Host '%s' has history of %d entries, but the wanted one is %d entries.", host->hostname, host->rrd_history_entries, history);
 305
 306         if(host->rrd_memory_mode != mode)
 307             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));
 308     }
 309
 310     rrdhost_cleanup_remote_stale(host);
 311
 312     return host;
 313 }
 314
 315 void rrdhost_cleanup_remote_stale(RRDHOST *protected) {
 316     rrd_wrlock();
 317
 318     RRDHOST *h;
 319     rrdhost_foreach_write(h) {
 320         if(h != protected
 321            && h != localhost
 322            && !h->connected_senders
 323            && h->senders_disconnected_time + rrdhost_free_orphan_time > now_realtime_sec()) {
 324             info("Host '%s' with machine guid '%s' is obsolete - cleaning up.", h->hostname, h->machine_guid);
 325             rrdhost_save(h);
 326             rrdhost_free(h);
 327             break;
 328         }
 329     }
 330
 331     rrd_unlock();
 332 }
 333
 334 // ----------------------------------------------------------------------------
 335 // RRDHOST global / startup initialization
 336
 337 void rrd_init(char *hostname) {
 338     health_init();
 339     registry_init();
 340     rrdpush_init();
 341
 342     debug(D_RRDHOST, "Initializing localhost with hostname '%s'", hostname);
 343     localhost = rrdhost_create(
 344             hostname
 345             , registry_get_this_machine_guid()
 346             , os_type
 347             , default_rrd_update_every
 348             , default_rrd_history_entries
 349             , default_rrd_memory_mode
 350             , default_health_enabled
 351             , default_rrdpush_enabled
 352             , default_rrdpush_destination
 353             , default_rrdpush_api_key
 354             , 1
 355     );
 356 }
 357
 358 // ----------------------------------------------------------------------------
 359 // RRDHOST - lock validations
 360 // there are only used when NETDATA_INTERNAL_CHECKS is set
 361
 362 void rrdhost_check_rdlock_int(RRDHOST *host, const char *file, const char *function, const unsigned long line) {
 363     debug(D_RRDHOST, "Checking read lock on host '%s'", host->hostname);
 364
 365     int ret = pthread_rwlock_trywrlock(&host->rrdhost_rwlock);
 366     if(ret == 0)
 367         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);
 368 }
 369
 370 void rrdhost_check_wrlock_int(RRDHOST *host, const char *file, const char *function, const unsigned long line) {
 371     debug(D_RRDHOST, "Checking write lock on host '%s'", host->hostname);
 372
 373     int ret = pthread_rwlock_tryrdlock(&host->rrdhost_rwlock);
 374     if(ret == 0)
 375         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);
 376 }
 377
 378 void rrd_check_rdlock_int(const char *file, const char *function, const unsigned long line) {
 379     debug(D_RRDHOST, "Checking read lock on all RRDs");
 380
 381     int ret = pthread_rwlock_trywrlock(&rrd_rwlock);
 382     if(ret == 0)
 383         fatal("RRDs should be read-locked, but it are not, at function %s() at line %lu of file '%s'", function, line, file);
 384 }
 385
 386 void rrd_check_wrlock_int(const char *file, const char *function, const unsigned long line) {
 387     debug(D_RRDHOST, "Checking write lock on all RRDs");
 388
 389     int ret = pthread_rwlock_tryrdlock(&rrd_rwlock);
 390     if(ret == 0)
 391         fatal("RRDs should be write-locked, but it are not, at function %s() at line %lu of file '%s'", function, line, file);
 392 }
 393
 394 // ----------------------------------------------------------------------------
 395 // RRDHOST - free
 396
 397 void rrdhost_free(RRDHOST *host) {
 398     if(!host) return;
 399
 400     info("Freeing all memory for host '%s'...", host->hostname);
 401
 402     rrd_check_wrlock();     // make sure the RRDs are write locked
 403     rrdhost_wrlock(host);   // lock this RRDHOST
 404
 405     // ------------------------------------------------------------------------
 406     // release its children resources
 407
 408     while(host->rrdset_root) rrdset_free(host->rrdset_root);
 409
 410     while(host->alarms) rrdcalc_free(host, host->alarms);
 411     while(host->templates) rrdcalctemplate_free(host, host->templates);
 412     health_alarm_log_free(host);
 413
 414
 415     // ------------------------------------------------------------------------
 416     // remove it from the indexes
 417
 418     if(rrdhost_index_del(host) != host)
 419         error("RRDHOST '%s' removed from index, deleted the wrong entry.", host->hostname);
 420
 421
 422     // ------------------------------------------------------------------------
 423     // unlink it from the host
 424
 425     if(host == localhost) {
 426         localhost = host->next;
 427     }
 428     else {
 429         // find the previous one
 430         RRDHOST *h;
 431         for(h = localhost; h && h->next != host ; h = h->next) ;
 432
 433         // bypass it
 434         if(h) h->next = host->next;
 435         else error("Request to free RRDHOST '%s': cannot find it", host->hostname);
 436     }
 437
 438     // ------------------------------------------------------------------------
 439     // free it
 440
 441     rrdpush_sender_thread_stop(host);
 442
 443     freez(host->os);
 444     freez(host->cache_dir);
 445     freez(host->varlib_dir);
 446     freez(host->rrdpush_api_key);
 447     freez(host->rrdpush_destination);
 448     freez(host->health_default_exec);
 449     freez(host->health_default_recipient);
 450     freez(host->health_log_filename);
 451     freez(host->hostname);
 452     rrdhost_unlock(host);
 453     freez(host);
 454
 455     rrd_hosts_available--;
 456 }
 457
 458 void rrdhost_free_all(void) {
 459     rrd_wrlock();
 460     while(localhost) rrdhost_free(localhost);
 461     rrd_unlock();
 462 }
 463
 464 // ----------------------------------------------------------------------------
 465 // RRDHOST - save
 466
 467 void rrdhost_save(RRDHOST *host) {
 468     if(!host) return;
 469
 470     info("Saving database of host '%s'...", host->hostname);
 471
 472     RRDSET *st;
 473     RRDDIM *rd;
 474
 475     // we get a write lock
 476     // to ensure only one thread is saving the database
 477     rrdhost_wrlock(host);
 478
 479     rrdset_foreach_write(st, host) {
 480         rrdset_rdlock(st);
 481
 482         if(st->rrd_memory_mode == RRD_MEMORY_MODE_SAVE) {
 483             debug(D_RRD_STATS, "Saving stats '%s' to '%s'.", st->name, st->cache_filename);
 484             savememory(st->cache_filename, st, st->memsize);
 485         }
 486
 487         rrddim_foreach_read(rd, st) {
 488             if(likely(rd->rrd_memory_mode == RRD_MEMORY_MODE_SAVE)) {
 489                 debug(D_RRD_STATS, "Saving dimension '%s' to '%s'.", rd->name, rd->cache_filename);
 490                 savememory(rd->cache_filename, rd, rd->memsize);
 491             }
 492         }
 493
 494         rrdset_unlock(st);
 495     }
 496
 497     rrdhost_unlock(host);
 498 }
 499
 500 void rrdhost_save_all(void) {
 501     info("Saving database [%zu hosts(s)]...", rrd_hosts_available);
 502
 503     rrd_rdlock();
 504
 505     RRDHOST *host;
 506     rrdhost_foreach_read(host)
 507         rrdhost_save(host);
 508
 509     rrd_unlock();
 510 }