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