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