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_machine_guid(RRDHOST *host, const char *machine_guid) {
  47     strncpy(host->machine_guid, machine_guid, GUID_LEN);
  48     host->machine_guid[GUID_LEN] = '\0';
  49     host->hash_machine_guid = simple_hash(host->machine_guid);
  50 }
  51
  52
  53 // ----------------------------------------------------------------------------
  54 // RRDHOST - add a host
  55
  56 RRDHOST *rrdhost_create(const char *hostname,
  57         const char *guid,
  58         int update_every,
  59         int entries,
  60         RRD_MEMORY_MODE memory_mode,
  61         int health_enabled) {
  62
  63     debug(D_RRDHOST, "Host '%s': adding with guid '%s'", hostname, guid);
  64
  65     RRDHOST *host = callocz(1, sizeof(RRDHOST));
  66
  67     host->rrd_update_every    = update_every;
  68     host->rrd_history_entries = entries;
  69     host->rrd_memory_mode     = memory_mode;
  70     host->health_enabled      = health_enabled;
  71
  72     pthread_rwlock_init(&(host->rrdhost_rwlock), NULL);
  73
  74     rrdhost_init_hostname(host, hostname);
  75     rrdhost_init_machine_guid(host, guid);
  76
  77     avl_init_lock(&(host->rrdset_root_index), rrdset_compare);
  78     avl_init_lock(&(host->rrdset_root_index_name), rrdset_compare_name);
  79     avl_init_lock(&(host->rrdfamily_root_index), rrdfamily_compare);
  80     avl_init_lock(&(host->variables_root_index), rrdvar_compare);
  81
  82     // ------------------------------------------------------------------------
  83     // initialize health variables
  84
  85     host->health_log.next_log_id = 1;
  86     host->health_log.next_alarm_id = 1;
  87     host->health_log.max = 1000;
  88     host->health_log.next_log_id =
  89     host->health_log.next_alarm_id = (uint32_t)now_realtime_sec();
  90
  91     long n = config_get_number("health", "in memory max health log entries", host->health_log.max);
  92     if(n < 10) {
  93         error("Host '%s': health configuration has invalid max log entries %ld. Using default %u", host->hostname, n, host->health_log.max);
  94         config_set_number("health", "in memory max health log entries", (long)host->health_log.max);
  95     }
  96     else
  97         host->health_log.max = (unsigned int)n;
  98
  99     pthread_rwlock_init(&(host->health_log.alarm_log_rwlock), NULL);
 100
 101     char filename[FILENAME_MAX + 1];
 102
 103     if(!localhost) {
 104         // this is localhost
 105
 106         host->cache_dir = strdupz(netdata_configured_cache_dir);
 107         host->varlib_dir = strdupz(netdata_configured_varlib_dir);
 108
 109         snprintfz(filename, FILENAME_MAX, "%s/health/health-log.db", host->varlib_dir);
 110         host->health_log_filename = strdupz(config_get("health", "health db file", filename));
 111
 112     }
 113     else {
 114         // this is not localhost - append our GUID to localhost path
 115
 116         snprintfz(filename, FILENAME_MAX, "%s/%s", netdata_configured_cache_dir, host->machine_guid);
 117         host->cache_dir = strdupz(filename);
 118
 119         if(host->rrd_memory_mode == RRD_MEMORY_MODE_MAP || host->rrd_memory_mode == RRD_MEMORY_MODE_SAVE) {
 120             int r = mkdir(host->cache_dir, 0775);
 121             if(r != 0 && errno != EEXIST)
 122                 error("Host '%s': cannot create directory '%s'", host->hostname, host->cache_dir);
 123         }
 124
 125         snprintfz(filename, FILENAME_MAX, "%s/%s", netdata_configured_varlib_dir, host->machine_guid);
 126         host->varlib_dir = strdupz(filename);
 127
 128         if(host->health_enabled) {
 129             int r = mkdir(host->varlib_dir, 0775);
 130             if(r != 0 && errno != EEXIST)
 131                 error("Host '%s': cannot create directory '%s'", host->hostname, host->varlib_dir);
 132         }
 133
 134         snprintfz(filename, FILENAME_MAX, "%s/health", host->varlib_dir);
 135         int r = mkdir(filename, 0775);
 136         if(r != 0 && errno != EEXIST)
 137             error("Host '%s': cannot create directory '%s'", host->hostname, filename);
 138
 139         snprintfz(filename, FILENAME_MAX, "%s/health/health-log.db", host->varlib_dir);
 140         host->health_log_filename = strdupz(filename);
 141
 142     }
 143
 144     snprintfz(filename, FILENAME_MAX, "%s/alarm-notify.sh", netdata_configured_plugins_dir);
 145     host->health_default_exec = strdupz(config_get("health", "script to execute on alarm", filename));
 146     host->health_default_recipient = strdup("root");
 147
 148
 149     // ------------------------------------------------------------------------
 150     // load health configuration
 151
 152     health_alarm_log_load(host);
 153     health_alarm_log_open(host);
 154
 155     rrdhost_wrlock(host);
 156     health_readdir(host, health_config_dir());
 157     rrdhost_unlock(host);
 158
 159
 160     // ------------------------------------------------------------------------
 161     // link it and add it to the index
 162
 163     rrd_wrlock();
 164
 165     if(localhost) {
 166         host->next = localhost->next;
 167         localhost->next = host;
 168     }
 169
 170     if(rrdhost_index_add(host) != host)
 171         fatal("Host '%s': cannot add host to index. It already exists.", hostname);
 172
 173     rrd_unlock();
 174
 175     debug(D_RRDHOST, "Host '%s', added with guid '%s'", host->hostname, host->machine_guid);
 176     return host;
 177 }
 178
 179 RRDHOST *rrdhost_find_or_create(const char *hostname, const char *guid) {
 180     debug(D_RRDHOST, "Searching for host '%s' with guid '%s'", hostname, guid);
 181
 182     RRDHOST *host = rrdhost_find(guid, 0);
 183     if(!host)
 184         host = rrdhost_create(hostname,
 185                 guid,
 186                 default_rrd_update_every,
 187                 default_rrd_history_entries,
 188                 default_rrd_memory_mode,
 189                 default_health_enabled
 190         );
 191
 192     return host;
 193 }
 194
 195 // ----------------------------------------------------------------------------
 196 // RRDHOST global / startup initialization
 197
 198 void rrd_init(char *hostname) {
 199     debug(D_RRDHOST, "Initializing localhost with hostname '%s'", hostname);
 200
 201     localhost = rrdhost_create(hostname,
 202             registry_get_this_machine_guid(),
 203             default_rrd_update_every,
 204             default_rrd_history_entries,
 205             default_rrd_memory_mode,
 206             default_health_enabled
 207     );
 208 }
 209
 210 // ----------------------------------------------------------------------------
 211 // RRDHOST - lock validations
 212 // there are only used when NETDATA_INTERNAL_CHECKS is set
 213
 214 void rrdhost_check_rdlock_int(RRDHOST *host, const char *file, const char *function, const unsigned long line) {
 215     debug(D_RRDHOST, "Checking read lock on host '%s'", host->hostname);
 216
 217     int ret = pthread_rwlock_trywrlock(&host->rrdhost_rwlock);
 218     if(ret == 0)
 219         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);
 220 }
 221
 222 void rrdhost_check_wrlock_int(RRDHOST *host, const char *file, const char *function, const unsigned long line) {
 223     debug(D_RRDHOST, "Checking write lock on host '%s'", host->hostname);
 224
 225     int ret = pthread_rwlock_tryrdlock(&host->rrdhost_rwlock);
 226     if(ret == 0)
 227         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);
 228 }
 229
 230 void rrd_check_rdlock_int(const char *file, const char *function, const unsigned long line) {
 231     debug(D_RRDHOST, "Checking read lock on all RRDs");
 232
 233     int ret = pthread_rwlock_trywrlock(&rrd_rwlock);
 234     if(ret == 0)
 235         fatal("RRDs should be read-locked, but it are not, at function %s() at line %lu of file '%s'", function, line, file);
 236 }
 237
 238 void rrd_check_wrlock_int(const char *file, const char *function, const unsigned long line) {
 239     debug(D_RRDHOST, "Checking write lock on all RRDs");
 240
 241     int ret = pthread_rwlock_tryrdlock(&rrd_rwlock);
 242     if(ret == 0)
 243         fatal("RRDs should be write-locked, but it are not, at function %s() at line %lu of file '%s'", function, line, file);
 244 }
 245
 246 // ----------------------------------------------------------------------------
 247 // RRDHOST - free
 248
 249 void rrdhost_free(RRDHOST *host) {
 250     if(!host) return;
 251
 252     info("Freeing all memory for host '%s'...", host->hostname);
 253
 254     rrd_check_wrlock();     // make sure the RRDs are write locked
 255     rrdhost_wrlock(host);   // lock this RRDHOST
 256
 257     // ------------------------------------------------------------------------
 258     // release its children resources
 259
 260     while(host->rrdset_root) rrdset_free(host->rrdset_root);
 261
 262     while(host->alarms) rrdcalc_free(host, host->alarms);
 263     while(host->templates) rrdcalctemplate_free(host, host->templates);
 264     health_alarm_log_free(host);
 265
 266
 267     // ------------------------------------------------------------------------
 268     // remove it from the indexes
 269
 270     if(rrdhost_index_del(host) != host)
 271         error("RRDHOST '%s' removed from index, deleted the wrong entry.", host->hostname);
 272
 273
 274     // ------------------------------------------------------------------------
 275     // unlink it from the host
 276
 277     if(host == localhost) {
 278         localhost = host->next;
 279     }
 280     else {
 281         // find the previous one
 282         RRDHOST *h;
 283         for(h = localhost; h && h->next != host ; h = h->next) ;
 284
 285         // bypass it
 286         if(h) h->next = host->next;
 287         else error("Request to free RRDHOST '%s': cannot find it", host->hostname);
 288     }
 289
 290     // ------------------------------------------------------------------------
 291     // free it
 292
 293     freez(host->cache_dir);
 294     freez(host->varlib_dir);
 295     freez(host->health_default_exec);
 296     freez(host->health_default_recipient);
 297     freez(host->health_log_filename);
 298     freez(host->hostname);
 299     rrdhost_unlock(host);
 300     freez(host);
 301
 302     info("Host memory cleanup completed...");
 303 }
 304
 305 void rrdhost_free_all(void) {
 306     rrd_wrlock();
 307     while(localhost) rrdhost_free(localhost);
 308     rrd_unlock();
 309 }
 310
 311 // ----------------------------------------------------------------------------
 312 // RRDHOST - save
 313
 314 void rrdhost_save(RRDHOST *host) {
 315     if(!host) return;
 316
 317     info("Saving host '%s' database...", host->hostname);
 318
 319     RRDSET *st;
 320     RRDDIM *rd;
 321
 322     // we get a write lock
 323     // to ensure only one thread is saving the database
 324     rrdhost_wrlock(host);
 325
 326     rrdset_foreach_write(st, host) {
 327         rrdset_rdlock(st);
 328
 329         if(st->rrd_memory_mode == RRD_MEMORY_MODE_SAVE) {
 330             debug(D_RRD_STATS, "Saving stats '%s' to '%s'.", st->name, st->cache_filename);
 331             savememory(st->cache_filename, st, st->memsize);
 332         }
 333
 334         rrddim_foreach_read(rd, st) {
 335             if(likely(rd->rrd_memory_mode == RRD_MEMORY_MODE_SAVE)) {
 336                 debug(D_RRD_STATS, "Saving dimension '%s' to '%s'.", rd->name, rd->cache_filename);
 337                 savememory(rd->cache_filename, rd, rd->memsize);
 338             }
 339         }
 340
 341         rrdset_unlock(st);
 342     }
 343
 344     rrdhost_unlock(host);
 345 }
 346
 347 void rrdhost_save_all(void) {
 348     info("Saving database...");
 349
 350     rrd_rdlock();
 351
 352     RRDHOST *host;
 353     rrdhost_foreach_read(host)
 354         rrdhost_save(host);
 355
 356     rrd_unlock();
 357 }