src/rrdcalc.c

   1 #define NETDATA_HEALTH_INTERNALS
   2 #include "common.h"
   3
   4 // ----------------------------------------------------------------------------
   5 // RRDCALC management
   6
   7 inline const char *rrdcalc_status2string(int status) {
   8     switch(status) {
   9         case RRDCALC_STATUS_REMOVED:
  10             return "REMOVED";
  11
  12         case RRDCALC_STATUS_UNDEFINED:
  13             return "UNDEFINED";
  14
  15         case RRDCALC_STATUS_UNINITIALIZED:
  16             return "UNINITIALIZED";
  17
  18         case RRDCALC_STATUS_CLEAR:
  19             return "CLEAR";
  20
  21         case RRDCALC_STATUS_RAISED:
  22             return "RAISED";
  23
  24         case RRDCALC_STATUS_WARNING:
  25             return "WARNING";
  26
  27         case RRDCALC_STATUS_CRITICAL:
  28             return "CRITICAL";
  29
  30         default:
  31             error("Unknown alarm status %d", status);
  32             return "UNKNOWN";
  33     }
  34 }
  35
  36 static void rrdsetcalc_link(RRDSET *st, RRDCALC *rc) {
  37     debug(D_HEALTH, "Health linking alarm '%s.%s' to chart '%s' of host '%s'", rc->chart?rc->chart:"NOCHART", rc->name, st->id, st->rrdhost->hostname);
  38
  39     rc->last_status_change = now_realtime_sec();
  40     rc->rrdset = st;
  41
  42     rc->rrdset_next = st->alarms;
  43     rc->rrdset_prev = NULL;
  44
  45     if(rc->rrdset_next)
  46         rc->rrdset_next->rrdset_prev = rc;
  47
  48     st->alarms = rc;
  49
  50     if(rc->update_every < rc->rrdset->update_every) {
  51         error("Health alarm '%s.%s' has update every %d, less than chart update every %d. Setting alarm update frequency to %d.", rc->rrdset->id, rc->name, rc->update_every, rc->rrdset->update_every, rc->rrdset->update_every);
  52         rc->update_every = rc->rrdset->update_every;
  53     }
  54
  55     if(!isnan(rc->green) && isnan(st->green)) {
  56         debug(D_HEALTH, "Health alarm '%s.%s' green threshold set from %Lf to %Lf.", rc->rrdset->id, rc->name, rc->rrdset->green, rc->green);
  57         st->green = rc->green;
  58     }
  59
  60     if(!isnan(rc->red) && isnan(st->red)) {
  61         debug(D_HEALTH, "Health alarm '%s.%s' red threshold set from %Lf to %Lf.", rc->rrdset->id, rc->name, rc->rrdset->red, rc->red);
  62         st->red = rc->red;
  63     }
  64
  65     rc->local  = rrdvar_create_and_index("local",  &st->variables_root_index, rc->name, RRDVAR_TYPE_CALCULATED, &rc->value);
  66     rc->family = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rc->name, RRDVAR_TYPE_CALCULATED, &rc->value);
  67
  68     char fullname[RRDVAR_MAX_LENGTH + 1];
  69     snprintfz(fullname, RRDVAR_MAX_LENGTH, "%s.%s", st->id, rc->name);
  70     rc->hostid   = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, fullname, RRDVAR_TYPE_CALCULATED, &rc->value);
  71
  72     snprintfz(fullname, RRDVAR_MAX_LENGTH, "%s.%s", st->name, rc->name);
  73     rc->hostname = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, fullname, RRDVAR_TYPE_CALCULATED, &rc->value);
  74
  75     if(!rc->units) rc->units = strdupz(st->units);
  76
  77     {
  78         time_t now = now_realtime_sec();
  79         health_alarm_log(
  80                 st->rrdhost,
  81                 rc->id,
  82                 rc->next_event_id++,
  83                 now,
  84                 rc->name,
  85                 rc->rrdset->id,
  86                 rc->rrdset->family,
  87                 rc->exec,
  88                 rc->recipient,
  89                 now - rc->last_status_change,
  90                 rc->old_value,
  91                 rc->value,
  92                 rc->status,
  93                 RRDCALC_STATUS_UNINITIALIZED,
  94                 rc->source,
  95                 rc->units,
  96                 rc->info,
  97                 0,
  98                 0
  99         );
 100     }
 101 }
 102
 103 static inline int rrdcalc_is_matching_this_rrdset(RRDCALC *rc, RRDSET *st) {
 104     if(     (rc->hash_chart == st->hash      && !strcmp(rc->chart, st->id)) ||
 105             (rc->hash_chart == st->hash_name && !strcmp(rc->chart, st->name)))
 106         return 1;
 107
 108     return 0;
 109 }
 110
 111 // this has to be called while the RRDHOST is locked
 112 inline void rrdsetcalc_link_matching(RRDSET *st) {
 113     // debug(D_HEALTH, "find matching alarms for chart '%s'", st->id);
 114
 115     RRDCALC *rc;
 116     for(rc = st->rrdhost->alarms; rc ; rc = rc->next) {
 117         if(unlikely(rc->rrdset))
 118             continue;
 119
 120         if(unlikely(rrdcalc_is_matching_this_rrdset(rc, st)))
 121             rrdsetcalc_link(st, rc);
 122     }
 123 }
 124
 125 // this has to be called while the RRDHOST is locked
 126 inline void rrdsetcalc_unlink(RRDCALC *rc) {
 127     RRDSET *st = rc->rrdset;
 128
 129     if(!st) {
 130         debug(D_HEALTH, "Requested to unlink RRDCALC '%s.%s' which is not linked to any RRDSET", rc->chart?rc->chart:"NOCHART", rc->name);
 131         error("Requested to unlink RRDCALC '%s.%s' which is not linked to any RRDSET", rc->chart?rc->chart:"NOCHART", rc->name);
 132         return;
 133     }
 134
 135     {
 136         time_t now = now_realtime_sec();
 137         health_alarm_log(
 138                 st->rrdhost,
 139                 rc->id,
 140                 rc->next_event_id++,
 141                 now,
 142                 rc->name,
 143                 rc->rrdset->id,
 144                 rc->rrdset->family,
 145                 rc->exec,
 146                 rc->recipient,
 147                 now - rc->last_status_change,
 148                 rc->old_value,
 149                 rc->value,
 150                 rc->status,
 151                 RRDCALC_STATUS_REMOVED,
 152                 rc->source,
 153                 rc->units,
 154                 rc->info,
 155                 0,
 156                 0
 157         );
 158     }
 159
 160     RRDHOST *host = st->rrdhost;
 161
 162     debug(D_HEALTH, "Health unlinking alarm '%s.%s' from chart '%s' of host '%s'", rc->chart?rc->chart:"NOCHART", rc->name, st->id, host->hostname);
 163
 164     // unlink it
 165     if(rc->rrdset_prev)
 166         rc->rrdset_prev->rrdset_next = rc->rrdset_next;
 167
 168     if(rc->rrdset_next)
 169         rc->rrdset_next->rrdset_prev = rc->rrdset_prev;
 170
 171     if(st->alarms == rc)
 172         st->alarms = rc->rrdset_next;
 173
 174     rc->rrdset_prev = rc->rrdset_next = NULL;
 175
 176     rrdvar_free(st->rrdhost, &st->variables_root_index, rc->local);
 177     rc->local = NULL;
 178
 179     rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rc->family);
 180     rc->family = NULL;
 181
 182     rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rc->hostid);
 183     rc->hostid = NULL;
 184
 185     rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rc->hostname);
 186     rc->hostname = NULL;
 187
 188     rc->rrdset = NULL;
 189
 190     // RRDCALC will remain in RRDHOST
 191     // so that if the matching chart is found in the future
 192     // it will be applied automatically
 193 }
 194
 195 RRDCALC *rrdcalc_find(RRDSET *st, const char *name) {
 196     RRDCALC *rc;
 197     uint32_t hash = simple_hash(name);
 198
 199     for( rc = st->alarms; rc ; rc = rc->rrdset_next ) {
 200         if(unlikely(rc->hash == hash && !strcmp(rc->name, name)))
 201             return rc;
 202     }
 203
 204     return NULL;
 205 }
 206
 207 inline int rrdcalc_exists(RRDHOST *host, const char *chart, const char *name, uint32_t hash_chart, uint32_t hash_name) {
 208     RRDCALC *rc;
 209
 210     if(unlikely(!chart)) {
 211         error("attempt to find RRDCALC '%s' without giving a chart name", name);
 212         return 1;
 213     }
 214
 215     if(unlikely(!hash_chart)) hash_chart = simple_hash(chart);
 216     if(unlikely(!hash_name))  hash_name  = simple_hash(name);
 217
 218     // make sure it does not already exist
 219     for(rc = host->alarms; rc ; rc = rc->next) {
 220         if (unlikely(rc->chart && rc->hash == hash_name && rc->hash_chart == hash_chart && !strcmp(name, rc->name) && !strcmp(chart, rc->chart))) {
 221             debug(D_HEALTH, "Health alarm '%s.%s' already exists in host '%s'.", chart, name, host->hostname);
 222             error("Health alarm '%s.%s' already exists in host '%s'.", chart, name, host->hostname);
 223             return 1;
 224         }
 225     }
 226
 227     return 0;
 228 }
 229
 230 inline uint32_t rrdcalc_get_unique_id(RRDHOST *host, const char *chart, const char *name, uint32_t *next_event_id) {
 231     if(chart && name) {
 232         uint32_t hash_chart = simple_hash(chart);
 233         uint32_t hash_name = simple_hash(name);
 234
 235         // re-use old IDs, by looking them up in the alarm log
 236         ALARM_ENTRY *ae;
 237         for(ae = host->health_log.alarms; ae ;ae = ae->next) {
 238             if(unlikely(ae->hash_name == hash_name && ae->hash_chart == hash_chart && !strcmp(name, ae->name) && !strcmp(chart, ae->chart))) {
 239                 if(next_event_id) *next_event_id = ae->alarm_event_id + 1;
 240                 return ae->alarm_id;
 241             }
 242         }
 243     }
 244
 245     return host->health_log.next_alarm_id++;
 246 }
 247
 248 inline void rrdcalc_create_part2(RRDHOST *host, RRDCALC *rc) {
 249     rrdhost_check_rdlock(host);
 250
 251     if(rc->calculation) {
 252         rc->calculation->status = &rc->status;
 253         rc->calculation->this = &rc->value;
 254         rc->calculation->after = &rc->db_after;
 255         rc->calculation->before = &rc->db_before;
 256         rc->calculation->rrdcalc = rc;
 257     }
 258
 259     if(rc->warning) {
 260         rc->warning->status = &rc->status;
 261         rc->warning->this = &rc->value;
 262         rc->warning->after = &rc->db_after;
 263         rc->warning->before = &rc->db_before;
 264         rc->warning->rrdcalc = rc;
 265     }
 266
 267     if(rc->critical) {
 268         rc->critical->status = &rc->status;
 269         rc->critical->this = &rc->value;
 270         rc->critical->after = &rc->db_after;
 271         rc->critical->before = &rc->db_before;
 272         rc->critical->rrdcalc = rc;
 273     }
 274
 275     // link it to the host
 276     if(likely(host->alarms)) {
 277         // append it
 278         RRDCALC *t;
 279         for(t = host->alarms; t && t->next ; t = t->next) ;
 280         t->next = rc;
 281     }
 282     else {
 283         host->alarms = rc;
 284     }
 285
 286     // link it to its chart
 287     RRDSET *st;
 288     rrdset_foreach_read(st, host) {
 289         if(rrdcalc_is_matching_this_rrdset(rc, st)) {
 290             rrdsetcalc_link(st, rc);
 291             break;
 292         }
 293     }
 294 }
 295
 296 inline RRDCALC *rrdcalc_create(RRDHOST *host, RRDCALCTEMPLATE *rt, const char *chart) {
 297
 298     debug(D_HEALTH, "Health creating dynamic alarm (from template) '%s.%s'", chart, rt->name);
 299
 300     if(rrdcalc_exists(host, chart, rt->name, 0, 0))
 301         return NULL;
 302
 303     RRDCALC *rc = callocz(1, sizeof(RRDCALC));
 304     rc->next_event_id = 1;
 305     rc->id = rrdcalc_get_unique_id(host, chart, rt->name, &rc->next_event_id);
 306     rc->name = strdupz(rt->name);
 307     rc->hash = simple_hash(rc->name);
 308     rc->chart = strdupz(chart);
 309     rc->hash_chart = simple_hash(rc->chart);
 310
 311     if(rt->dimensions) rc->dimensions = strdupz(rt->dimensions);
 312
 313     rc->green = rt->green;
 314     rc->red = rt->red;
 315     rc->value = NAN;
 316     rc->old_value = NAN;
 317
 318     rc->delay_up_duration = rt->delay_up_duration;
 319     rc->delay_down_duration = rt->delay_down_duration;
 320     rc->delay_max_duration = rt->delay_max_duration;
 321     rc->delay_multiplier = rt->delay_multiplier;
 322
 323     rc->group = rt->group;
 324     rc->after = rt->after;
 325     rc->before = rt->before;
 326     rc->update_every = rt->update_every;
 327     rc->options = rt->options;
 328
 329     if(rt->exec) rc->exec = strdupz(rt->exec);
 330     if(rt->recipient) rc->recipient = strdupz(rt->recipient);
 331     if(rt->source) rc->source = strdupz(rt->source);
 332     if(rt->units) rc->units = strdupz(rt->units);
 333     if(rt->info) rc->info = strdupz(rt->info);
 334
 335     if(rt->calculation) {
 336         rc->calculation = expression_parse(rt->calculation->source, NULL, NULL);
 337         if(!rc->calculation)
 338             error("Health alarm '%s.%s': failed to parse calculation expression '%s'", chart, rt->name, rt->calculation->source);
 339     }
 340     if(rt->warning) {
 341         rc->warning = expression_parse(rt->warning->source, NULL, NULL);
 342         if(!rc->warning)
 343             error("Health alarm '%s.%s': failed to re-parse warning expression '%s'", chart, rt->name, rt->warning->source);
 344     }
 345     if(rt->critical) {
 346         rc->critical = expression_parse(rt->critical->source, NULL, NULL);
 347         if(!rc->critical)
 348             error("Health alarm '%s.%s': failed to re-parse critical expression '%s'", chart, rt->name, rt->critical->source);
 349     }
 350
 351     debug(D_HEALTH, "Health runtime added alarm '%s.%s': exec '%s', recipient '%s', green %Lf, red %Lf, lookup: group %d, after %d, before %d, options %u, dimensions '%s', update every %d, calculation '%s', warning '%s', critical '%s', source '%s', delay up %d, delay down %d, delay max %d, delay_multiplier %f",
 352             (rc->chart)?rc->chart:"NOCHART",
 353             rc->name,
 354             (rc->exec)?rc->exec:"DEFAULT",
 355             (rc->recipient)?rc->recipient:"DEFAULT",
 356             rc->green,
 357             rc->red,
 358             rc->group,
 359             rc->after,
 360             rc->before,
 361             rc->options,
 362             (rc->dimensions)?rc->dimensions:"NONE",
 363             rc->update_every,
 364             (rc->calculation)?rc->calculation->parsed_as:"NONE",
 365             (rc->warning)?rc->warning->parsed_as:"NONE",
 366             (rc->critical)?rc->critical->parsed_as:"NONE",
 367             rc->source,
 368             rc->delay_up_duration,
 369             rc->delay_down_duration,
 370             rc->delay_max_duration,
 371             rc->delay_multiplier
 372     );
 373
 374     rrdcalc_create_part2(host, rc);
 375     return rc;
 376 }
 377
 378 void rrdcalc_free(RRDHOST *host, RRDCALC *rc) {
 379     if(!rc) return;
 380
 381     debug(D_HEALTH, "Health removing alarm '%s.%s' of host '%s'", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);
 382
 383     // unlink it from RRDSET
 384     if(rc->rrdset) rrdsetcalc_unlink(rc);
 385
 386     // unlink it from RRDHOST
 387     if(unlikely(rc == host->alarms))
 388         host->alarms = rc->next;
 389
 390     else {
 391         RRDCALC *t;
 392         for(t = host->alarms; t && t->next != rc; t = t->next) ;
 393         if(t) {
 394             t->next = rc->next;
 395             rc->next = NULL;
 396         }
 397         else
 398             error("Cannot unlink alarm '%s.%s' from host '%s': not found", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);
 399     }
 400
 401     expression_free(rc->calculation);
 402     expression_free(rc->warning);
 403     expression_free(rc->critical);
 404
 405     freez(rc->name);
 406     freez(rc->chart);
 407     freez(rc->family);
 408     freez(rc->dimensions);
 409     freez(rc->exec);
 410     freez(rc->recipient);
 411     freez(rc->source);
 412     freez(rc->units);
 413     freez(rc->info);
 414     freez(rc);
 415 }