src/health.c

   1 #include "common.h"
   2
   3 #define RRDVAR_MAX_LENGTH 1024
   4
   5 static const char *health_default_exec = PLUGINS_DIR "/alarm-email.sh";
   6 static const char *health_default_recipient = "root";
   7 int health_enabled = 1;
   8
   9 // ----------------------------------------------------------------------------
  10 // RRDVAR management
  11
  12 static inline int rrdvar_fix_name(char *variable) {
  13     int fixed = 0;
  14     while(*variable) {
  15         if (!isalnum(*variable) && *variable != '.' && *variable != '_') {
  16             *variable++ = '_';
  17             fixed++;
  18         }
  19         else
  20             variable++;
  21     }
  22
  23     return fixed;
  24 }
  25
  26 int rrdvar_compare(void* a, void* b) {
  27     if(((RRDVAR *)a)->hash < ((RRDVAR *)b)->hash) return -1;
  28     else if(((RRDVAR *)a)->hash > ((RRDVAR *)b)->hash) return 1;
  29     else return strcmp(((RRDVAR *)a)->name, ((RRDVAR *)b)->name);
  30 }
  31
  32 static inline RRDVAR *rrdvar_index_add(avl_tree_lock *tree, RRDVAR *rv) {
  33     RRDVAR *ret = (RRDVAR *)avl_insert_lock(tree, (avl *)(rv));
  34     if(ret != rv)
  35         debug(D_VARIABLES, "Request to insert RRDVAR '%s' into index failed. Already exists.", rv->name);
  36
  37     return ret;
  38 }
  39
  40 static inline RRDVAR *rrdvar_index_del(avl_tree_lock *tree, RRDVAR *rv) {
  41     RRDVAR *ret = (RRDVAR *)avl_remove_lock(tree, (avl *)(rv));
  42     if(!ret)
  43         error("Request to remove RRDVAR '%s' from index failed. Not Found.", rv->name);
  44
  45     return ret;
  46 }
  47
  48 static inline RRDVAR *rrdvar_index_find(avl_tree_lock *tree, const char *name, uint32_t hash) {
  49     RRDVAR tmp;
  50     tmp.name = (char *)name;
  51     tmp.hash = (hash)?hash:simple_hash(tmp.name);
  52
  53     return (RRDVAR *)avl_search_lock(tree, (avl *)&tmp);
  54 }
  55
  56 static inline void rrdvar_free(RRDHOST *host, avl_tree_lock *tree, RRDVAR *rv) {
  57     (void)host;
  58
  59     if(!rv) return;
  60
  61     if(tree)
  62         rrdvar_index_del(tree, rv);
  63
  64     freez(rv->name);
  65     freez(rv);
  66 }
  67
  68 static inline RRDVAR *rrdvar_create_and_index(const char *scope, avl_tree_lock *tree, const char *name, int type, calculated_number *value) {
  69     char *variable = strdupz(name);
  70     rrdvar_fix_name(variable);
  71     uint32_t hash = simple_hash(variable);
  72
  73     RRDVAR *rv = rrdvar_index_find(tree, variable, hash);
  74     if(unlikely(!rv)) {
  75         debug(D_VARIABLES, "Variable '%s' not found in scope '%s'. Creating a new one.", variable, scope);
  76
  77         rv = callocz(1, sizeof(RRDVAR));
  78         rv->name = variable;
  79         rv->hash = hash;
  80         rv->type = type;
  81         rv->value = value;
  82
  83         RRDVAR *ret = rrdvar_index_add(tree, rv);
  84         if(unlikely(ret != rv)) {
  85             debug(D_VARIABLES, "Variable '%s' in scope '%s' already exists", variable, scope);
  86             rrdvar_free(NULL, NULL, rv);
  87             rv = NULL;
  88         }
  89         else
  90             debug(D_VARIABLES, "Variable '%s' created in scope '%s'", variable, scope);
  91     }
  92     else {
  93         // already exists
  94         freez(variable);
  95         rv = NULL;
  96     }
  97
  98     return rv;
  99 }
 100
 101 // ----------------------------------------------------------------------------
 102 // RRDVAR lookup
 103
 104 calculated_number rrdvar2number(RRDVAR *rv) {
 105     switch(rv->type) {
 106         case RRDVAR_TYPE_CALCULATED: {
 107             calculated_number *n = (calculated_number *)rv->value;
 108             return *n;
 109         }
 110
 111         case RRDVAR_TYPE_TIME_T: {
 112             time_t *n = (time_t *)rv->value;
 113             return *n;
 114         }
 115
 116         case RRDVAR_TYPE_COLLECTED: {
 117             collected_number *n = (collected_number *)rv->value;
 118             return *n;
 119         }
 120
 121         case RRDVAR_TYPE_TOTAL: {
 122             total_number *n = (total_number *)rv->value;
 123             return *n;
 124         }
 125
 126         case RRDVAR_TYPE_INT: {
 127             int *n = (int *)rv->value;
 128             return *n;
 129         }
 130
 131         default:
 132             error("I don't know how to convert RRDVAR type %d to calculated_number", rv->type);
 133             return NAN;
 134     }
 135 }
 136
 137 void dump_variable(void *data) {
 138     RRDVAR *rv = (RRDVAR *)data;
 139     debug(D_HEALTH, "%50s : %20.5Lf", rv->name, rrdvar2number(rv));
 140 }
 141
 142 int health_variable_lookup(const char *variable, uint32_t hash, RRDCALC *rc, calculated_number *result) {
 143     RRDSET *st = rc->rrdset;
 144     RRDVAR *rv;
 145
 146     if(!st) return 0;
 147
 148     rv = rrdvar_index_find(&st->variables_root_index, variable, hash);
 149     if(rv) {
 150         *result = rrdvar2number(rv);
 151         return 1;
 152     }
 153
 154     rv = rrdvar_index_find(&st->rrdfamily->variables_root_index, variable, hash);
 155     if(rv) {
 156         *result = rrdvar2number(rv);
 157         return 1;
 158     }
 159
 160     rv = rrdvar_index_find(&st->rrdhost->variables_root_index, variable, hash);
 161     if(rv) {
 162         *result = rrdvar2number(rv);
 163         return 1;
 164     }
 165
 166     debug(D_HEALTH, "Available local chart '%s' variables:", st->id);
 167     avl_traverse_lock(&st->variables_root_index, dump_variable);
 168
 169     debug(D_HEALTH, "Available family '%s' variables:", st->rrdfamily->family);
 170     avl_traverse_lock(&st->rrdfamily->variables_root_index, dump_variable);
 171
 172     debug(D_HEALTH, "Available host '%s' variables:", st->rrdhost->hostname);
 173     avl_traverse_lock(&st->rrdhost->variables_root_index, dump_variable);
 174
 175     return 0;
 176 }
 177
 178 // ----------------------------------------------------------------------------
 179 // RRDSETVAR management
 180
 181 RRDSETVAR *rrdsetvar_create(RRDSET *st, const char *variable, int type, void *value, uint32_t options) {
 182     debug(D_VARIABLES, "RRDVARSET create for chart id '%s' name '%s' with variable name '%s'", st->id, st->name, variable);
 183     RRDSETVAR *rs = (RRDSETVAR *)callocz(1, sizeof(RRDSETVAR));
 184
 185     char buffer[RRDVAR_MAX_LENGTH + 1];
 186     snprintfz(buffer, RRDVAR_MAX_LENGTH, "%s.%s", st->id, variable);
 187     rs->fullid = strdupz(buffer);
 188
 189     snprintfz(buffer, RRDVAR_MAX_LENGTH, "%s.%s", st->name, variable);
 190     rs->fullname = strdupz(buffer);
 191
 192     rs->variable = strdupz(variable);
 193
 194     rs->type = type;
 195     rs->value = value;
 196     rs->options = options;
 197     rs->rrdset = st;
 198
 199     rs->local       = rrdvar_create_and_index("local",  &st->variables_root_index, rs->variable, rs->type, rs->value);
 200     rs->family      = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->fullid, rs->type, rs->value);
 201     rs->host        = rrdvar_create_and_index("host",   &st->rrdhost->variables_root_index, rs->fullid, rs->type, rs->value);
 202     rs->family_name = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->fullname, rs->type, rs->value);
 203     rs->host_name   = rrdvar_create_and_index("host",   &st->rrdhost->variables_root_index, rs->fullname, rs->type, rs->value);
 204
 205     rs->next = st->variables;
 206     st->variables = rs;
 207
 208     return rs;
 209 }
 210
 211 void rrdsetvar_rename_all(RRDSET *st) {
 212     debug(D_VARIABLES, "RRDSETVAR rename for chart id '%s' name '%s'", st->id, st->name);
 213
 214     // only these 2 can change name
 215     // rs->family_name
 216     // rs->host_name
 217
 218     char buffer[RRDVAR_MAX_LENGTH + 1];
 219     RRDSETVAR *rs, *next = st->variables;
 220     while((rs = next)) {
 221         next = rs->next;
 222
 223         snprintfz(buffer, RRDVAR_MAX_LENGTH, "%s.%s", st->name, rs->variable);
 224
 225         if (strcmp(buffer, rs->fullname)) {
 226             // name changed
 227             rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family_name);
 228             rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_name);
 229
 230             freez(rs->fullname);
 231             rs->fullname = strdupz(st->name);
 232             rs->family_name = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->fullname, rs->type, rs->value);
 233             rs->host_name   = rrdvar_create_and_index("host",   &st->rrdhost->variables_root_index, rs->fullname, rs->type, rs->value);
 234         }
 235     }
 236
 237     rrdsetcalc_link_matching(st);
 238 }
 239
 240 void rrdsetvar_free(RRDSETVAR *rs) {
 241     RRDSET *st = rs->rrdset;
 242     debug(D_VARIABLES, "RRDSETVAR free for chart id '%s' name '%s', variable '%s'", st->id, st->name, rs->variable);
 243
 244     rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local);
 245     rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family);
 246     rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host);
 247     rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family_name);
 248     rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_name);
 249
 250     if(st->variables == rs) {
 251         st->variables = rs->next;
 252     }
 253     else {
 254         RRDSETVAR *t;
 255         for (t = st->variables; t && t->next != rs; t = t->next);
 256         if(!t) error("RRDSETVAR '%s' not found in chart '%s' variables linked list", rs->fullname, st->id);
 257         else t->next = rs->next;
 258     }
 259
 260     freez(rs->fullid);
 261     freez(rs->fullname);
 262     freez(rs->variable);
 263     freez(rs);
 264 }
 265
 266 // ----------------------------------------------------------------------------
 267 // RRDDIMVAR management
 268
 269 #define RRDDIMVAR_ID_MAX 1024
 270
 271 RRDDIMVAR *rrddimvar_create(RRDDIM *rd, int type, const char *prefix, const char *suffix, void *value, uint32_t options) {
 272     RRDSET *st = rd->rrdset;
 273
 274     debug(D_VARIABLES, "RRDDIMSET create for chart id '%s' name '%s', dimension id '%s', name '%s%s%s'", st->id, st->name, rd->id, (prefix)?prefix:"", rd->name, (suffix)?suffix:"");
 275
 276     if(!prefix) prefix = "";
 277     if(!suffix) suffix = "";
 278
 279     char buffer[RRDDIMVAR_ID_MAX + 1];
 280     RRDDIMVAR *rs = (RRDDIMVAR *)callocz(1, sizeof(RRDDIMVAR));
 281
 282     rs->prefix = strdupz(prefix);
 283     rs->suffix = strdupz(suffix);
 284
 285     snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s%s%s", rs->prefix, rd->id, rs->suffix);
 286     rs->id = strdupz(buffer);
 287
 288     snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s%s%s", rs->prefix, rd->name, rs->suffix);
 289     rs->name = strdupz(buffer);
 290
 291     snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->id, rs->id);
 292     rs->fullidid = strdupz(buffer);
 293
 294     snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->id, rs->name);
 295     rs->fullidname = strdupz(buffer);
 296
 297     snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->name, rs->id);
 298     rs->fullnameid = strdupz(buffer);
 299
 300     snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", rd->rrdset->name, rs->name);
 301     rs->fullnamename = strdupz(buffer);
 302
 303     rs->type = type;
 304     rs->value = value;
 305     rs->options = options;
 306     rs->rrddim = rd;
 307
 308     rs->local_id     = rrdvar_create_and_index("local", &st->variables_root_index, rs->id, rs->type, rs->value);
 309     rs->local_name   = rrdvar_create_and_index("local", &st->variables_root_index, rs->name, rs->type, rs->value);
 310
 311     rs->family_id    = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->id, rs->type, rs->value);
 312     rs->family_name  = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rs->name, rs->type, rs->value);
 313
 314     rs->host_fullidid     = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullidid, rs->type, rs->value);
 315     rs->host_fullidname   = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullidname, rs->type, rs->value);
 316     rs->host_fullnameid   = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullnameid, rs->type, rs->value);
 317     rs->host_fullnamename = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, rs->fullnamename, rs->type, rs->value);
 318
 319     rs->next = rd->variables;
 320     rd->variables = rs;
 321
 322     return rs;
 323 }
 324
 325 void rrddimvar_rename_all(RRDDIM *rd) {
 326     RRDSET *st = rd->rrdset;
 327     debug(D_VARIABLES, "RRDDIMSET rename for chart id '%s' name '%s', dimension id '%s', name '%s'", st->id, st->name, rd->id, rd->name);
 328
 329     RRDDIMVAR *rs, *next = rd->variables;
 330     while((rs = next)) {
 331         next = rs->next;
 332
 333         if (strcmp(rd->name, rs->name)) {
 334             char buffer[RRDDIMVAR_ID_MAX + 1];
 335             // name changed
 336
 337             // name
 338             rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local_name);
 339             freez(rs->name);
 340             snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s%s%s", rs->prefix, rd->name, rs->suffix);
 341             rs->name = strdupz(buffer);
 342             rs->local_name = rrdvar_create_and_index("local", &st->variables_root_index, rs->name, rs->type, rs->value);
 343
 344             rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullidname);
 345             freez(rs->fullidname);
 346             snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->id, rs->name);
 347             rs->fullidname = strdupz(buffer);
 348             rs->host_fullidname = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index,
 349                                                              rs->fullidname, rs->type, rs->value);
 350
 351             // fullnameid
 352             rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnameid);
 353             freez(rs->fullnameid);
 354             snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->name, rs->id);
 355             rs->fullnameid = strdupz(buffer);
 356             rs->host_fullnameid = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index,
 357                                                           rs->fullnameid, rs->type, rs->value);
 358
 359             // fullnamename
 360             rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnamename);
 361             freez(rs->fullnamename);
 362             snprintfz(buffer, RRDDIMVAR_ID_MAX, "%s.%s", st->name, rs->name);
 363             rs->fullnamename = strdupz(buffer);
 364             rs->host_fullnamename = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index,
 365                                                           rs->fullnamename, rs->type, rs->value);
 366         }
 367     }
 368 }
 369
 370 void rrddimvar_free(RRDDIMVAR *rs) {
 371     RRDDIM *rd = rs->rrddim;
 372     RRDSET *st = rd->rrdset;
 373     debug(D_VARIABLES, "RRDDIMSET free for chart id '%s' name '%s', dimension id '%s', name '%s', prefix='%s', suffix='%s'", st->id, st->name, rd->id, rd->name, rs->prefix, rs->suffix);
 374
 375     rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local_id);
 376     rrdvar_free(st->rrdhost, &st->variables_root_index, rs->local_name);
 377
 378     rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family_id);
 379     rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rs->family_name);
 380
 381     rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullidid);
 382     rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullidname);
 383     rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnameid);
 384     rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rs->host_fullnamename);
 385
 386     if(rd->variables == rs) {
 387         debug(D_VARIABLES, "RRDDIMSET removing first entry for chart id '%s' name '%s', dimension id '%s', name '%s'", st->id, st->name, rd->id, rd->name);
 388         rd->variables = rs->next;
 389     }
 390     else {
 391         debug(D_VARIABLES, "RRDDIMSET removing non-first entry for chart id '%s' name '%s', dimension id '%s', name '%s'", st->id, st->name, rd->id, rd->name);
 392         RRDDIMVAR *t;
 393         for (t = rd->variables; t && t->next != rs; t = t->next) ;
 394         if(!t) error("RRDDIMVAR '%s' not found in dimension '%s/%s' variables linked list", rs->name, st->id, rd->id);
 395         else t->next = rs->next;
 396     }
 397
 398     freez(rs->prefix);
 399     freez(rs->suffix);
 400     freez(rs->id);
 401     freez(rs->name);
 402     freez(rs->fullidid);
 403     freez(rs->fullidname);
 404     freez(rs->fullnameid);
 405     freez(rs->fullnamename);
 406     freez(rs);
 407 }
 408
 409 // ----------------------------------------------------------------------------
 410 // RRDCALC management
 411
 412 static inline const char *rrdcalc_status2string(int status) {
 413     switch(status) {
 414         case RRDCALC_STATUS_UNINITIALIZED:
 415             return "UNINITIALIZED";
 416
 417         case RRDCALC_STATUS_UNDEFINED:
 418             return "UNDEFINED";
 419
 420         case RRDCALC_STATUS_CLEAR:
 421             return "CLEAR";
 422
 423         case RRDCALC_STATUS_RAISED:
 424             return "RAISED";
 425
 426         case RRDCALC_STATUS_WARNING:
 427             return "WARNING";
 428
 429         case RRDCALC_STATUS_CRITICAL:
 430             return "CRITICAL";
 431
 432         default:
 433             return "UNKNOWN";
 434     }
 435 }
 436
 437 static void rrdsetcalc_link(RRDSET *st, RRDCALC *rc) {
 438     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);
 439
 440     rc->last_status_change = time(NULL);
 441     rc->rrdset = st;
 442
 443     rc->rrdset_next = st->alarms;
 444     rc->rrdset_prev = NULL;
 445
 446     if(rc->rrdset_next)
 447         rc->rrdset_next->rrdset_prev = rc;
 448
 449     st->alarms = rc;
 450
 451     if(rc->update_every < rc->rrdset->update_every) {
 452         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);
 453         rc->update_every = rc->rrdset->update_every;
 454     }
 455
 456     if(!isnan(rc->green) && isnan(st->green)) {
 457         debug(D_HEALTH, "Health alarm '%s.%s' green threshold set from %Lf to %Lf.", rc->rrdset->id, rc->name, rc->rrdset->green, rc->green);
 458         st->green = rc->green;
 459     }
 460
 461     if(!isnan(rc->red) && isnan(st->red)) {
 462         debug(D_HEALTH, "Health alarm '%s.%s' red threshold set from %Lf to %Lf.", rc->rrdset->id, rc->name, rc->rrdset->red, rc->red);
 463         st->red = rc->red;
 464     }
 465
 466     rc->local  = rrdvar_create_and_index("local",  &st->variables_root_index, rc->name, RRDVAR_TYPE_CALCULATED, &rc->value);
 467     rc->family = rrdvar_create_and_index("family", &st->rrdfamily->variables_root_index, rc->name, RRDVAR_TYPE_CALCULATED, &rc->value);
 468
 469     char fullname[RRDVAR_MAX_LENGTH + 1];
 470     snprintfz(fullname, RRDVAR_MAX_LENGTH, "%s.%s", st->id, rc->name);
 471     rc->hostid   = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, fullname, RRDVAR_TYPE_CALCULATED, &rc->value);
 472
 473     snprintfz(fullname, RRDVAR_MAX_LENGTH, "%s.%s", st->name, rc->name);
 474     rc->hostname = rrdvar_create_and_index("host", &st->rrdhost->variables_root_index, fullname, RRDVAR_TYPE_CALCULATED, &rc->value);
 475
 476         if(!rc->units) rc->units = strdupz(st->units);
 477 }
 478
 479 static inline int rrdcalc_is_matching_this_rrdset(RRDCALC *rc, RRDSET *st) {
 480     if(     (rc->hash_chart == st->hash      && !strcmp(rc->chart, st->id)) ||
 481             (rc->hash_chart == st->hash_name && !strcmp(rc->chart, st->name)))
 482         return 1;
 483
 484     return 0;
 485 }
 486
 487 // this has to be called while the RRDHOST is locked
 488 inline void rrdsetcalc_link_matching(RRDSET *st) {
 489     // debug(D_HEALTH, "find matching alarms for chart '%s'", st->id);
 490
 491     RRDCALC *rc;
 492     for(rc = st->rrdhost->alarms; rc ; rc = rc->next) {
 493         if(unlikely(rc->rrdset))
 494             continue;
 495
 496         if(unlikely(rrdcalc_is_matching_this_rrdset(rc, st)))
 497             rrdsetcalc_link(st, rc);
 498     }
 499 }
 500
 501 // this has to be called while the RRDHOST is locked
 502 inline void rrdsetcalc_unlink(RRDCALC *rc) {
 503     RRDSET *st = rc->rrdset;
 504
 505     if(!st) {
 506         debug(D_HEALTH, "Requested to unlink RRDCALC '%s.%s' which is not linked to any RRDSET", rc->chart?rc->chart:"NOCHART", rc->name);
 507         error("Requested to unlink RRDCALC '%s.%s' which is not linked to any RRDSET", rc->chart?rc->chart:"NOCHART", rc->name);
 508         return;
 509     }
 510
 511     RRDHOST *host = st->rrdhost;
 512
 513     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);
 514
 515     // unlink it
 516     if(rc->rrdset_prev)
 517         rc->rrdset_prev->rrdset_next = rc->rrdset_next;
 518
 519     if(rc->rrdset_next)
 520         rc->rrdset_next->rrdset_prev = rc->rrdset_prev;
 521
 522     if(st->alarms == rc)
 523         st->alarms = rc->rrdset_next;
 524
 525     rc->rrdset_prev = rc->rrdset_next = NULL;
 526
 527     rrdvar_free(st->rrdhost, &st->variables_root_index, rc->local);
 528     rc->local = NULL;
 529
 530     rrdvar_free(st->rrdhost, &st->rrdfamily->variables_root_index, rc->family);
 531     rc->family = NULL;
 532
 533     rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rc->hostid);
 534     rc->hostid = NULL;
 535
 536     rrdvar_free(st->rrdhost, &st->rrdhost->variables_root_index, rc->hostname);
 537     rc->hostname = NULL;
 538
 539     rc->rrdset = NULL;
 540
 541     // RRDCALC will remain in RRDHOST
 542     // so that if the matching chart is found in the future
 543     // it will be applied automatically
 544 }
 545
 546 RRDCALC *rrdcalc_find(RRDSET *st, const char *name) {
 547     RRDCALC *rc;
 548     uint32_t hash = simple_hash(name);
 549
 550     for( rc = st->alarms; rc ; rc = rc->rrdset_next ) {
 551         if(unlikely(rc->hash == hash && !strcmp(rc->name, name)))
 552             return rc;
 553     }
 554
 555     return NULL;
 556 }
 557
 558 static inline int rrdcalc_exists(RRDHOST *host, const char *chart, const char *name, uint32_t hash_chart, uint32_t hash_name) {
 559     RRDCALC *rc;
 560
 561     if(unlikely(!chart)) {
 562         error("attempt to find RRDCALC '%s' without giving a chart name", name);
 563         return 1;
 564     }
 565
 566     if(unlikely(!hash_chart)) hash_chart = simple_hash(chart);
 567     if(unlikely(!hash_name))  hash_name  = simple_hash(name);
 568
 569     // make sure it does not already exist
 570     for(rc = host->alarms; rc ; rc = rc->next) {
 571         if (unlikely(rc->chart && rc->hash == hash_name && rc->hash_chart == hash_chart && !strcmp(name, rc->name) && !strcmp(chart, rc->chart))) {
 572             debug(D_HEALTH, "Health alarm '%s.%s' already exists in host '%s'.", chart, name, host->hostname);
 573             error("Health alarm '%s.%s' already exists in host '%s'.", chart, name, host->hostname);
 574             return 1;
 575         }
 576     }
 577
 578     return 0;
 579 }
 580
 581 static inline void rrdcalc_create_part2(RRDHOST *host, RRDCALC *rc) {
 582     rrdhost_check_rdlock(host);
 583
 584     if(rc->calculation) {
 585         rc->calculation->this = &rc->value;
 586         rc->calculation->after = &rc->db_after;
 587         rc->calculation->before = &rc->db_before;
 588         rc->calculation->rrdcalc = rc;
 589     }
 590
 591     if(rc->warning) {
 592         rc->warning->this = &rc->value;
 593         rc->warning->after = &rc->db_after;
 594         rc->warning->before = &rc->db_before;
 595         rc->warning->rrdcalc = rc;
 596     }
 597
 598     if(rc->critical) {
 599         rc->critical->this = &rc->value;
 600         rc->critical->after = &rc->db_after;
 601         rc->critical->before = &rc->db_before;
 602         rc->critical->rrdcalc = rc;
 603     }
 604
 605     // link it to the host
 606     if(likely(host->alarms)) {
 607         // append it
 608         RRDCALC *t;
 609         for(t = host->alarms; t && t->next ; t = t->next) ;
 610         t->next = rc;
 611     }
 612     else {
 613         host->alarms = rc;
 614     }
 615
 616     // link it to its chart
 617     RRDSET *st;
 618     for(st = host->rrdset_root; st ; st = st->next) {
 619         if(rrdcalc_is_matching_this_rrdset(rc, st)) {
 620             rrdsetcalc_link(st, rc);
 621             break;
 622         }
 623     }
 624 }
 625
 626 static inline RRDCALC *rrdcalc_create(RRDHOST *host, const char *name, const char *chart, const char *dimensions,
 627                         const char *units, const char *info,
 628                         int group_method, int after, int before, int update_every, uint32_t options,
 629                         calculated_number green, calculated_number red,
 630                         const char *exec, const char *recipient, const char *source,
 631                         const char *calc, const char *warn, const char *crit) {
 632
 633     debug(D_HEALTH, "Health creating dynamic alarm (from template) '%s.%s'", chart, name);
 634
 635     if(rrdcalc_exists(host, chart, name, 0, 0))
 636         return NULL;
 637
 638     RRDCALC *rc = callocz(1, sizeof(RRDCALC));
 639     rc->id = host->health_log.next_alarm_id++;
 640     rc->next_event_id = 1;
 641     rc->name = strdupz(name);
 642     rc->hash = simple_hash(rc->name);
 643     rc->chart = strdupz(chart);
 644     rc->hash_chart = simple_hash(rc->chart);
 645
 646     if(dimensions) rc->dimensions = strdupz(dimensions);
 647
 648     rc->green = green;
 649     rc->red = red;
 650     rc->value = NAN;
 651     rc->old_value = NAN;
 652
 653     rc->group = group_method;
 654     rc->after = after;
 655     rc->before = before;
 656     rc->update_every = update_every;
 657     rc->options = options;
 658
 659     if(exec) rc->exec = strdupz(exec);
 660     if(recipient) rc->recipient = strdupz(recipient);
 661     if(source) rc->source = strdupz(source);
 662     if(units) rc->units = strdupz(units);
 663     if(info) rc->info = strdupz(info);
 664
 665     if(calc) {
 666         rc->calculation = expression_parse(calc, NULL, NULL);
 667         if(!rc->calculation)
 668             error("Health alarm '%s.%s': failed to parse calculation expression '%s'", chart, name, calc);
 669     }
 670     if(warn) {
 671         rc->warning = expression_parse(warn, NULL, NULL);
 672         if(!rc->warning)
 673             error("Health alarm '%s.%s': failed to re-parse warning expression '%s'", chart, name, warn);
 674     }
 675     if(crit) {
 676         rc->critical = expression_parse(crit, NULL, NULL);
 677         if(!rc->critical)
 678             error("Health alarm '%s.%s': failed to re-parse critical expression '%s'", chart, name, crit);
 679     }
 680
 681     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",
 682           (rc->chart)?rc->chart:"NOCHART",
 683           rc->name,
 684           (rc->exec)?rc->exec:"DEFAULT",
 685           (rc->recipient)?rc->recipient:"DEFAULT",
 686           rc->green,
 687           rc->red,
 688           rc->group,
 689           rc->after,
 690           rc->before,
 691           rc->options,
 692           (rc->dimensions)?rc->dimensions:"NONE",
 693           rc->update_every,
 694           (rc->calculation)?rc->calculation->parsed_as:"NONE",
 695           (rc->warning)?rc->warning->parsed_as:"NONE",
 696           (rc->critical)?rc->critical->parsed_as:"NONE",
 697           rc->source
 698     );
 699
 700     rrdcalc_create_part2(host, rc);
 701     return rc;
 702 }
 703
 704 void rrdcalc_free(RRDHOST *host, RRDCALC *rc) {
 705     if(!rc) return;
 706
 707     debug(D_HEALTH, "Health removing alarm '%s.%s' of host '%s'", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);
 708
 709     // unlink it from RRDSET
 710     if(rc->rrdset) rrdsetcalc_unlink(rc);
 711
 712     // unlink it from RRDHOST
 713     if(unlikely(rc == host->alarms))
 714         host->alarms = rc->next;
 715
 716     else if(likely(host->alarms)) {
 717         RRDCALC *t, *last = host->alarms;
 718         for(t = last->next; t && t != rc; last = t, t = t->next) ;
 719         if(last && last->next == rc)
 720             last->next = rc->next;
 721         else
 722             error("Cannot unlink alarm '%s.%s' from host '%s': not found", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);
 723     }
 724     else
 725         error("Cannot unlink unlink '%s.%s' from host '%s': This host does not have any calculations", rc->chart?rc->chart:"NOCHART", rc->name, host->hostname);
 726
 727     expression_free(rc->calculation);
 728     expression_free(rc->warning);
 729     expression_free(rc->critical);
 730
 731     freez(rc->name);
 732     freez(rc->chart);
 733     freez(rc->family);
 734     freez(rc->dimensions);
 735     freez(rc->exec);
 736     freez(rc->recipient);
 737     freez(rc->source);
 738     freez(rc->units);
 739     freez(rc->info);
 740     freez(rc);
 741 }
 742
 743 // ----------------------------------------------------------------------------
 744 // RRDCALCTEMPLATE management
 745
 746 void rrdcalctemplate_link_matching(RRDSET *st) {
 747     RRDCALCTEMPLATE *rt;
 748
 749     for(rt = st->rrdhost->templates; rt ; rt = rt->next) {
 750         if(rt->hash_context == st->hash_context && !strcmp(rt->context, st->context)) {
 751             RRDCALC *rc = rrdcalc_create(st->rrdhost, rt->name, st->id,
 752                            rt->dimensions, rt->units, rt->info, rt->group, rt->after, rt->before, rt->update_every, rt->options,
 753                            rt->green, rt->red,
 754                            rt->exec, rt->recipient, rt->source,
 755                            (rt->calculation)?rt->calculation->source:NULL,
 756                            (rt->warning)?rt->warning->source:NULL,
 757                            (rt->critical)?rt->critical->source:NULL);
 758
 759             if(unlikely(!rc))
 760                 error("Health tried to create alarm from template '%s', but it failed", rt->name);
 761
 762 #ifdef NETDATA_INTERNAL_CHECKS
 763             else if(rc->rrdset != st)
 764                 error("Health alarm '%s.%s' should be linked to chart '%s', but it is not", rc->chart?rc->chart:"NOCHART", rc->name, st->id);
 765 #endif
 766         }
 767     }
 768 }
 769
 770 static inline void rrdcalctemplate_free(RRDHOST *host, RRDCALCTEMPLATE *rt) {
 771     debug(D_HEALTH, "Health removing template '%s' of host '%s'", rt->name, host->hostname);
 772
 773     if(host->templates) {
 774         if(host->templates == rt) {
 775             host->templates = rt->next;
 776         }
 777         else {
 778             RRDCALCTEMPLATE *t, *last = host->templates;
 779             for (t = last->next; t && t != rt; last = t, t = t->next ) ;
 780             if(last && last->next == rt) {
 781                 last->next = rt->next;
 782                 rt->next = NULL;
 783             }
 784             else
 785                 error("Cannot find RRDCALCTEMPLATE '%s' linked in host '%s'", rt->name, host->hostname);
 786         }
 787     }
 788
 789     expression_free(rt->calculation);
 790     expression_free(rt->warning);
 791     expression_free(rt->critical);
 792
 793     freez(rt->name);
 794     freez(rt->exec);
 795     freez(rt->recipient);
 796     freez(rt->context);
 797     freez(rt->source);
 798     freez(rt->units);
 799     freez(rt->info);
 800     freez(rt->dimensions);
 801     freez(rt);
 802 }
 803
 804 // ----------------------------------------------------------------------------
 805 // load health configuration
 806
 807 #define HEALTH_CONF_MAX_LINE 4096
 808
 809 #define HEALTH_ALARM_KEY "alarm"
 810 #define HEALTH_TEMPLATE_KEY "template"
 811 #define HEALTH_ON_KEY "on"
 812 #define HEALTH_LOOKUP_KEY "lookup"
 813 #define HEALTH_CALC_KEY "calc"
 814 #define HEALTH_EVERY_KEY "every"
 815 #define HEALTH_GREEN_KEY "green"
 816 #define HEALTH_RED_KEY "red"
 817 #define HEALTH_WARN_KEY "warn"
 818 #define HEALTH_CRIT_KEY "crit"
 819 #define HEALTH_EXEC_KEY "exec"
 820 #define HEALTH_RECIPIENT_KEY "to"
 821 #define HEALTH_UNITS_KEY "units"
 822 #define HEALTH_INFO_KEY "info"
 823
 824 static inline int rrdcalc_add_alarm_from_config(RRDHOST *host, RRDCALC *rc) {
 825     if(!rc->chart) {
 826         error("Health configuration for alarm '%s' does not have a chart", rc->name);
 827         return 0;
 828     }
 829
 830     if(!rc->update_every) {
 831         error("Health configuration for alarm '%s.%s' has no frequency (parameter 'every'). Ignoring it.", rc->chart?rc->chart:"NOCHART", rc->name);
 832         return 0;
 833     }
 834
 835     if(!RRDCALC_HAS_DB_LOOKUP(rc) && !rc->warning && !rc->critical) {
 836         error("Health configuration for alarm '%s.%s' is useless (no calculation, no warning and no critical evaluation)", rc->chart?rc->chart:"NOCHART", rc->name);
 837         return 0;
 838     }
 839
 840     if (rrdcalc_exists(host, rc->chart, rc->name, rc->hash_chart, rc->hash))
 841         return 0;
 842
 843     debug(D_HEALTH, "Health configuration adding 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",
 844           rc->chart?rc->chart:"NOCHART",
 845           rc->name,
 846           (rc->exec)?rc->exec:"DEFAULT",
 847           (rc->recipient)?rc->recipient:"DEFAULT",
 848           rc->green,
 849           rc->red,
 850           rc->group,
 851           rc->after,
 852           rc->before,
 853           rc->options,
 854           (rc->dimensions)?rc->dimensions:"NONE",
 855           rc->update_every,
 856           (rc->calculation)?rc->calculation->parsed_as:"NONE",
 857           (rc->warning)?rc->warning->parsed_as:"NONE",
 858           (rc->critical)?rc->critical->parsed_as:"NONE",
 859           rc->source
 860     );
 861
 862     rrdcalc_create_part2(host, rc);
 863     return 1;
 864 }
 865
 866 static inline int rrdcalctemplate_add_template_from_config(RRDHOST *host, RRDCALCTEMPLATE *rt) {
 867     if(unlikely(!rt->context)) {
 868         error("Health configuration for template '%s' does not have a context", rt->name);
 869         return 0;
 870     }
 871
 872     if(unlikely(!rt->update_every)) {
 873         error("Health configuration for template '%s' has no frequency (parameter 'every'). Ignoring it.", rt->name);
 874         return 0;
 875     }
 876
 877     if(unlikely(!RRDCALCTEMPLATE_HAS_CALCULATION(rt) && !rt->warning && !rt->critical)) {
 878         error("Health configuration for template '%s' is useless (no calculation, no warning and no critical evaluation)", rt->name);
 879         return 0;
 880     }
 881
 882     RRDCALCTEMPLATE *t, *last = NULL;
 883     for (t = host->templates; t ; last = t, t = t->next) {
 884         if(unlikely(t->hash_name == rt->hash_name && !strcmp(t->name, rt->name))) {
 885             error("Health configuration template '%s' already exists for host '%s'.", rt->name, host->hostname);
 886             return 0;
 887         }
 888     }
 889
 890     debug(D_HEALTH, "Health configuration adding template '%s': context '%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'",
 891           rt->name,
 892           (rt->context)?rt->context:"NONE",
 893           (rt->exec)?rt->exec:"DEFAULT",
 894           (rt->recipient)?rt->recipient:"DEFAULT",
 895           rt->green,
 896           rt->red,
 897           rt->group,
 898           rt->after,
 899           rt->before,
 900           rt->options,
 901           (rt->dimensions)?rt->dimensions:"NONE",
 902           rt->update_every,
 903           (rt->calculation)?rt->calculation->parsed_as:"NONE",
 904           (rt->warning)?rt->warning->parsed_as:"NONE",
 905           (rt->critical)?rt->critical->parsed_as:"NONE",
 906           rt->source
 907     );
 908
 909     if(likely(last)) {
 910         last->next = rt;
 911     }
 912     else {
 913         rt->next = host->templates;
 914         host->templates = rt;
 915     }
 916
 917     return 1;
 918 }
 919
 920 static inline int health_parse_duration(char *string, int *result) {
 921     // make sure it is a number
 922     if(!*string || !(isdigit(*string) || *string == '+' || *string == '-')) {
 923         *result = 0;
 924         return 0;
 925     }
 926
 927     char *e = NULL;
 928     calculated_number n = strtold(string, &e);
 929     if(e && *e) {
 930         switch (*e) {
 931             case 'Y':
 932                 *result = (int) (n * 86400 * 365);
 933                 break;
 934             case 'M':
 935                 *result = (int) (n * 86400 * 30);
 936                 break;
 937             case 'w':
 938                 *result = (int) (n * 86400 * 7);
 939                 break;
 940             case 'd':
 941                 *result = (int) (n * 86400);
 942                 break;
 943             case 'h':
 944                 *result = (int) (n * 3600);
 945                 break;
 946             case 'm':
 947                 *result = (int) (n * 60);
 948                 break;
 949
 950             default:
 951             case 's':
 952                 *result = (int) (n);
 953                 break;
 954         }
 955     }
 956     else
 957        *result = (int)(n);
 958
 959     return 1;
 960 }
 961
 962 static inline int health_parse_db_lookup(
 963         size_t line, const char *path, const char *file, char *string,
 964         int *group_method, int *after, int *before, int *every,
 965         uint32_t *options, char **dimensions
 966 ) {
 967     debug(D_HEALTH, "Health configuration parsing database lookup %zu@%s/%s: %s", line, path, file, string);
 968
 969     if(*dimensions) freez(*dimensions);
 970     *dimensions = NULL;
 971     *after = 0;
 972     *before = 0;
 973     *every = 0;
 974     *options = 0;
 975
 976     char *s = string, *key;
 977
 978     // first is the group method
 979     key = s;
 980     while(*s && !isspace(*s)) s++;
 981     while(*s && isspace(*s)) *s++ = '\0';
 982     if(!*s) {
 983         error("Health configuration invalid chart calculation at line %zu of file '%s/%s': expected group method followed by the 'after' time, but got '%s'",
 984               line, path, file, key);
 985         return 0;
 986     }
 987
 988     if((*group_method = web_client_api_request_v1_data_group(key, -1)) == -1) {
 989         error("Health configuration at line %zu of file '%s/%s': invalid group method '%s'",
 990               line, path, file, key);
 991         return 0;
 992     }
 993
 994     // then is the 'after' time
 995     key = s;
 996     while(*s && !isspace(*s)) s++;
 997     while(*s && isspace(*s)) *s++ = '\0';
 998
 999     if(!health_parse_duration(key, after)) {
1000         error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' after group method",
1001               line, path, file, key);
1002         return 0;
1003     }
1004
1005     // sane defaults
1006     *every = abs(*after);
1007
1008     // now we may have optional parameters
1009     while(*s) {
1010         key = s;
1011         while(*s && !isspace(*s)) s++;
1012         while(*s && isspace(*s)) *s++ = '\0';
1013         if(!*key) break;
1014
1015         if(!strcasecmp(key, "at")) {
1016             char *value = s;
1017             while(*s && !isspace(*s)) s++;
1018             while(*s && isspace(*s)) *s++ = '\0';
1019
1020             if (!health_parse_duration(value, before)) {
1021                 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' for '%s' keyword",
1022                       line, path, file, value, key);
1023             }
1024         }
1025         else if(!strcasecmp(key, HEALTH_EVERY_KEY)) {
1026             char *value = s;
1027             while(*s && !isspace(*s)) s++;
1028             while(*s && isspace(*s)) *s++ = '\0';
1029
1030             if (!health_parse_duration(value, every)) {
1031                 error("Health configuration at line %zu of file '%s/%s': invalid duration '%s' for '%s' keyword",
1032                       line, path, file, value, key);
1033             }
1034         }
1035         else if(!strcasecmp(key, "absolute") || !strcasecmp(key, "abs") || !strcasecmp(key, "absolute_sum")) {
1036             *options |= RRDR_OPTION_ABSOLUTE;
1037         }
1038         else if(!strcasecmp(key, "min2max")) {
1039             *options |= RRDR_OPTION_MIN2MAX;
1040         }
1041         else if(!strcasecmp(key, "null2zero")) {
1042             *options |= RRDR_OPTION_NULL2ZERO;
1043         }
1044         else if(!strcasecmp(key, "percentage")) {
1045             *options |= RRDR_OPTION_PERCENTAGE;
1046         }
1047         else if(!strcasecmp(key, "unaligned")) {
1048             *options |= RRDR_OPTION_NOT_ALIGNED;
1049         }
1050         else if(!strcasecmp(key, "of")) {
1051             if(*s && strcasecmp(s, "all"))
1052                *dimensions = strdupz(s);
1053             break;
1054         }
1055         else {
1056             error("Health configuration at line %zu of file '%s/%s': unknown keyword '%s'",
1057                   line, path, file, key);
1058         }
1059     }
1060
1061     return 1;
1062 }
1063
1064 static inline char *health_source_file(size_t line, const char *path, const char *filename) {
1065     char buffer[FILENAME_MAX + 1];
1066     snprintfz(buffer, FILENAME_MAX, "%zu@%s/%s", line, path, filename);
1067     return strdupz(buffer);
1068 }
1069
1070 static inline void strip_quotes(char *s) {
1071     while(*s) {
1072         if(*s == '\'' || *s == '"') *s = ' ';
1073         s++;
1074     }
1075 }
1076
1077 int health_readfile(const char *path, const char *filename) {
1078     debug(D_HEALTH, "Health configuration reading file '%s/%s'", path, filename);
1079
1080     static uint32_t hash_alarm = 0, hash_template = 0, hash_on = 0, hash_calc = 0, hash_green = 0, hash_red = 0, hash_warn = 0, hash_crit = 0, hash_exec = 0, hash_every = 0, hash_lookup = 0, hash_units = 0, hash_info = 0, hash_recipient = 0;
1081     char buffer[HEALTH_CONF_MAX_LINE + 1];
1082
1083     if(unlikely(!hash_alarm)) {
1084         hash_alarm = simple_uhash(HEALTH_ALARM_KEY);
1085         hash_template = simple_uhash(HEALTH_TEMPLATE_KEY);
1086         hash_on = simple_uhash(HEALTH_ON_KEY);
1087         hash_calc = simple_uhash(HEALTH_CALC_KEY);
1088         hash_lookup = simple_uhash(HEALTH_LOOKUP_KEY);
1089         hash_green = simple_uhash(HEALTH_GREEN_KEY);
1090         hash_red = simple_uhash(HEALTH_RED_KEY);
1091         hash_warn = simple_uhash(HEALTH_WARN_KEY);
1092         hash_crit = simple_uhash(HEALTH_CRIT_KEY);
1093         hash_exec = simple_uhash(HEALTH_EXEC_KEY);
1094         hash_every = simple_uhash(HEALTH_EVERY_KEY);
1095         hash_units = simple_hash(HEALTH_UNITS_KEY);
1096         hash_info = simple_hash(HEALTH_INFO_KEY);
1097         hash_recipient = simple_hash(HEALTH_RECIPIENT_KEY);
1098     }
1099
1100     snprintfz(buffer, HEALTH_CONF_MAX_LINE, "%s/%s", path, filename);
1101     FILE *fp = fopen(buffer, "r");
1102     if(!fp) {
1103         error("Health configuration cannot read file '%s'.", buffer);
1104         return 0;
1105     }
1106
1107     RRDCALC *rc = NULL;
1108     RRDCALCTEMPLATE *rt = NULL;
1109
1110     size_t line = 0, append = 0;
1111     char *s;
1112     while((s = fgets(&buffer[append], (int)(HEALTH_CONF_MAX_LINE - append), fp)) || append) {
1113         int stop_appending = !s;
1114         line++;
1115         // info("Line %zu of file '%s/%s': '%s'", line, path, filename, s);
1116         s = trim(buffer);
1117         if(!s) continue;
1118         // info("Trimmed line %zu of file '%s/%s': '%s'", line, path, filename, s);
1119
1120         append = strlen(s);
1121         if(!stop_appending && s[append - 1] == '\\') {
1122             s[append - 1] = ' ';
1123             append = &s[append] - buffer;
1124             if(append < HEALTH_CONF_MAX_LINE)
1125                 continue;
1126             else {
1127                 error("Health configuration has too long muli-line at line %zu of file '%s/%s'.", line, path, filename);
1128             }
1129         }
1130         append = 0;
1131
1132         char *key = s;
1133         while(*s && *s != ':') s++;
1134         if(!*s) {
1135             error("Health configuration has invalid line %zu of file '%s/%s'. It does not contain a ':'. Ignoring it.", line, path, filename);
1136             continue;
1137         }
1138         *s = '\0';
1139         s++;
1140
1141         char *value = s;
1142         key = trim(key);
1143         value = trim(value);
1144
1145         if(!key) {
1146             error("Health configuration has invalid line %zu of file '%s/%s'. Keyword is empty. Ignoring it.", line, path, filename);
1147             continue;
1148         }
1149
1150         if(!value) {
1151             error("Health configuration has invalid line %zu of file '%s/%s'. value is empty. Ignoring it.", line, path, filename);
1152             continue;
1153         }
1154
1155         // info("Health file '%s/%s', key '%s', value '%s'", path, filename, key, value);
1156         uint32_t hash = simple_uhash(key);
1157
1158         if(hash == hash_alarm && !strcasecmp(key, HEALTH_ALARM_KEY)) {
1159             if(rc && !rrdcalc_add_alarm_from_config(&localhost, rc))
1160                 rrdcalc_free(&localhost, rc);
1161
1162             if(rt) {
1163                 if (!rrdcalctemplate_add_template_from_config(&localhost, rt))
1164                     rrdcalctemplate_free(&localhost, rt);
1165                 rt = NULL;
1166             }
1167
1168             rc = callocz(1, sizeof(RRDCALC));
1169             rc->id = localhost.health_log.next_alarm_id++;
1170             rc->next_event_id = 1;
1171             rc->name = strdupz(value);
1172             rc->hash = simple_hash(rc->name);
1173             rc->source = health_source_file(line, path, filename);
1174             rc->green = NAN;
1175             rc->red = NAN;
1176             rc->value = NAN;
1177             rc->old_value = NAN;
1178
1179             if(rrdvar_fix_name(rc->name))
1180                 error("Health configuration renamed alarm '%s' to '%s'", value, rc->name);
1181         }
1182         else if(hash == hash_template && !strcasecmp(key, HEALTH_TEMPLATE_KEY)) {
1183             if(rc) {
1184                 if(!rrdcalc_add_alarm_from_config(&localhost, rc))
1185                     rrdcalc_free(&localhost, rc);
1186                 rc = NULL;
1187             }
1188
1189             if(rt && !rrdcalctemplate_add_template_from_config(&localhost, rt))
1190                 rrdcalctemplate_free(&localhost, rt);
1191
1192             rt = callocz(1, sizeof(RRDCALCTEMPLATE));
1193             rt->name = strdupz(value);
1194             rt->hash_name = simple_hash(rt->name);
1195             rt->source = health_source_file(line, path, filename);
1196             rt->green = NAN;
1197             rt->red = NAN;
1198
1199             if(rrdvar_fix_name(rt->name))
1200                 error("Health configuration renamed template '%s' to '%s'", value, rt->name);
1201         }
1202         else if(rc) {
1203             if(hash == hash_on && !strcasecmp(key, HEALTH_ON_KEY)) {
1204                 if(rc->chart) {
1205                     if(strcmp(rc->chart, value))
1206                         info("Health configuration at line %zu of file '%s/%s' for alarm '%s' has key '%s' twice, once with value '%s' and later with value '%s'. Using ('%s').",
1207                              line, path, filename, rc->name, key, rc->chart, value, value);
1208
1209                     freez(rc->chart);
1210                 }
1211                 rc->chart = strdupz(value);
1212                 rc->hash_chart = simple_hash(rc->chart);
1213             }
1214             else if(hash == hash_lookup && !strcasecmp(key, HEALTH_LOOKUP_KEY)) {
1215                 health_parse_db_lookup(line, path, filename, value, &rc->group, &rc->after, &rc->before,
1216                                        &rc->update_every,
1217                                        &rc->options, &rc->dimensions);
1218             }
1219             else if(hash == hash_every && !strcasecmp(key, HEALTH_EVERY_KEY)) {
1220                 if(!health_parse_duration(value, &rc->update_every))
1221                     info("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' cannot parse duration: '%s'.",
1222                          line, path, filename, rc->name, key, value);
1223             }
1224             else if(hash == hash_green && !strcasecmp(key, HEALTH_GREEN_KEY)) {
1225                 char *e;
1226                 rc->green = strtold(value, &e);
1227                 if(e && *e) {
1228                     info("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' leaves this string unmatched: '%s'.",
1229                          line, path, filename, rc->name, key, e);
1230                 }
1231             }
1232             else if(hash == hash_red && !strcasecmp(key, HEALTH_RED_KEY)) {
1233                 char *e;
1234                 rc->red = strtold(value, &e);
1235                 if(e && *e) {
1236                     info("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' leaves this string unmatched: '%s'.",
1237                          line, path, filename, rc->name, key, e);
1238                 }
1239             }
1240             else if(hash == hash_calc && !strcasecmp(key, HEALTH_CALC_KEY)) {
1241                 const char *failed_at = NULL;
1242                 int error = 0;
1243                 rc->calculation = expression_parse(value, &failed_at, &error);
1244                 if(!rc->calculation) {
1245                     error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1246                           line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
1247                 }
1248             }
1249             else if(hash == hash_warn && !strcasecmp(key, HEALTH_WARN_KEY)) {
1250                 const char *failed_at = NULL;
1251                 int error = 0;
1252                 rc->warning = expression_parse(value, &failed_at, &error);
1253                 if(!rc->warning) {
1254                     error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1255                           line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
1256                 }
1257             }
1258             else if(hash == hash_crit && !strcasecmp(key, HEALTH_CRIT_KEY)) {
1259                 const char *failed_at = NULL;
1260                 int error = 0;
1261                 rc->critical = expression_parse(value, &failed_at, &error);
1262                 if(!rc->critical) {
1263                     error("Health configuration at line %zu of file '%s/%s' for alarm '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1264                           line, path, filename, rc->name, key, value, expression_strerror(error), failed_at);
1265                 }
1266             }
1267             else if(hash == hash_exec && !strcasecmp(key, HEALTH_EXEC_KEY)) {
1268                 if(rc->exec) {
1269                     if(strcmp(rc->exec, value))
1270                         info("Health configuration at line %zu of file '%s/%s' for alarm '%s' has key '%s' twice, once with value '%s' and later with value '%s'. Using ('%s').",
1271                              line, path, filename, rc->name, key, rc->exec, value, value);
1272
1273                     freez(rc->exec);
1274                 }
1275                 rc->exec = strdupz(value);
1276             }
1277             else if(hash == hash_recipient && !strcasecmp(key, HEALTH_RECIPIENT_KEY)) {
1278                 if(rc->recipient) {
1279                     if(strcmp(rc->recipient, value))
1280                         info("Health configuration at line %zu of file '%s/%s' for alarm '%s' has key '%s' twice, once with value '%s' and later with value '%s'. Using ('%s').",
1281                              line, path, filename, rc->name, key, rc->recipient, value, value);
1282
1283                     freez(rc->recipient);
1284                 }
1285                 rc->recipient = strdupz(value);
1286             }
1287             else if(hash == hash_units && !strcasecmp(key, HEALTH_UNITS_KEY)) {
1288                 if(rc->units) {
1289                     if(strcmp(rc->units, value))
1290                         info("Health configuration at line %zu of file '%s/%s' for alarm '%s' has key '%s' twice, once with value '%s' and later with value '%s'. Using ('%s').",
1291                              line, path, filename, rc->name, key, rc->units, value, value);
1292
1293                     freez(rc->units);
1294                 }
1295                 rc->units = strdupz(value);
1296                 strip_quotes(rc->units);
1297             }
1298             else if(hash == hash_info && !strcasecmp(key, HEALTH_INFO_KEY)) {
1299                 if(rc->info) {
1300                     if(strcmp(rc->info, value))
1301                         info("Health configuration at line %zu of file '%s/%s' for alarm '%s' has key '%s' twice, once with value '%s' and later with value '%s'. Using ('%s').",
1302                              line, path, filename, rc->name, key, rc->info, value, value);
1303
1304                     freez(rc->info);
1305                 }
1306                 rc->info = strdupz(value);
1307                 strip_quotes(rc->info);
1308             }
1309             else {
1310                 error("Health configuration at line %zu of file '%s/%s' for alarm '%s' has unknown key '%s'.",
1311                      line, path, filename, rc->name, key);
1312             }
1313         }
1314         else if(rt) {
1315             if(hash == hash_on && !strcasecmp(key, HEALTH_ON_KEY)) {
1316                 if(rt->context) {
1317                     if(strcmp(rt->context, value))
1318                         info("Health configuration at line %zu of file '%s/%s' for template '%s' has key '%s' twice, once with value '%s' and later with value '%s'. Using ('%s').",
1319                              line, path, filename, rt->name, key, rt->context, value, value);
1320
1321                     freez(rt->context);
1322                 }
1323                 rt->context = strdupz(value);
1324                 rt->hash_context = simple_hash(rt->context);
1325             }
1326             else if(hash == hash_lookup && !strcasecmp(key, HEALTH_LOOKUP_KEY)) {
1327                 health_parse_db_lookup(line, path, filename, value, &rt->group, &rt->after, &rt->before,
1328                                        &rt->update_every,
1329                                        &rt->options, &rt->dimensions);
1330             }
1331             else if(hash == hash_every && !strcasecmp(key, HEALTH_EVERY_KEY)) {
1332                 if(!health_parse_duration(value, &rt->update_every))
1333                     info("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' cannot parse duration: '%s'.",
1334                          line, path, filename, rt->name, key, value);
1335             }
1336             else if(hash == hash_green && !strcasecmp(key, HEALTH_GREEN_KEY)) {
1337                 char *e;
1338                 rt->green = strtold(value, &e);
1339                 if(e && *e) {
1340                     info("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' leaves this string unmatched: '%s'.",
1341                          line, path, filename, rt->name, key, e);
1342                 }
1343             }
1344             else if(hash == hash_red && !strcasecmp(key, HEALTH_RED_KEY)) {
1345                 char *e;
1346                 rt->red = strtold(value, &e);
1347                 if(e && *e) {
1348                     info("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' leaves this string unmatched: '%s'.",
1349                          line, path, filename, rt->name, key, e);
1350                 }
1351             }
1352             else if(hash == hash_calc && !strcasecmp(key, HEALTH_CALC_KEY)) {
1353                 const char *failed_at = NULL;
1354                 int error = 0;
1355                 rt->calculation = expression_parse(value, &failed_at, &error);
1356                 if(!rt->calculation) {
1357                     error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1358                           line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
1359                 }
1360             }
1361             else if(hash == hash_warn && !strcasecmp(key, HEALTH_WARN_KEY)) {
1362                 const char *failed_at = NULL;
1363                 int error = 0;
1364                 rt->warning = expression_parse(value, &failed_at, &error);
1365                 if(!rt->warning) {
1366                     error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1367                           line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
1368                 }
1369             }
1370             else if(hash == hash_crit && !strcasecmp(key, HEALTH_CRIT_KEY)) {
1371                 const char *failed_at = NULL;
1372                 int error = 0;
1373                 rt->critical = expression_parse(value, &failed_at, &error);
1374                 if(!rt->critical) {
1375                     error("Health configuration at line %zu of file '%s/%s' for template '%s' at key '%s' has unparse-able expression '%s': %s at '%s'",
1376                           line, path, filename, rt->name, key, value, expression_strerror(error), failed_at);
1377                 }
1378             }
1379             else if(hash == hash_exec && !strcasecmp(key, HEALTH_EXEC_KEY)) {
1380                 if(rt->exec) {
1381                     if(strcmp(rt->exec, value))
1382                         info("Health configuration at line %zu of file '%s/%s' for template '%s' has key '%s' twice, once with value '%s' and later with value '%s'. Using ('%s').",
1383                              line, path, filename, rt->name, key, rt->exec, value, value);
1384
1385                     freez(rt->exec);
1386                 }
1387                 rt->exec = strdupz(value);
1388             }
1389             else if(hash == hash_recipient && !strcasecmp(key, HEALTH_RECIPIENT_KEY)) {
1390                 if(rt->recipient) {
1391                     if(strcmp(rt->recipient, value))
1392                         info("Health configuration at line %zu of file '%s/%s' for template '%s' has key '%s' twice, once with value '%s' and later with value '%s'. Using ('%s').",
1393                              line, path, filename, rt->name, key, rt->recipient, value, value);
1394
1395                     freez(rt->recipient);
1396                 }
1397                 rt->recipient = strdupz(value);
1398             }
1399             else if(hash == hash_units && !strcasecmp(key, HEALTH_UNITS_KEY)) {
1400                 if(rt->units) {
1401                     if(strcmp(rt->units, value))
1402                         info("Health configuration at line %zu of file '%s/%s' for template '%s' has key '%s' twice, once with value '%s' and later with value '%s'. Using ('%s').",
1403                              line, path, filename, rt->name, key, rt->units, value, value);
1404
1405                     freez(rt->units);
1406                 }
1407                 rt->units = strdupz(value);
1408                 strip_quotes(rt->units);
1409             }
1410             else if(hash == hash_info && !strcasecmp(key, HEALTH_INFO_KEY)) {
1411                 if(rt->info) {
1412                     if(strcmp(rt->info, value))
1413                         info("Health configuration at line %zu of file '%s/%s' for template '%s' has key '%s' twice, once with value '%s' and later with value '%s'. Using ('%s').",
1414                              line, path, filename, rt->name, key, rt->info, value, value);
1415
1416                     freez(rt->info);
1417                 }
1418                 rt->info = strdupz(value);
1419                 strip_quotes(rt->info);
1420             }
1421             else {
1422                 error("Health configuration at line %zu of file '%s/%s' for template '%s' has unknown key '%s'.",
1423                       line, path, filename, rt->name, key);
1424             }
1425         }
1426         else {
1427             error("Health configuration at line %zu of file '%s/%s' has unknown key '%s'. Expected either '" HEALTH_ALARM_KEY "' or '" HEALTH_TEMPLATE_KEY "'.",
1428                   line, path, filename, key);
1429         }
1430     }
1431
1432     if(rc && !rrdcalc_add_alarm_from_config(&localhost, rc))
1433         rrdcalc_free(&localhost, rc);
1434
1435     if(rt && !rrdcalctemplate_add_template_from_config(&localhost, rt))
1436         rrdcalctemplate_free(&localhost, rt);
1437
1438     fclose(fp);
1439     return 1;
1440 }
1441
1442 void health_readdir(const char *path) {
1443     size_t pathlen = strlen(path);
1444
1445     debug(D_HEALTH, "Health configuration reading directory '%s'", path);
1446
1447     DIR *dir = opendir(path);
1448     if (!dir) {
1449         error("Health configuration cannot open directory '%s'.", path);
1450         return;
1451     }
1452
1453     struct dirent *de = NULL;
1454     while ((de = readdir(dir))) {
1455         size_t len = strlen(de->d_name);
1456
1457         if(de->d_type == DT_DIR
1458            && (
1459                    (de->d_name[0] == '.' && de->d_name[1] == '\0')
1460                    || (de->d_name[0] == '.' && de->d_name[1] == '.' && de->d_name[2] == '\0')
1461            )) {
1462             debug(D_HEALTH, "Ignoring directory '%s'", de->d_name);
1463             continue;
1464         }
1465
1466         else if(de->d_type == DT_DIR) {
1467             char *s = mallocz(pathlen + strlen(de->d_name) + 2);
1468             strcpy(s, path);
1469             strcat(s, "/");
1470             strcat(s, de->d_name);
1471             health_readdir(s);
1472             freez(s);
1473             continue;
1474         }
1475
1476         else if((de->d_type == DT_LNK || de->d_type == DT_REG) &&
1477                 len > 5 && !strcmp(&de->d_name[len - 5], ".conf")) {
1478             health_readfile(path, de->d_name);
1479         }
1480
1481         else debug(D_HEALTH, "Ignoring file '%s'", de->d_name);
1482     }
1483
1484     closedir(dir);
1485 }
1486
1487 static inline char *health_config_dir(void) {
1488     char buffer[FILENAME_MAX + 1];
1489     snprintfz(buffer, FILENAME_MAX, "%s/health.d", config_get("global", "config directory", CONFIG_DIR));
1490     return config_get("health", "health configuration directory", buffer);
1491 }
1492
1493 void health_init(void) {
1494     debug(D_HEALTH, "Health configuration initializing");
1495
1496     if(!(health_enabled = config_get_boolean("health", "enabled", 1))) {
1497         debug(D_HEALTH, "Health is disabled.");
1498         return;
1499     }
1500
1501     char *path = health_config_dir();
1502
1503     {
1504         char buffer[FILENAME_MAX + 1];
1505         snprintfz(buffer, FILENAME_MAX, "%s/alarm-email.sh", config_get("global", "plugins directory", PLUGINS_DIR));
1506         health_default_exec = config_get("health", "script to execute on alarm", buffer);
1507     }
1508
1509     long n = config_get_number("health", "in memory max health log entries", (long)localhost.health_log.max);
1510     if(n < 2) {
1511         error("Health configuration has invalid max log entries %ld. Using default %u", n, localhost.health_log.max);
1512         config_set_number("health", "in memory max health log entries", (long)localhost.health_log.max);
1513     }
1514     else localhost.health_log.max = (unsigned int)n;
1515
1516     rrdhost_rwlock(&localhost);
1517     health_readdir(path);
1518     rrdhost_unlock(&localhost);
1519 }
1520
1521 // ----------------------------------------------------------------------------
1522 // JSON generation
1523
1524 static inline void health_string2json(BUFFER *wb, const char *prefix, const char *label, const char *value, const char *suffix) {
1525     if(value && *value)
1526         buffer_sprintf(wb, "%s\"%s\":\"%s\"%s", prefix, label, value, suffix);
1527     else
1528         buffer_sprintf(wb, "%s\"%s\":null%s", prefix, label, suffix);
1529 }
1530
1531 static inline void health_alarm_entry2json_nolock(BUFFER *wb, ALARM_ENTRY *ae, RRDHOST *host) {
1532     buffer_sprintf(wb, "\n\t{\n"
1533                            "\t\t\"hostname\":\"%s\",\n"
1534                            "\t\t\"unique_id\":%u,\n"
1535                            "\t\t\"alarm_id\":%u,\n"
1536                            "\t\t\"alarm_event_id\":%u,\n"
1537                            "\t\t\"name\":\"%s\",\n"
1538                            "\t\t\"chart\":\"%s\",\n"
1539                            "\t\t\"family\":\"%s\",\n"
1540                            "\t\t\"processed\":%s,\n"
1541                            "\t\t\"updated\":%s,\n"
1542                            "\t\t\"exec_run\":%s,\n"
1543                            "\t\t\"exec_failed\":%s,\n"
1544                            "\t\t\"exec\":\"%s\",\n"
1545                            "\t\t\"recipient\":\"%s\",\n"
1546                            "\t\t\"exec_code\":%d,\n"
1547                            "\t\t\"source\":\"%s\",\n"
1548                            "\t\t\"units\":\"%s\",\n"
1549                            "\t\t\"info\":\"%s\",\n"
1550                            "\t\t\"when\":%lu,\n"
1551                            "\t\t\"duration\":%lu,\n"
1552                            "\t\t\"non_clear_duration\":%lu,\n"
1553                            "\t\t\"status\":\"%s\",\n"
1554                            "\t\t\"old_status\":\"%s\",\n",
1555                    host->hostname,
1556                    ae->unique_id,
1557                    ae->alarm_id,
1558                    ae->alarm_event_id,
1559                    ae->name,
1560                    ae->chart,
1561                    ae->family,
1562                    (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_PROCESSED)?"true":"false",
1563                    (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_UPDATED)?"true":"false",
1564                    (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_EXEC_RUN)?"true":"false",
1565                    (ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_EXEC_FAILED)?"true":"false",
1566                    ae->exec?ae->exec:health_default_exec,
1567                    ae->recipient?ae->recipient:health_default_recipient,
1568                    ae->exec_code,
1569                    ae->source,
1570                    ae->units?ae->units:"",
1571                    ae->info?ae->info:"",
1572                    (unsigned long)ae->when,
1573                    (unsigned long)ae->duration,
1574                    (unsigned long)ae->non_clear_duration,
1575                    rrdcalc_status2string(ae->new_status),
1576                    rrdcalc_status2string(ae->old_status)
1577     );
1578
1579     buffer_strcat(wb, "\t\t\"value\":");
1580     buffer_rrd_value(wb, ae->new_value);
1581     buffer_strcat(wb, ",\n");
1582
1583     buffer_strcat(wb, "\t\t\"old_value\":");
1584     buffer_rrd_value(wb, ae->old_value);
1585     buffer_strcat(wb, "\n");
1586
1587     buffer_strcat(wb, "\t}");
1588 }
1589
1590 void health_alarm_log2json(RRDHOST *host, BUFFER *wb, uint32_t after) {
1591     pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
1592
1593     buffer_strcat(wb, "[");
1594
1595     unsigned int max = host->health_log.max;
1596     unsigned int count = 0;
1597     ALARM_ENTRY *ae;
1598     for(ae = host->health_log.alarms; ae && count < max ; count++, ae = ae->next) {
1599         if(ae->unique_id > after) {
1600             if(likely(count)) buffer_strcat(wb, ",");
1601             health_alarm_entry2json_nolock(wb, ae, host);
1602         }
1603     }
1604
1605     buffer_strcat(wb, "\n]\n");
1606
1607     pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1608 }
1609
1610 static inline void health_rrdcalc2json_nolock(BUFFER *wb, RRDCALC *rc) {
1611     buffer_sprintf(wb,
1612            "\t\t\"%s.%s\": {\n"
1613                    "\t\t\t\"name\": \"%s\",\n"
1614                    "\t\t\t\"chart\": \"%s\",\n"
1615                    "\t\t\t\"family\": \"%s\",\n"
1616                    "\t\t\t\"active\": %s,\n"
1617                    "\t\t\t\"exec\": \"%s\",\n"
1618                    "\t\t\t\"recipient\": \"%s\",\n"
1619                    "\t\t\t\"source\": \"%s\",\n"
1620                    "\t\t\t\"units\": \"%s\",\n"
1621                    "\t\t\t\"info\": \"%s\",\n"
1622                                    "\t\t\t\"status\": \"%s\",\n"
1623                    "\t\t\t\"last_status_change\": %lu,\n"
1624                    "\t\t\t\"last_updated\": %lu,\n"
1625                    "\t\t\t\"next_update\": %lu,\n"
1626                    "\t\t\t\"update_every\": %d,\n"
1627             , rc->chart, rc->name
1628             , rc->name
1629             , rc->chart
1630             , (rc->rrdset && rc->rrdset->family)?rc->rrdset->family:""
1631             , (rc->rrdset)?"true":"false"
1632             , rc->exec?rc->exec:health_default_exec
1633             , rc->recipient?rc->recipient:health_default_recipient
1634             , rc->source
1635             , rc->units?rc->units:""
1636             , rc->info?rc->info:""
1637             , rrdcalc_status2string(rc->status)
1638             , (unsigned long)rc->last_status_change
1639             , (unsigned long)rc->last_updated
1640             , (unsigned long)rc->next_update
1641             , rc->update_every
1642     );
1643
1644     if(RRDCALC_HAS_DB_LOOKUP(rc)) {
1645         if(rc->dimensions && *rc->dimensions)
1646             health_string2json(wb, "\t\t\t", "lookup_dimensions", rc->dimensions, ",\n");
1647
1648         buffer_sprintf(wb,
1649                        "\t\t\t\"db_after\": %lu,\n"
1650                        "\t\t\t\"db_before\": %lu,\n"
1651                        "\t\t\t\"lookup_method\": \"%s\",\n"
1652                        "\t\t\t\"lookup_after\": %d,\n"
1653                        "\t\t\t\"lookup_before\": %d,\n"
1654                        "\t\t\t\"lookup_options\": \"",
1655                        (unsigned long) rc->db_after,
1656                        (unsigned long) rc->db_before,
1657                        group_method2string(rc->group),
1658                        rc->after,
1659                        rc->before
1660         );
1661         buffer_data_options2string(wb, rc->options);
1662         buffer_strcat(wb, "\",\n");
1663     }
1664
1665     if(rc->calculation) {
1666         health_string2json(wb, "\t\t\t", "calc", rc->calculation->source, ",\n");
1667         health_string2json(wb, "\t\t\t", "calc_parsed", rc->calculation->parsed_as, ",\n");
1668     }
1669
1670     if(rc->warning) {
1671         health_string2json(wb, "\t\t\t", "warn", rc->warning->source, ",\n");
1672         health_string2json(wb, "\t\t\t", "warn_parsed", rc->warning->parsed_as, ",\n");
1673     }
1674
1675     if(rc->critical) {
1676         health_string2json(wb, "\t\t\t", "crit", rc->critical->source, ",\n");
1677         health_string2json(wb, "\t\t\t", "crit_parsed", rc->critical->parsed_as, ",\n");
1678     }
1679
1680     buffer_strcat(wb, "\t\t\t\"green\":");
1681     buffer_rrd_value(wb, rc->green);
1682     buffer_strcat(wb, ",\n");
1683
1684     buffer_strcat(wb, "\t\t\t\"red\":");
1685     buffer_rrd_value(wb, rc->red);
1686     buffer_strcat(wb, ",\n");
1687
1688     buffer_strcat(wb, "\t\t\t\"value\":");
1689     buffer_rrd_value(wb, rc->value);
1690     buffer_strcat(wb, "\n");
1691
1692     buffer_strcat(wb, "\t\t}");
1693 }
1694
1695 //void health_rrdcalctemplate2json_nolock(BUFFER *wb, RRDCALCTEMPLATE *rt) {
1696 //
1697 //}
1698
1699 void health_alarms2json(RRDHOST *host, BUFFER *wb, int all) {
1700     int i;
1701     rrdhost_rdlock(&localhost);
1702
1703     buffer_strcat(wb, "{\n\t\"alarms\": {\n");
1704     RRDCALC *rc;
1705     for(i = 0, rc = host->alarms; rc ; rc = rc->next) {
1706         if(unlikely(!rc->rrdset || !rc->rrdset->last_collected_time.tv_sec))
1707             continue;
1708
1709         if(likely(!all && !(rc->status == RRDCALC_STATUS_WARNING || rc->status == RRDCALC_STATUS_CRITICAL)))
1710             continue;
1711
1712         if(likely(i)) buffer_strcat(wb, ",\n");
1713         health_rrdcalc2json_nolock(wb, rc);
1714         i++;
1715     }
1716
1717 //    buffer_strcat(wb, "\n\t},\n\t\"templates\": {");
1718
1719 //    RRDCALCTEMPLATE *rt;
1720 //    for(rt = host->templates; rt ; rt = rt->next)
1721 //        health_rrdcalctemplate2json_nolock(wb, rt);
1722
1723     buffer_sprintf(wb, "\n\t},\n\t\"now\": %lu\n}\n", (unsigned long)time(NULL));
1724     rrdhost_unlock(&localhost);
1725 }
1726
1727
1728 // ----------------------------------------------------------------------------
1729 // re-load health configuration
1730
1731 static inline void health_free_all_nolock(RRDHOST *host) {
1732     while(host->templates)
1733         rrdcalctemplate_free(host, host->templates);
1734
1735     while(host->alarms)
1736         rrdcalc_free(host, host->alarms);
1737 }
1738
1739 void health_reload(void) {
1740     if(!health_enabled) {
1741         error("Health reload is requested, but health is not enabled.");
1742         return;
1743     }
1744
1745     char *path = health_config_dir();
1746
1747     // free all running alarms
1748     rrdhost_rwlock(&localhost);
1749     health_free_all_nolock(&localhost);
1750     rrdhost_unlock(&localhost);
1751
1752     // invalidate all previous entries in the alarm log
1753     ALARM_ENTRY *t;
1754     for(t = localhost.health_log.alarms ; t ; t = t->next) {
1755         t->notifications |= HEALTH_ENTRY_NOTIFICATIONS_UPDATED;
1756     }
1757
1758     // reset all thresholds to all charts
1759     RRDSET *st;
1760     for(st = localhost.rrdset_root; st ; st = st->next) {
1761         st->green = NAN;
1762         st->red = NAN;
1763     }
1764
1765     // load the new alarms
1766     rrdhost_rwlock(&localhost);
1767     health_readdir(path);
1768     rrdhost_unlock(&localhost);
1769
1770     // link the loaded alarms to their charts
1771     for(st = localhost.rrdset_root; st ; st = st->next) {
1772         rrdhost_rwlock(&localhost);
1773
1774         rrdsetcalc_link_matching(st);
1775         rrdcalctemplate_link_matching(st);
1776
1777         rrdhost_unlock(&localhost);
1778     }
1779 }
1780
1781
1782 // ----------------------------------------------------------------------------
1783 // health main thread and friends
1784
1785 static inline int rrdcalc_value2status(calculated_number n) {
1786     if(isnan(n)) return RRDCALC_STATUS_UNDEFINED;
1787     if(n) return RRDCALC_STATUS_RAISED;
1788     return RRDCALC_STATUS_CLEAR;
1789 }
1790
1791 static inline void health_alarm_execute(RRDHOST *host, ALARM_ENTRY *ae) {
1792     if(ae->old_status == RRDCALC_STATUS_UNINITIALIZED && ae->new_status == RRDCALC_STATUS_CLEAR)
1793         return;
1794
1795     char buffer[FILENAME_MAX + 1];
1796     pid_t command_pid;
1797
1798     const char *exec = ae->exec;
1799     if(!exec) exec = health_default_exec;
1800
1801     const char *recipient = ae->recipient;
1802     if(!recipient) recipient = health_default_recipient;
1803
1804     snprintfz(buffer, FILENAME_MAX, "exec %s '%s' '%s' '%u' '%u' '%u' '%zu' '%s' '%s' '%s' '%s' '%s' '%0.0Lf' '%0.0Lf' '%s' '%u' '%u' '%s' '%s'",
1805               exec,
1806               recipient,
1807               host->hostname,
1808               ae->unique_id,
1809               ae->alarm_id,
1810               ae->alarm_event_id,
1811               ae->when,
1812               ae->name,
1813               ae->chart?ae->chart:"NOCAHRT",
1814               ae->family?ae->family:"NOFAMILY",
1815               rrdcalc_status2string(ae->new_status),
1816               rrdcalc_status2string(ae->old_status),
1817               ae->new_value,
1818               ae->old_value,
1819               ae->source?ae->source:"UNKNOWN",
1820               (uint32_t)ae->duration,
1821               (uint32_t)ae->non_clear_duration,
1822               ae->units?ae->units:"",
1823               ae->info?ae->info:""
1824     );
1825
1826     ae->notifications |= HEALTH_ENTRY_NOTIFICATIONS_EXEC_RUN;
1827
1828     debug(D_HEALTH, "executing command '%s'", buffer);
1829     FILE *fp = mypopen(buffer, &command_pid);
1830     if(!fp) {
1831         error("HEALTH: Cannot popen(\"%s\", \"r\").", buffer);
1832         return;
1833     }
1834     debug(D_HEALTH, "HEALTH reading from command");
1835     char *s = fgets(buffer, FILENAME_MAX, fp);
1836     (void)s;
1837     debug(D_HEALTH, "HEALTH closing command");
1838     ae->exec_code = mypclose(fp, command_pid);
1839     debug(D_HEALTH, "done executing command - returned with code %d", ae->exec_code);
1840
1841     if(ae->exec_code != 0)
1842         ae->notifications |= HEALTH_ENTRY_NOTIFICATIONS_EXEC_FAILED;
1843 }
1844
1845 static inline void health_process_notifications(RRDHOST *host, ALARM_ENTRY *ae) {
1846     info("Health alarm '%s.%s' = %0.2Lf - changed status from %s to %s",
1847          ae->chart?ae->chart:"NOCHART", ae->name,
1848          ae->new_value,
1849          rrdcalc_status2string(ae->old_status),
1850          rrdcalc_status2string(ae->new_status)
1851     );
1852
1853     health_alarm_execute(host, ae);
1854 }
1855
1856 static inline void health_alarm_log(RRDHOST *host,
1857                 uint32_t alarm_id, uint32_t alarm_event_id,
1858                 time_t when,
1859                 const char *name, const char *chart, const char *family,
1860                 const char *exec, const char *recipient, time_t duration,
1861                 calculated_number old_value, calculated_number new_value,
1862                 int old_status, int new_status,
1863                 const char *source,
1864                 const char *units,
1865                 const char *info
1866 ) {
1867     ALARM_ENTRY *ae = callocz(1, sizeof(ALARM_ENTRY));
1868     ae->name = strdupz(name);
1869     ae->hash_name = simple_hash(ae->name);
1870
1871     if(chart) {
1872         ae->chart = strdupz(chart);
1873         ae->hash_chart = simple_hash(ae->chart);
1874     }
1875
1876     if(family)
1877         ae->family = strdupz(family);
1878
1879     if(exec) ae->exec = strdupz(exec);
1880     if(recipient) ae->recipient = strdupz(recipient);
1881     if(source) ae->source = strdupz(source);
1882     if(units) ae->units = strdupz(units);
1883     if(info) ae->info = strdupz(info);
1884
1885     ae->unique_id = host->health_log.next_log_id++;
1886     ae->alarm_id = alarm_id;
1887     ae->alarm_event_id = alarm_event_id;
1888     ae->when = when;
1889     ae->old_value = old_value;
1890     ae->new_value = new_value;
1891     ae->old_status = old_status;
1892     ae->new_status = new_status;
1893     ae->duration = duration;
1894
1895     if(ae->old_status == RRDCALC_STATUS_WARNING || ae->old_status == RRDCALC_STATUS_CRITICAL)
1896         ae->non_clear_duration += ae->duration;
1897
1898     // link it
1899     pthread_rwlock_wrlock(&host->health_log.alarm_log_rwlock);
1900     ae->next = host->health_log.alarms;
1901     host->health_log.alarms = ae;
1902     host->health_log.count++;
1903     pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1904
1905     // match previous alarms
1906     pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
1907     ALARM_ENTRY *t;
1908     for(t = host->health_log.alarms ; t ; t = t->next) {
1909         if(t != ae &&
1910                 t->hash_name == ae->hash_name &&
1911                 t->hash_chart == ae->hash_chart &&
1912                 !strcmp(t->name, ae->name) &&
1913                 t->chart && ae->chart && !strcmp(t->chart, ae->chart)) {
1914
1915             if(!(t->notifications & HEALTH_ENTRY_NOTIFICATIONS_UPDATED) && !t->updated_by) {
1916                 t->notifications |= HEALTH_ENTRY_NOTIFICATIONS_UPDATED;
1917                 t->updated_by = ae;
1918
1919                 if((t->new_status == RRDCALC_STATUS_WARNING || t->new_status == RRDCALC_STATUS_CRITICAL) &&
1920                    (t->old_status == RRDCALC_STATUS_WARNING || t->old_status == RRDCALC_STATUS_CRITICAL))
1921                     ae->non_clear_duration += t->non_clear_duration;
1922             }
1923             else {
1924                 // no need to continue
1925                 break;
1926             }
1927         }
1928     }
1929     pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1930 }
1931
1932 static inline void health_alarm_log_process(RRDHOST *host) {
1933     static uint32_t last_processed = 0;
1934     ALARM_ENTRY *ae;
1935
1936     pthread_rwlock_rdlock(&host->health_log.alarm_log_rwlock);
1937
1938     for(ae = host->health_log.alarms; ae ;ae = ae->next) {
1939         if(last_processed >= ae->unique_id) break;
1940
1941         if(!(ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_PROCESSED) &&
1942                 !(ae->notifications & HEALTH_ENTRY_NOTIFICATIONS_UPDATED)) {
1943             ae->notifications |= HEALTH_ENTRY_NOTIFICATIONS_PROCESSED;
1944             health_process_notifications(host, ae);
1945         }
1946     }
1947
1948     if(host->health_log.alarms)
1949         last_processed = host->health_log.alarms->unique_id;
1950
1951     pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1952
1953     if(host->health_log.count <= host->health_log.max)
1954         return;
1955
1956     // cleanup excess entries in the log
1957     pthread_rwlock_wrlock(&host->health_log.alarm_log_rwlock);
1958
1959     ALARM_ENTRY *last = NULL;
1960     unsigned int count = host->health_log.max;
1961     for(ae = host->health_log.alarms; ae && count ; count--, last = ae, ae = ae->next) ;
1962
1963     if(ae && last && last->next == ae)
1964         last->next = NULL;
1965     else
1966         ae = NULL;
1967
1968     while(ae) {
1969         ALARM_ENTRY *t = ae->next;
1970
1971         freez(ae->name);
1972         freez(ae->chart);
1973         freez(ae->family);
1974         freez(ae->exec);
1975         freez(ae->recipient);
1976         freez(ae->source);
1977         freez(ae->units);
1978         freez(ae->info);
1979         freez(ae);
1980
1981         ae = t;
1982     }
1983
1984     pthread_rwlock_unlock(&host->health_log.alarm_log_rwlock);
1985 }
1986
1987 static inline int rrdcalc_isrunnable(RRDCALC *rc, time_t now, time_t *next_run) {
1988     if (unlikely(!rc->rrdset)) {
1989         debug(D_HEALTH, "Health not running alarm '%s.%s'. It is not linked to a chart.", rc->chart?rc->chart:"NOCHART", rc->name);
1990         return 0;
1991     }
1992
1993     if (unlikely(!rc->rrdset->last_collected_time.tv_sec)) {
1994         debug(D_HEALTH, "Health not running alarm '%s.%s'. Chart is not yet collected.", rc->chart?rc->chart:"NOCHART", rc->name);
1995         return 0;
1996     }
1997
1998     if (unlikely(!rc->update_every)) {
1999         debug(D_HEALTH, "Health not running alarm '%s.%s'. It does not have an update frequency", rc->chart?rc->chart:"NOCHART", rc->name);
2000         return 0;
2001     }
2002
2003     if (unlikely(rc->next_update > now)) {
2004         if (unlikely(*next_run > rc->next_update))
2005             *next_run = rc->next_update;
2006
2007         debug(D_HEALTH, "Health not examining alarm '%s.%s' yet (will do in %d secs).", rc->chart?rc->chart:"NOCHART", rc->name, (int) (rc->next_update - now));
2008         return 0;
2009     }
2010
2011     // FIXME
2012     // we should check that the DB lookup is possible
2013     // i.e.
2014     // - the duration of the chart includes the required timeframe
2015     // we SHOULD NOT check the dimensions - there might be alarms that refer non-existing dimensions (e.g. cpu steal)
2016
2017     return 1;
2018 }
2019
2020 void *health_main(void *ptr) {
2021     (void)ptr;
2022
2023     info("HEALTH thread created with task id %d", gettid());
2024
2025     if(pthread_setcanceltype(PTHREAD_CANCEL_DEFERRED, NULL) != 0)
2026         error("Cannot set pthread cancel type to DEFERRED.");
2027
2028     if(pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, NULL) != 0)
2029         error("Cannot set pthread cancel state to ENABLE.");
2030
2031     int min_run_every = (int)config_get_number("health", "run at least every seconds", 10);
2032     if(min_run_every < 1) min_run_every = 1;
2033
2034     BUFFER *wb = buffer_create(100);
2035
2036     unsigned int loop = 0;
2037     while(health_enabled) {
2038         loop++;
2039         debug(D_HEALTH, "Health monitoring iteration no %u started", loop);
2040
2041         int oldstate, runnable = 0;
2042         time_t now = time(NULL);
2043         time_t next_run = now + min_run_every;
2044         RRDCALC *rc;
2045
2046         if (unlikely(pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate) != 0))
2047             error("Cannot set pthread cancel state to DISABLE.");
2048
2049         rrdhost_rdlock(&localhost);
2050
2051         // the first loop is to lookup values from the db
2052         for (rc = localhost.alarms; rc; rc = rc->next) {
2053             if (unlikely(!rrdcalc_isrunnable(rc, now, &next_run)))
2054                 continue;
2055
2056             runnable++;
2057             rc->old_value = rc->value;
2058
2059             // 1. if there is database lookup, do it
2060             // 2. if there is calculation expression, run it
2061
2062             if (unlikely(RRDCALC_HAS_DB_LOOKUP(rc))) {
2063                 time_t old_db_timestamp = rc->db_before;
2064                 int value_is_null = 0;
2065
2066                 int ret = rrd2value(rc->rrdset, wb, &rc->value,
2067                                     rc->dimensions, 1, rc->after, rc->before, rc->group,
2068                                     rc->options, &rc->db_after, &rc->db_before, &value_is_null);
2069
2070                 if (unlikely(ret != 200)) {
2071                     // database lookup failed
2072                     rc->value = NAN;
2073
2074                     debug(D_HEALTH, "Health alarm '%s.%s': database lookup returned error %d", rc->chart?rc->chart:"NOCHART", rc->name, ret);
2075
2076                     if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_ERROR))) {
2077                         rc->rrdcalc_flags |= RRDCALC_FLAG_DB_ERROR;
2078                         error("Health alarm '%s.%s': database lookup returned error %d", rc->chart?rc->chart:"NOCHART", rc->name, ret);
2079                     }
2080                 }
2081                 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_ERROR))
2082                     rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_ERROR;
2083
2084                 if (unlikely(old_db_timestamp == rc->db_before)) {
2085                     // database is stale
2086
2087                     debug(D_HEALTH, "Health alarm '%s.%s': database is stale", rc->chart?rc->chart:"NOCHART", rc->name);
2088
2089                     if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_STALE))) {
2090                         rc->rrdcalc_flags |= RRDCALC_FLAG_DB_STALE;
2091                         error("Health alarm '%s.%s': database is stale", rc->chart?rc->chart:"NOCHART", rc->name);
2092                     }
2093                 }
2094                 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_STALE))
2095                     rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_STALE;
2096
2097                 if (unlikely(value_is_null)) {
2098                     // collected value is null
2099
2100                     rc->value = NAN;
2101
2102                     debug(D_HEALTH, "Health alarm '%s.%s': database lookup returned empty value (possibly value is not collected yet)",
2103                           rc->chart?rc->chart:"NOCHART", rc->name);
2104
2105                     if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_DB_NAN))) {
2106                         rc->rrdcalc_flags |= RRDCALC_FLAG_DB_NAN;
2107                         error("Health alarm '%s.%s': database lookup returned empty value (possibly value is not collected yet)",
2108                               rc->chart?rc->chart:"NOCHART", rc->name);
2109                     }
2110                 }
2111                 else if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_DB_NAN))
2112                     rc->rrdcalc_flags &= ~RRDCALC_FLAG_DB_NAN;
2113
2114                 debug(D_HEALTH, "Health alarm '%s.%s': database lookup gave value "
2115                         CALCULATED_NUMBER_FORMAT, rc->chart?rc->chart:"NOCHART", rc->name, rc->value);
2116             }
2117
2118             if(unlikely(rc->calculation)) {
2119                 if (unlikely(!expression_evaluate(rc->calculation))) {
2120                     // calculation failed
2121
2122                     rc->value = NAN;
2123
2124                     debug(D_HEALTH, "Health alarm '%s.%s': failed to evaluate calculation with error: %s",
2125                           rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->calculation->error_msg));
2126
2127                     if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_CALC_ERROR))) {
2128                         rc->rrdcalc_flags |= RRDCALC_FLAG_CALC_ERROR;
2129                         error("Health alarm '%s.%s': failed to evaluate calculation with error: %s",
2130                               rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->calculation->error_msg));
2131                     }
2132                 }
2133                 else {
2134                     if (unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_CALC_ERROR))
2135                         rc->rrdcalc_flags &= ~RRDCALC_FLAG_CALC_ERROR;
2136
2137                     debug(D_HEALTH, "Health alarm '%s.%s': calculation expression gave value "
2138                             CALCULATED_NUMBER_FORMAT
2139                             ": %s (source: %s)",
2140                           rc->chart?rc->chart:"NOCHART", rc->name,
2141                           rc->calculation->result,
2142                           buffer_tostring(rc->calculation->error_msg),
2143                           rc->source
2144                     );
2145
2146                     rc->value = rc->calculation->result;
2147                 }
2148             }
2149         }
2150         rrdhost_unlock(&localhost);
2151
2152         if (runnable) {
2153             rrdhost_rdlock(&localhost);
2154
2155             for (rc = localhost.alarms; rc; rc = rc->next) {
2156                 if (unlikely(!rrdcalc_isrunnable(rc, now, &next_run)))
2157                     continue;
2158
2159                 int warning_status  = RRDCALC_STATUS_UNDEFINED;
2160                 int critical_status = RRDCALC_STATUS_UNDEFINED;
2161
2162                 if(likely(rc->warning)) {
2163                     if(unlikely(!expression_evaluate(rc->warning))) {
2164                         // calculation failed
2165
2166                         debug(D_HEALTH, "Health alarm '%s.%s': warning expression failed with error: %s",
2167                               rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->warning->error_msg));
2168
2169                         if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_WARN_ERROR))) {
2170                             rc->rrdcalc_flags |= RRDCALC_FLAG_WARN_ERROR;
2171                             error("Health alarm '%s.%s': warning expression failed with error: %s",
2172                                   rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->warning->error_msg));
2173                         }
2174                     }
2175                     else {
2176                         if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_WARN_ERROR))
2177                             rc->rrdcalc_flags &= ~RRDCALC_FLAG_WARN_ERROR;
2178
2179                         debug(D_HEALTH, "Health alarm '%s.%s': warning expression gave value "
2180                                 CALCULATED_NUMBER_FORMAT
2181                                 ": %s (source: %s)",
2182                               rc->chart?rc->chart:"NOCHART", rc->name,
2183                               rc->warning->result,
2184                               buffer_tostring(rc->warning->error_msg),
2185                               rc->source
2186                         );
2187
2188                         warning_status = rrdcalc_value2status(rc->warning->result);
2189                     }
2190                 }
2191
2192                 if(likely(rc->critical)) {
2193                     if(unlikely(!expression_evaluate(rc->critical))) {
2194                         // calculation failed
2195
2196                         debug(D_HEALTH, "Health alarm '%s.%s': critical expression failed with error: %s",
2197                               rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->critical->error_msg));
2198
2199                         if (unlikely(!(rc->rrdcalc_flags & RRDCALC_FLAG_CRIT_ERROR))) {
2200                             rc->rrdcalc_flags |= RRDCALC_FLAG_CRIT_ERROR;
2201                             error("Health alarm '%s.%s': critical expression failed with error: %s",
2202                                   rc->chart?rc->chart:"NOCHART", rc->name, buffer_tostring(rc->critical->error_msg));
2203                         }
2204                     }
2205                     else {
2206                         if(unlikely(rc->rrdcalc_flags & RRDCALC_FLAG_CRIT_ERROR))
2207                             rc->rrdcalc_flags &= ~RRDCALC_FLAG_CRIT_ERROR;
2208
2209                         debug(D_HEALTH, "Health alarm '%s.%s': critical expression gave value "
2210                                 CALCULATED_NUMBER_FORMAT
2211                                 ": %s (source: %s)",
2212                               rc->chart?rc->chart:"NOCHART", rc->name,
2213                               rc->critical->result,
2214                               buffer_tostring(rc->critical->error_msg),
2215                               rc->source
2216                         );
2217
2218                         critical_status = rrdcalc_value2status(rc->critical->result);
2219                     }
2220                 }
2221
2222                 int status = RRDCALC_STATUS_UNDEFINED;
2223
2224                 switch(warning_status) {
2225                     case RRDCALC_STATUS_CLEAR:
2226                         status = RRDCALC_STATUS_CLEAR;
2227                         break;
2228
2229                     case RRDCALC_STATUS_RAISED:
2230                         status = RRDCALC_STATUS_WARNING;
2231                         break;
2232
2233                     default:
2234                         break;
2235                 }
2236
2237                 switch(critical_status) {
2238                     case RRDCALC_STATUS_CLEAR:
2239                         if(status == RRDCALC_STATUS_UNDEFINED)
2240                             status = RRDCALC_STATUS_CLEAR;
2241                         break;
2242
2243                     case RRDCALC_STATUS_RAISED:
2244                         status = RRDCALC_STATUS_CRITICAL;
2245                         break;
2246
2247                     default:
2248                         break;
2249                 }
2250
2251                 if(status != rc->status) {
2252                     health_alarm_log(&localhost, rc->id, rc->next_event_id++, time(NULL), rc->name, rc->rrdset->id, rc->rrdset->family, rc->exec, rc->recipient, now - rc->last_status_change, rc->old_value, rc->value, rc->status, status, rc->source, rc->units, rc->info);
2253                     rc->last_status_change = now;
2254                     rc->status = status;
2255                 }
2256
2257                 rc->last_updated = now;
2258                 rc->next_update = now + rc->update_every;
2259
2260                 if (next_run > rc->next_update)
2261                     next_run = rc->next_update;
2262             }
2263
2264             rrdhost_unlock(&localhost);
2265         }
2266
2267         if (unlikely(pthread_setcancelstate(oldstate, NULL) != 0))
2268             error("Cannot set pthread cancel state to RESTORE (%d).", oldstate);
2269
2270         // execute notifications
2271         // and cleanup
2272         health_alarm_log_process(&localhost);
2273
2274         now = time(NULL);
2275         if(now < next_run) {
2276             debug(D_HEALTH, "Health monitoring iteration no %u done. Next iteration in %d secs",
2277                   loop, (int) (next_run - now));
2278             sleep_usec(1000000 * (unsigned long long) (next_run - now));
2279         }
2280         else {
2281             debug(D_HEALTH, "Health monitoring iteration no %u done. Next iteration now", loop);
2282         }
2283     }
2284
2285     buffer_free(wb);
2286
2287     info("HEALTH thread exiting");
2288     pthread_exit(NULL);
2289     return NULL;
2290 }