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