| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | #include "health.h" |
| 4 | #include "health_internals.h" |
| 5 | |
| 6 | struct variable_lookup_score { |
| 7 | RRDSET *st; |
| 8 | const char *source; |
| 9 | NETDATA_DOUBLE value; |
| 10 | size_t score; |
| 11 | }; |
| 12 | |
| 13 | struct variable_lookup_job { |
| 14 | RRDCALC *rc; |
| 15 | RRDHOST *host; |
| 16 | STRING *variable; |
| 17 | STRING *dim; |
| 18 | const char *dimension; |
| 19 | size_t dimension_length; |
| 20 | enum { |
| 21 | DIM_SELECT_NORMAL, |
| 22 | DIM_SELECT_RAW, |
| 23 | DIM_SELECT_LAST_COLLECTED, |
| 24 | } dimension_selection; |
| 25 | |
| 26 | struct { |
| 27 | size_t size; |
| 28 | size_t used; |
| 29 | struct variable_lookup_score *array; |
| 30 | } result; |
| 31 | |
| 32 | struct { |
| 33 | RRDSET *last_rrdset; |
| 34 | size_t last_score; |
| 35 | } score; |
| 36 | }; |
| 37 | |
| 38 | static void variable_lookup_add_result_with_score(struct variable_lookup_job *vbd, NETDATA_DOUBLE n, RRDSET *st, const char *source __maybe_unused) { |
| 39 | if(vbd->score.last_rrdset != st && vbd->rc->rrdset) { |
| 40 | vbd->score.last_rrdset = st; |
| 41 | vbd->score.last_score = rrdlabels_common_count(vbd->rc->rrdset->rrdlabels, st->rrdlabels); |
| 42 | } |
| 43 | |
| 44 | if(vbd->result.used >= vbd->result.size) { |
| 45 | if(!vbd->result.size) |
| 46 | vbd->result.size = 1; |
| 47 | |
| 48 | vbd->result.size *= 2; |
| 49 | vbd->result.array = reallocz(vbd->result.array, sizeof(struct variable_lookup_score) * vbd->result.size); |
| 50 | } |
| 51 | |
| 52 | vbd->result.array[vbd->result.used++] = (struct variable_lookup_score) { |
| 53 | .value = n, |
| 54 | .score = vbd->score.last_score, |
| 55 | .st = st, |
| 56 | .source = source, |
| 57 | }; |
| 58 | } |
| 59 | |
| 60 | static bool variable_lookup_in_chart(struct variable_lookup_job *vbd, RRDSET *st, bool stop_on_match) { |
| 61 | bool found = false; |
| 62 | const DICTIONARY_ITEM *item = NULL; |
| 63 | RRDDIM *rd = NULL; |
| 64 | dfe_start_read(st->rrddim_root_index, rd) { |
| 65 | if(rd->id == vbd->dim || rd->name == vbd->dim) { |
| 66 | item = dictionary_acquired_item_dup(st->rrddim_root_index, rd_dfe.item); |
| 67 | break; |
| 68 | } |
| 69 | } |
| 70 | dfe_done(rd); |
| 71 | |
| 72 | if (item) { |
| 73 | switch (vbd->dimension_selection) { |
| 74 | case DIM_SELECT_NORMAL: |
| 75 | variable_lookup_add_result_with_score(vbd, (NETDATA_DOUBLE)rd->collector.last_stored_value, st, "last stored value of dimension"); |
| 76 | break; |
| 77 | case DIM_SELECT_RAW: |
| 78 | variable_lookup_add_result_with_score(vbd, rrddim_last_collected_as_double(rd), st, "last collected value of dimension"); |
| 79 | break; |
| 80 | case DIM_SELECT_LAST_COLLECTED: |
| 81 | variable_lookup_add_result_with_score(vbd, (NETDATA_DOUBLE)rd->collector.last_collected_time.tv_sec, st, "last collected time of dimension"); |
| 82 | break; |
| 83 | } |
| 84 | |
| 85 | dictionary_acquired_item_release(st->rrddim_root_index, item); |
| 86 | found = true; |
| 87 | } |
| 88 | if(found && stop_on_match) goto cleanup; |
| 89 | |
| 90 | // chart variable |
| 91 | { |
| 92 | NETDATA_DOUBLE n; |
| 93 | if(rrdvar_get_custom_chart_variable_value(st, vbd->variable, &n)) { |
| 94 | variable_lookup_add_result_with_score(vbd, n, st, "chart variable"); |
| 95 | found = true; |
| 96 | } |
| 97 | } |
| 98 | if(found && stop_on_match) goto cleanup; |
| 99 | |
| 100 | cleanup: |
| 101 | return found; |
| 102 | } |
| 103 | |
| 104 | static int foreach_instance_in_context_cb(RRDSET *st, void *data) { |
| 105 | struct variable_lookup_job *vbd = data; |
| 106 | return variable_lookup_in_chart(vbd, st, false) ? 1 : 0; |
| 107 | } |
| 108 | |
| 109 | static bool variable_lookup_context(struct variable_lookup_job *vbd, const char *chart_or_context, const char *dim_id_or_name) { |
| 110 | struct variable_lookup_job vbd_back = *vbd; |
| 111 | |
| 112 | vbd->dimension = dim_id_or_name; |
| 113 | vbd->dim = string_strdupz(vbd->dimension); |
| 114 | vbd->dimension_length = string_strlen(vbd->dim); |
| 115 | // vbd->dimension_selection = DIM_SELECT_NORMAL; |
| 116 | |
| 117 | bool found = false; |
| 118 | |
| 119 | // lookup chart in host |
| 120 | |
| 121 | RRDSET_ACQUIRED *rsa = rrdset_find_and_acquire(vbd->host, chart_or_context, false); |
| 122 | if(rsa) { |
| 123 | if(variable_lookup_in_chart(vbd, rrdset_acquired_to_rrdset(rsa), false)) |
| 124 | found = true; |
| 125 | rrdset_acquired_release(rsa); |
| 126 | } |
| 127 | |
| 128 | // lookup context in contexts, then foreach chart |
| 129 | |
| 130 | if(rrdcontext_foreach_instance_with_rrdset_in_context(vbd->host, chart_or_context, foreach_instance_in_context_cb, vbd) > 0) |
| 131 | found = true; |
| 132 | |
| 133 | string_freez(vbd->dim); |
| 134 | |
| 135 | vbd->dimension = vbd_back.dimension; |
| 136 | vbd->dim = vbd_back.dim; |
| 137 | vbd->dimension_length = vbd_back.dimension_length; |
| 138 | // vbd->dimension_selection = vbd_back.dimension_selection; |
| 139 | |
| 140 | return found; |
| 141 | } |
| 142 | |
| 143 | bool alert_variable_from_running_alerts(struct variable_lookup_job *vbd) { |
| 144 | bool found = false; |
| 145 | RRDCALC *rc; |
| 146 | foreach_rrdcalc_in_rrdhost_read(vbd->host, rc) { |
| 147 | if(rc->config.name == vbd->variable && rc->rrdset) { |
| 148 | variable_lookup_add_result_with_score(vbd, (NETDATA_DOUBLE)rc->value, rc->rrdset, "alarm value"); |
| 149 | found = true; |
| 150 | } |
| 151 | } |
| 152 | foreach_rrdcalc_in_rrdhost_done(rc); |
| 153 | return found; |
| 154 | } |
| 155 | |
| 156 | static STRING *this_string = NULL, |
| 157 | *now_string = NULL, |
| 158 | *after_string = NULL, |
| 159 | *before_string = NULL, |
| 160 | *status_string = NULL, |
| 161 | *removed_string = NULL, |
| 162 | *uninitialized_string = NULL, |
| 163 | *undefined_string = NULL, |
| 164 | *clear_string = NULL, |
| 165 | *warning_string = NULL, |
| 166 | *critical_string = NULL, |
| 167 | *last_collected_t_string = NULL, |
| 168 | *update_every_string = NULL; |
| 169 | |
| 170 | void alert_variable_lookup_cleanup(void) { |
| 171 | string_freez(this_string); |
| 172 | string_freez(now_string); |
| 173 | string_freez(after_string); |
| 174 | string_freez(before_string); |
| 175 | string_freez(status_string); |
| 176 | string_freez(removed_string); |
| 177 | string_freez(uninitialized_string); |
| 178 | string_freez(undefined_string); |
| 179 | string_freez(clear_string); |
| 180 | string_freez(warning_string); |
| 181 | string_freez(critical_string); |
| 182 | string_freez(last_collected_t_string); |
| 183 | string_freez(update_every_string); |
| 184 | this_string = NULL; |
| 185 | now_string = NULL; |
| 186 | after_string = NULL; |
| 187 | before_string = NULL; |
| 188 | status_string = NULL; |
| 189 | removed_string = NULL; |
| 190 | uninitialized_string = NULL; |
| 191 | undefined_string = NULL; |
| 192 | clear_string = NULL; |
| 193 | warning_string = NULL; |
| 194 | critical_string = NULL; |
| 195 | last_collected_t_string = NULL; |
| 196 | update_every_string = NULL; |
| 197 | } |
| 198 | |
| 199 | bool alert_variable_lookup_internal(STRING *variable, void *data, NETDATA_DOUBLE *result, BUFFER *wb) { |
| 200 | struct variable_lookup_job vbd = { 0 }; |
| 201 | |
| 202 | // const char *v_name = string2str(variable); |
| 203 | // bool trace_this = false; |
| 204 | // if(strcmp(v_name, "btrfs_allocated") == 0) |
| 205 | // trace_this = true; |
| 206 | |
| 207 | bool found = false; |
| 208 | |
| 209 | const char *source = NULL; |
| 210 | RRDSET *source_st = NULL; |
| 211 | |
| 212 | RRDCALC *rc = data; |
| 213 | RRDSET *st = rc->rrdset; |
| 214 | |
| 215 | if(!st) |
| 216 | return false; |
| 217 | |
| 218 | if(unlikely(!last_collected_t_string)) { |
| 219 | this_string = string_strdupz("this"); |
| 220 | now_string = string_strdupz("now"); |
| 221 | after_string = string_strdupz("after"); |
| 222 | before_string = string_strdupz("before"); |
| 223 | status_string = string_strdupz("status"); |
| 224 | removed_string = string_strdupz("REMOVED"); |
| 225 | undefined_string = string_strdupz("UNDEFINED"); |
| 226 | uninitialized_string = string_strdupz("UNINITIALIZED"); |
| 227 | clear_string = string_strdupz("CLEAR"); |
| 228 | warning_string = string_strdupz("WARNING"); |
| 229 | critical_string = string_strdupz("CRITICAL"); |
| 230 | last_collected_t_string = string_strdupz("last_collected_t"); |
| 231 | update_every_string = string_strdupz("update_every"); |
| 232 | } |
| 233 | |
| 234 | if(unlikely(variable == this_string)) { |
| 235 | *result = (NETDATA_DOUBLE)rc->value; |
| 236 | source = "current alert value"; |
| 237 | source_st = st; |
| 238 | found = true; |
| 239 | goto log; |
| 240 | } |
| 241 | |
| 242 | if(unlikely(variable == after_string)) { |
| 243 | *result = (NETDATA_DOUBLE)rc->db_after; |
| 244 | source = "current alert query start time"; |
| 245 | source_st = st; |
| 246 | found = true; |
| 247 | goto log; |
| 248 | } |
| 249 | |
| 250 | if(unlikely(variable == before_string)) { |
| 251 | *result = (NETDATA_DOUBLE)rc->db_before; |
| 252 | source = "current alert query end time"; |
| 253 | source_st = st; |
| 254 | found = true; |
| 255 | goto log; |
| 256 | } |
| 257 | |
| 258 | if(unlikely(variable == now_string)) { |
| 259 | *result = (NETDATA_DOUBLE)now_realtime_sec(); |
| 260 | source = "current wall-time clock timestamp"; |
| 261 | source_st = st; |
| 262 | found = true; |
| 263 | goto log; |
| 264 | } |
| 265 | |
| 266 | if(unlikely(variable == status_string)) { |
| 267 | *result = (NETDATA_DOUBLE)rc->status; |
| 268 | source = "current alert status"; |
| 269 | source_st = st; |
| 270 | found = true; |
| 271 | goto log; |
| 272 | } |
| 273 | |
| 274 | if(unlikely(variable == removed_string)) { |
| 275 | *result = (NETDATA_DOUBLE)RRDCALC_STATUS_REMOVED; |
| 276 | source = "removed status constant"; |
| 277 | source_st = st; |
| 278 | found = true; |
| 279 | goto log; |
| 280 | } |
| 281 | |
| 282 | if(unlikely(variable == uninitialized_string)) { |
| 283 | *result = (NETDATA_DOUBLE)RRDCALC_STATUS_UNINITIALIZED; |
| 284 | source = "uninitialized status constant"; |
| 285 | source_st = st; |
| 286 | found = true; |
| 287 | goto log; |
| 288 | } |
| 289 | |
| 290 | if(unlikely(variable == undefined_string)) { |
| 291 | *result = (NETDATA_DOUBLE)RRDCALC_STATUS_UNDEFINED; |
| 292 | source = "undefined status constant"; |
| 293 | source_st = st; |
| 294 | found = true; |
| 295 | goto log; |
| 296 | } |
| 297 | |
| 298 | if(unlikely(variable == clear_string)) { |
| 299 | *result = (NETDATA_DOUBLE)RRDCALC_STATUS_CLEAR; |
| 300 | source = "clear status constant"; |
| 301 | source_st = st; |
| 302 | found = true; |
| 303 | goto log; |
| 304 | } |
| 305 | |
| 306 | if(unlikely(variable == warning_string)) { |
| 307 | *result = (NETDATA_DOUBLE)RRDCALC_STATUS_WARNING; |
| 308 | source = "warning status constant"; |
| 309 | source_st = st; |
| 310 | found = true; |
| 311 | goto log; |
| 312 | } |
| 313 | |
| 314 | if(unlikely(variable == critical_string)) { |
| 315 | *result = (NETDATA_DOUBLE)RRDCALC_STATUS_CRITICAL; |
| 316 | source = "critical status constant"; |
| 317 | source_st = st; |
| 318 | found = true; |
| 319 | goto log; |
| 320 | } |
| 321 | |
| 322 | if(unlikely(variable == last_collected_t_string)) { |
| 323 | *result = (NETDATA_DOUBLE)st->last_collected_time.tv_sec; |
| 324 | source = "current instance last_collected_t"; |
| 325 | source_st = st; |
| 326 | found = true; |
| 327 | goto log; |
| 328 | } |
| 329 | |
| 330 | if(unlikely(variable == update_every_string)) { |
| 331 | *result = (NETDATA_DOUBLE)st->update_every; |
| 332 | source = "current instance update_every"; |
| 333 | source_st = st; |
| 334 | found = true; |
| 335 | goto log; |
| 336 | } |
| 337 | |
| 338 | // find the dimension id/name |
| 339 | |
| 340 | vbd = (struct variable_lookup_job){ |
| 341 | .rc = rc, |
| 342 | .host = st->rrdhost, |
| 343 | .variable = variable, |
| 344 | .dimension = string2str(variable), |
| 345 | .dimension_length = string_strlen(variable), |
| 346 | .dimension_selection = DIM_SELECT_NORMAL, |
| 347 | .dim = string_dup(variable), |
| 348 | .result = { 0 }, |
| 349 | }; |
| 350 | if (strendswith_lengths(vbd.dimension, vbd.dimension_length, "_raw", 4)) { |
| 351 | vbd.dimension_length -= 4; |
| 352 | vbd.dimension_selection = DIM_SELECT_RAW; |
| 353 | vbd.dim = string_strndupz(vbd.dimension, vbd.dimension_length); |
| 354 | } else if (strendswith_lengths(vbd.dimension, vbd.dimension_length, "_last_collected_t", 17)) { |
| 355 | vbd.dimension_length -= 17; |
| 356 | vbd.dimension_selection = DIM_SELECT_LAST_COLLECTED; |
| 357 | vbd.dim = string_strndupz(vbd.dimension, vbd.dimension_length); |
| 358 | } |
| 359 | |
| 360 | if(variable_lookup_in_chart(&vbd, st, true)) { |
| 361 | found = true; |
| 362 | goto find_best_scored; |
| 363 | } |
| 364 | |
| 365 | // host variables |
| 366 | { |
| 367 | NETDATA_DOUBLE n; |
| 368 | found = rrdvar_get_custom_host_variable_value(vbd.host, vbd.variable, &n); |
| 369 | if(found) { |
| 370 | variable_lookup_add_result_with_score(&vbd, n, st, "host variable"); |
| 371 | goto find_best_scored; |
| 372 | } |
| 373 | } |
| 374 | |
| 375 | // alert names |
| 376 | if(alert_variable_from_running_alerts(&vbd)) { |
| 377 | found = true; |
| 378 | goto find_best_scored; |
| 379 | } |
| 380 | |
| 381 | // find the components of the variable |
| 382 | { |
| 383 | char *id = strdupz(string2str(vbd.dim)); |
| 384 | |
| 385 | char *dot = strrchr(id, '.'); |
| 386 | while(dot) { |
| 387 | *dot = '\0'; |
| 388 | |
| 389 | if(strchr(id, '.') == NULL) break; |
| 390 | |
| 391 | if(variable_lookup_context(&vbd, id, dot + 1)) |
| 392 | found = true; |
| 393 | |
| 394 | char *dot2 = strrchr(id, '.'); |
| 395 | *dot = '.'; |
| 396 | dot = dot2; |
| 397 | } |
| 398 | |
| 399 | freez(id); |
| 400 | } |
| 401 | |
| 402 | find_best_scored: |
| 403 | if(found && vbd.result.array) { |
| 404 | struct variable_lookup_score *best = &vbd.result.array[0]; |
| 405 | for (size_t i = 1; i < vbd.result.used; i++) |
| 406 | if (vbd.result.array[i].score > best->score) |
| 407 | best = &vbd.result.array[i]; |
| 408 | |
| 409 | source = best->source; |
| 410 | source_st = best->st; |
| 411 | *result = best->value; |
| 412 | freez(vbd.result.array); |
| 413 | } |
| 414 | else { |
| 415 | found = false; |
| 416 | *result = NAN; |
| 417 | } |
| 418 | |
| 419 | log: |
| 420 | #ifdef NETDATA_LOG_HEALTH_VARIABLES_LOOKUP |
| 421 | if(found) { |
| 422 | nd_log(NDLS_DAEMON, NDLP_INFO, |
| 423 | "HEALTH_VARIABLE_LOOKUP: variable '%s' of alert '%s' of chart '%s', context '%s', host '%s' " |
| 424 | "resolved with %s of chart '%s' and context '%s'", |
| 425 | string2str(variable), |
| 426 | string2str(rc->config.name), |
| 427 | string2str(rc->rrdset->id), |
| 428 | string2str(rc->rrdset->context), |
| 429 | string2str(rc->rrdset->rrdhost->hostname), |
| 430 | source, |
| 431 | string2str(source_st->id), |
| 432 | string2str(source_st->context) |
| 433 | ); |
| 434 | } |
| 435 | else { |
| 436 | nd_log(NDLS_DAEMON, NDLP_INFO, |
| 437 | "HEALTH_VARIABLE_LOOKUP: variable '%s' of alert '%s' of chart '%s', context '%s', host '%s' " |
| 438 | "could not be resolved", |
| 439 | string2str(variable), |
| 440 | string2str(rc->config.name), |
| 441 | string2str(rc->rrdset->id), |
| 442 | string2str(rc->rrdset->context), |
| 443 | string2str(rc->rrdset->rrdhost->hostname) |
| 444 | ); |
| 445 | } |
| 446 | #endif |
| 447 | |
| 448 | if(unlikely(wb)) { |
| 449 | buffer_json_member_add_string(wb, "variable", string2str(variable)); |
| 450 | buffer_json_member_add_string(wb, "instance", string2str(st->id)); |
| 451 | buffer_json_member_add_string(wb, "context", string2str(st->context)); |
| 452 | buffer_json_member_add_boolean(wb, "found", found); |
| 453 | |
| 454 | if (found) { |
| 455 | buffer_json_member_add_double(wb, "value", *result); |
| 456 | buffer_json_member_add_object(wb, "source"); |
| 457 | { |
| 458 | buffer_json_member_add_string(wb, "description", source); |
| 459 | buffer_json_member_add_string(wb, "instance", string2str(source_st->id)); |
| 460 | buffer_json_member_add_string(wb, "context", string2str(source_st->context)); |
| 461 | buffer_json_member_add_uint64(wb, "candidates", vbd.result.used ? vbd.result.used : 1); |
| 462 | } |
| 463 | buffer_json_object_close(wb); // source |
| 464 | } |
| 465 | } |
| 466 | |
| 467 | string_freez(vbd.dim); |
| 468 | |
| 469 | return found; |
| 470 | } |
| 471 | |
| 472 | bool alert_variable_lookup(STRING *variable, void *data, NETDATA_DOUBLE *result) { |
| 473 | return alert_variable_lookup_internal(variable, data, result, NULL); |
| 474 | } |
| 475 | |
| 476 | int alert_variable_lookup_trace(RRDHOST *host __maybe_unused, RRDSET *st, const char *variable, BUFFER *wb) { |
| 477 | int code = HTTP_RESP_INTERNAL_SERVER_ERROR; |
| 478 | |
| 479 | buffer_flush(wb); |
| 480 | buffer_json_initialize(wb, "\"", "\"", 0, true, BUFFER_JSON_OPTIONS_DEFAULT); |
| 481 | |
| 482 | STRING *v = string_strdupz(variable); |
| 483 | RRDCALC rc = { |
| 484 | .rrdset = st, |
| 485 | }; |
| 486 | |
| 487 | NETDATA_DOUBLE n; |
| 488 | alert_variable_lookup_internal(v, &rc, &n, wb); |
| 489 | |
| 490 | string_freez(v); |
| 491 | |
| 492 | buffer_json_finalize(wb); |
| 493 | return code; |
| 494 | } |