| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | #include "rrdcontext-internal.h" |
| 4 | |
| 5 | static struct { |
| 6 | bool enabled; |
| 7 | size_t db_rotations; // count of actual dbengine rotations only |
| 8 | size_t instances_count; |
| 9 | size_t active_vs_archived_percentage; |
| 10 | } extreme_cardinality = { |
| 11 | .enabled = true, // this value is ignored - there is a dynamic condition to enable it |
| 12 | .db_rotations = 0, |
| 13 | .instances_count = 1000, |
| 14 | .active_vs_archived_percentage = 50, |
| 15 | }; |
| 16 | |
| 17 | void rrdcontext_count_db_rotation(void) { |
| 18 | __atomic_add_fetch(&extreme_cardinality.db_rotations, 1, __ATOMIC_RELAXED); |
| 19 | } |
| 20 | |
| 21 | static uint64_t rrdcontext_get_next_version(RRDCONTEXT *rc); |
| 22 | |
| 23 | static void rrdcontext_garbage_collect_for_all_hosts(void); |
| 24 | |
| 25 | extern usec_t rrdcontext_next_db_rotation_ut; |
| 26 | extern size_t rrdcontext_full_gc_rerun_requested; |
| 27 | |
| 28 | // ---------------------------------------------------------------------------- |
| 29 | // version hash calculation |
| 30 | |
| 31 | uint64_t rrdcontext_version_hash_with_callback( |
| 32 | RRDHOST *host, |
| 33 | void (*callback)(RRDCONTEXT *, bool, void *), |
| 34 | bool snapshot, |
| 35 | void *bundle) { |
| 36 | |
| 37 | if(unlikely(!host || !host->rrdctx.contexts)) return 0; |
| 38 | |
| 39 | RRDCONTEXT *rc; |
| 40 | uint64_t hash = 0; |
| 41 | |
| 42 | // loop through all contexts of the host |
| 43 | dfe_start_read(host->rrdctx.contexts, rc) { |
| 44 | |
| 45 | rrdcontext_lock(rc); |
| 46 | |
| 47 | if(unlikely(rrd_flag_check(rc, RRD_FLAG_HIDDEN))) { |
| 48 | rrdcontext_unlock(rc); |
| 49 | continue; |
| 50 | } |
| 51 | |
| 52 | if(unlikely(callback)) |
| 53 | callback(rc, snapshot, bundle); |
| 54 | |
| 55 | // skip any deleted contexts |
| 56 | if(unlikely(rrd_flag_is_deleted(rc))) { |
| 57 | rrdcontext_unlock(rc); |
| 58 | continue; |
| 59 | } |
| 60 | |
| 61 | // we use rc->hub.* which has the latest |
| 62 | // metadata we have sent to the hub |
| 63 | |
| 64 | // if a context is currently queued, rc->hub.* does NOT |
| 65 | // reflect the queued changes. rc->hub.* is updated with |
| 66 | // their metadata, after messages are dispatched to hub. |
| 67 | |
| 68 | // when the context is being collected, |
| 69 | // rc->hub.last_time_t is already zero |
| 70 | |
| 71 | hash += rc->hub.version + rc->hub.last_time_s - rc->hub.first_time_s; |
| 72 | |
| 73 | rrdcontext_unlock(rc); |
| 74 | |
| 75 | } |
| 76 | dfe_done(rc); |
| 77 | |
| 78 | return hash; |
| 79 | } |
| 80 | |
| 81 | // ---------------------------------------------------------------------------- |
| 82 | // retention recalculation |
| 83 | |
| 84 | static void rrdhost_update_cached_retention(RRDHOST *host, time_t first_time_s, time_t last_time_s, bool global) { |
| 85 | if(unlikely(!host)) |
| 86 | return; |
| 87 | |
| 88 | spinlock_lock(&host->retention.spinlock); |
| 89 | |
| 90 | time_t old_first_time_s = host->retention.first_time_s; |
| 91 | |
| 92 | if(global) { |
| 93 | host->retention.first_time_s = first_time_s; |
| 94 | host->retention.last_time_s = last_time_s; |
| 95 | } |
| 96 | else { |
| 97 | if(!host->retention.first_time_s || (first_time_s && first_time_s < host->retention.first_time_s)) |
| 98 | host->retention.first_time_s = first_time_s; |
| 99 | |
| 100 | if(!host->retention.last_time_s || last_time_s > host->retention.last_time_s) |
| 101 | host->retention.last_time_s = last_time_s; |
| 102 | } |
| 103 | |
| 104 | bool stream_path_update_required = old_first_time_s != host->retention.first_time_s; |
| 105 | |
| 106 | spinlock_unlock(&host->retention.spinlock); |
| 107 | |
| 108 | if(stream_path_update_required) |
| 109 | stream_path_retention_updated(host); |
| 110 | } |
| 111 | |
| 112 | void rrdcontext_recalculate_context_retention(RRDCONTEXT *rc, RRD_FLAGS reason, bool worker_jobs) { |
| 113 | bool forcefully_removed_instances = false; |
| 114 | do { |
| 115 | forcefully_removed_instances = rrdcontext_post_process_updates(rc, true, reason, worker_jobs); |
| 116 | } while(forcefully_removed_instances); |
| 117 | } |
| 118 | |
| 119 | void rrdcontext_recalculate_host_retention(RRDHOST *host, RRD_FLAGS reason, bool worker_jobs) { |
| 120 | if(unlikely(!host || !host->rrdctx.contexts)) return; |
| 121 | |
| 122 | time_t first_time_s = 0; |
| 123 | time_t last_time_s = 0; |
| 124 | |
| 125 | RRDCONTEXT *rc; |
| 126 | dfe_start_read(host->rrdctx.contexts, rc) { |
| 127 | rrdcontext_recalculate_context_retention(rc, reason, worker_jobs); |
| 128 | |
| 129 | if(!first_time_s || (rc->first_time_s && rc->first_time_s < first_time_s)) |
| 130 | first_time_s = rc->first_time_s; |
| 131 | |
| 132 | if(!last_time_s || rc->last_time_s > last_time_s) |
| 133 | last_time_s = rc->last_time_s; |
| 134 | } |
| 135 | dfe_done(rc); |
| 136 | |
| 137 | rrdhost_update_cached_retention(host, first_time_s, last_time_s, true); |
| 138 | } |
| 139 | |
| 140 | static void rrdcontext_recalculate_retention_all_hosts(void) { |
| 141 | // Don't pre-clear rrdcontext_next_db_rotation_ut here -- the caller in |
| 142 | // rrdcontext_main clears it via CAS at the end of the pass, expecting |
| 143 | // the same deadline value it observed. Pre-clearing would let a |
| 144 | // concurrent rrdcontext_request_full_gc() arm a new deadline that the |
| 145 | // caller's unconditional store-to-zero would then silently overwrite, |
| 146 | // dropping the request. |
| 147 | RRDHOST *host; |
| 148 | dfe_start_reentrant(rrdhost_root_index, host) { |
| 149 | worker_is_busy(WORKER_JOB_RETENTION); |
| 150 | rrdcontext_recalculate_host_retention(host, RRD_FLAG_UPDATE_REASON_DB_ROTATION, true); |
| 151 | } |
| 152 | dfe_done(host); |
| 153 | } |
| 154 | |
| 155 | // ---------------------------------------------------------------------------- |
| 156 | // garbage collector |
| 157 | |
| 158 | void get_metric_retention_by_id(RRDHOST *host, UUIDMAP_ID id, time_t *min_first_time_t, time_t *max_last_time_t, bool *tier0_retention) { |
| 159 | *min_first_time_t = LONG_MAX; |
| 160 | *max_last_time_t = 0; |
| 161 | |
| 162 | for (size_t tier = 0; tier < nd_profile.storage_tiers; tier++) { |
| 163 | STORAGE_ENGINE *eng = host->db[tier].eng; |
| 164 | |
| 165 | time_t first_time_t = 0, last_time_t = 0; |
| 166 | if (eng->api.metric_retention_by_id(host->db[tier].si, id, &first_time_t, &last_time_t)) { |
| 167 | if (first_time_t > 0 && first_time_t < *min_first_time_t) |
| 168 | *min_first_time_t = first_time_t; |
| 169 | |
| 170 | if (last_time_t > *max_last_time_t) |
| 171 | *max_last_time_t = last_time_t; |
| 172 | } |
| 173 | |
| 174 | if(tier == 0 && tier0_retention) |
| 175 | *tier0_retention = first_time_t || last_time_t; |
| 176 | } |
| 177 | } |
| 178 | |
| 179 | bool rrdmetric_update_retention(RRDMETRIC *rm) { |
| 180 | time_t min_first_time_t = LONG_MAX, max_last_time_t = 0; |
| 181 | RRDDIM *rd = rrdmetric_rrddim_get_and_lock(rm); |
| 182 | |
| 183 | if(rd) { |
| 184 | min_first_time_t = rrddim_first_entry_s(rd); |
| 185 | max_last_time_t = rrddim_last_entry_s(rd); |
| 186 | rrdmetric_rrddim_unlock(rd); |
| 187 | rrd_flag_clear(rm, RRD_FLAG_NO_TIER0_RETENTION); |
| 188 | } |
| 189 | else { |
| 190 | bool tier0_retention; |
| 191 | get_metric_retention_by_id(rm->ri->rc->rrdhost, rm->uuid, &min_first_time_t, &max_last_time_t, &tier0_retention); |
| 192 | |
| 193 | if(tier0_retention) |
| 194 | rrd_flag_clear(rm, RRD_FLAG_NO_TIER0_RETENTION); |
| 195 | else |
| 196 | rrd_flag_set(rm, RRD_FLAG_NO_TIER0_RETENTION); |
| 197 | } |
| 198 | |
| 199 | if(min_first_time_t == LONG_MAX) |
| 200 | min_first_time_t = 0; |
| 201 | |
| 202 | if(min_first_time_t > max_last_time_t) { |
| 203 | internal_error(true, "RRDMETRIC: retention of '%s' is flipped, first_time_t = %ld, last_time_t = %ld", string2str(rm->id), min_first_time_t, max_last_time_t); |
| 204 | SWAP(min_first_time_t, max_last_time_t); |
| 205 | } |
| 206 | |
| 207 | // check if retention changed |
| 208 | |
| 209 | if (min_first_time_t != rm->first_time_s) { |
| 210 | rm->first_time_s = min_first_time_t; |
| 211 | rrd_flag_set_updated(rm, RRD_FLAG_UPDATE_REASON_CHANGED_FIRST_TIME_T); |
| 212 | } |
| 213 | |
| 214 | if (max_last_time_t != rm->last_time_s) { |
| 215 | rm->last_time_s = max_last_time_t; |
| 216 | rrd_flag_set_updated(rm, RRD_FLAG_UPDATE_REASON_CHANGED_LAST_TIME_T); |
| 217 | } |
| 218 | |
| 219 | if(unlikely(!rm->first_time_s && !rm->last_time_s)) |
| 220 | rrdmetric_set_deleted(rm, RRD_FLAG_UPDATE_REASON_ZERO_RETENTION); |
| 221 | |
| 222 | rrd_flag_set(rm, RRD_FLAG_LIVE_RETENTION); |
| 223 | |
| 224 | return true; |
| 225 | } |
| 226 | |
| 227 | static inline bool rrdmetric_should_be_deleted(RRDMETRIC *rm) { |
| 228 | if(likely(!rrd_flag_check(rm, RRD_FLAGS_REQUIRED_FOR_DELETIONS))) |
| 229 | return false; |
| 230 | |
| 231 | if(likely(rrd_flag_check(rm, RRD_FLAGS_PREVENTING_DELETIONS))) |
| 232 | return false; |
| 233 | |
| 234 | if(likely(rrdmetric_rrddim_atomic_load(rm))) |
| 235 | return false; |
| 236 | |
| 237 | rrdmetric_update_retention(rm); |
| 238 | if(rm->first_time_s || rm->last_time_s) |
| 239 | return false; |
| 240 | |
| 241 | return true; |
| 242 | } |
| 243 | |
| 244 | static inline bool rrdinstance_should_be_deleted(RRDINSTANCE *ri) { |
| 245 | if(likely(!rrd_flag_check(ri, RRD_FLAGS_REQUIRED_FOR_DELETIONS))) |
| 246 | return false; |
| 247 | |
| 248 | if(likely(rrd_flag_check(ri, RRD_FLAGS_PREVENTING_DELETIONS))) |
| 249 | return false; |
| 250 | |
| 251 | if(likely(ri->rrdset)) |
| 252 | return false; |
| 253 | |
| 254 | if(unlikely(dictionary_referenced_items(ri->rrdmetrics) != 0)) |
| 255 | return false; |
| 256 | |
| 257 | if(unlikely(dictionary_entries(ri->rrdmetrics) != 0)) |
| 258 | return false; |
| 259 | |
| 260 | if(ri->first_time_s || ri->last_time_s) |
| 261 | return false; |
| 262 | |
| 263 | return true; |
| 264 | } |
| 265 | |
| 266 | bool rrdcontext_should_be_deleted(RRDCONTEXT *rc) { |
| 267 | if(likely(!rrd_flag_check(rc, RRD_FLAGS_REQUIRED_FOR_DELETIONS))) |
| 268 | return false; |
| 269 | |
| 270 | if(likely(rrd_flag_check(rc, RRD_FLAGS_PREVENTING_DELETIONS))) |
| 271 | return false; |
| 272 | |
| 273 | if(unlikely(dictionary_referenced_items(rc->rrdinstances) != 0)) |
| 274 | return false; |
| 275 | |
| 276 | if(unlikely(dictionary_entries(rc->rrdinstances) != 0)) |
| 277 | return false; |
| 278 | |
| 279 | if(unlikely(rc->first_time_s || rc->last_time_s)) |
| 280 | return false; |
| 281 | |
| 282 | return true; |
| 283 | } |
| 284 | |
| 285 | void rrdcontext_delete_from_sql_unsafe(RRDCONTEXT *rc) { |
| 286 | // we need to refresh the string pointers in rc->hub |
| 287 | // in case the context changed values |
| 288 | rc->hub.id = string2str(rc->id); |
| 289 | rc->hub.title = string2str(rc->title); |
| 290 | rc->hub.units = string2str(rc->units); |
| 291 | rc->hub.family = string2str(rc->family); |
| 292 | |
| 293 | if (rc->rrdhost->rrd_memory_mode != RRD_DB_MODE_DBENGINE) |
| 294 | return; |
| 295 | |
| 296 | // delete it from SQL |
| 297 | if(ctx_delete_context(&rc->rrdhost->host_id.uuid, &rc->hub) != 0) |
| 298 | netdata_log_error("RRDCONTEXT: failed to delete context '%s' version %"PRIu64" from SQL.", |
| 299 | rc->hub.id, rc->hub.version); |
| 300 | } |
| 301 | |
| 302 | void rrdcontext_garbage_collect_single_host(RRDHOST *host, bool worker_jobs) { |
| 303 | |
| 304 | internal_error(true, "RRDCONTEXT: garbage collecting context structures of host '%s'", rrdhost_hostname(host)); |
| 305 | |
| 306 | RRDCONTEXT *rc; |
| 307 | dfe_start_reentrant(host->rrdctx.contexts, rc) { |
| 308 | if(unlikely(worker_jobs && !service_running(SERVICE_CONTEXT))) break; |
| 309 | |
| 310 | if(worker_jobs) worker_is_busy(WORKER_JOB_CLEANUP); |
| 311 | |
| 312 | rrdcontext_lock(rc); |
| 313 | |
| 314 | RRDINSTANCE *ri; |
| 315 | dfe_start_reentrant(rc->rrdinstances, ri) { |
| 316 | if(unlikely(worker_jobs && !service_running(SERVICE_CONTEXT))) break; |
| 317 | |
| 318 | RRDMETRIC *rm; |
| 319 | dfe_start_write(ri->rrdmetrics, rm) { |
| 320 | if(rrdmetric_should_be_deleted(rm)) { |
| 321 | if(worker_jobs) worker_is_busy(WORKER_JOB_CLEANUP_DELETE); |
| 322 | if(!dictionary_del(ri->rrdmetrics, string2str(rm->id))) |
| 323 | netdata_log_error("RRDCONTEXT: metric '%s' of instance '%s' of context '%s' of host '%s', failed to be deleted from rrdmetrics dictionary.", |
| 324 | string2str(rm->id), |
| 325 | string2str(ri->id), |
| 326 | string2str(rc->id), |
| 327 | rrdhost_hostname(host)); |
| 328 | else |
| 329 | internal_error( |
| 330 | true, |
| 331 | "RRDCONTEXT: metric '%s' of instance '%s' of context '%s' of host '%s', deleted from rrdmetrics dictionary.", |
| 332 | string2str(rm->id), |
| 333 | string2str(ri->id), |
| 334 | string2str(rc->id), |
| 335 | rrdhost_hostname(host)); |
| 336 | } |
| 337 | } |
| 338 | dfe_done(rm); |
| 339 | |
| 340 | if(rrdinstance_should_be_deleted(ri)) { |
| 341 | if(worker_jobs) worker_is_busy(WORKER_JOB_CLEANUP_DELETE); |
| 342 | if(!dictionary_del(rc->rrdinstances, string2str(ri->id))) |
| 343 | netdata_log_error("RRDCONTEXT: instance '%s' of context '%s' of host '%s', failed to be deleted from rrdmetrics dictionary.", |
| 344 | string2str(ri->id), |
| 345 | string2str(rc->id), |
| 346 | rrdhost_hostname(host)); |
| 347 | else |
| 348 | internal_error( |
| 349 | true, |
| 350 | "RRDCONTEXT: instance '%s' of context '%s' of host '%s', deleted from rrdmetrics dictionary.", |
| 351 | string2str(ri->id), |
| 352 | string2str(rc->id), |
| 353 | rrdhost_hostname(host)); |
| 354 | } |
| 355 | |
| 356 | dictionary_garbage_collect(ri->rrdmetrics); |
| 357 | } |
| 358 | dfe_done(ri); |
| 359 | dictionary_garbage_collect(rc->rrdinstances); |
| 360 | |
| 361 | if(unlikely(rrdcontext_should_be_deleted(rc))) { |
| 362 | if(worker_jobs) worker_is_busy(WORKER_JOB_CLEANUP_DELETE); |
| 363 | rrdcontext_dequeue_from_post_processing(rc); |
| 364 | rrdcontext_delete_from_sql_unsafe(rc); |
| 365 | |
| 366 | if(!dictionary_del(host->rrdctx.contexts, string2str(rc->id))) |
| 367 | netdata_log_error("RRDCONTEXT: context '%s' of host '%s', failed to be deleted from rrdmetrics dictionary.", |
| 368 | string2str(rc->id), |
| 369 | rrdhost_hostname(host)); |
| 370 | else |
| 371 | internal_error( |
| 372 | true, |
| 373 | "RRDCONTEXT: context '%s' of host '%s', deleted from rrdmetrics dictionary.", |
| 374 | string2str(rc->id), |
| 375 | rrdhost_hostname(host)); |
| 376 | } |
| 377 | |
| 378 | // the item is referenced in the dictionary |
| 379 | // so, it is still here to unlock, even if we have deleted it |
| 380 | rrdcontext_unlock(rc); |
| 381 | } |
| 382 | dfe_done(rc); |
| 383 | |
| 384 | dictionary_garbage_collect(host->rrdctx.contexts); |
| 385 | } |
| 386 | |
| 387 | static void rrdcontext_garbage_collect_for_all_hosts(void) { |
| 388 | RRDHOST *host; |
| 389 | dfe_start_reentrant(rrdhost_root_index, host) { |
| 390 | rrdcontext_garbage_collect_single_host(host, true); |
| 391 | } |
| 392 | dfe_done(host); |
| 393 | } |
| 394 | |
| 395 | // ---------------------------------------------------------------------------- |
| 396 | // post processing |
| 397 | |
| 398 | static void rrdmetric_process_updates(RRDMETRIC *rm, bool force, RRD_FLAGS reason, bool worker_jobs) { |
| 399 | if(reason != RRD_FLAG_NONE) |
| 400 | rrd_flag_set_updated(rm, reason); |
| 401 | |
| 402 | if(!force && !rrd_flag_is_updated(rm) && rrd_flag_check(rm, RRD_FLAG_LIVE_RETENTION) && !rrd_flag_check(rm, RRD_FLAG_UPDATE_REASON_UPDATE_RETENTION)) |
| 403 | return; |
| 404 | |
| 405 | if(worker_jobs) |
| 406 | worker_is_busy(WORKER_JOB_PP_METRIC); |
| 407 | |
| 408 | if(reason & RRD_FLAG_UPDATE_REASON_DISCONNECTED_CHILD) { |
| 409 | rrdmetric_set_archived(rm); |
| 410 | rrd_flag_set(rm, RRD_FLAG_UPDATE_REASON_DISCONNECTED_CHILD); |
| 411 | } |
| 412 | if(rrd_flag_is_deleted(rm) && (reason & RRD_FLAG_UPDATE_REASON_UPDATE_RETENTION)) |
| 413 | rrdmetric_set_archived(rm); |
| 414 | |
| 415 | rrdmetric_update_retention(rm); |
| 416 | |
| 417 | rrd_flag_unset_updated(rm); |
| 418 | } |
| 419 | |
| 420 | static void rrdinstance_post_process_updates(RRDINSTANCE *ri, bool force, RRD_FLAGS reason, bool worker_jobs) { |
| 421 | if(reason != RRD_FLAG_NONE) |
| 422 | rrd_flag_set_updated(ri, reason); |
| 423 | |
| 424 | if(!force && !rrd_flag_is_updated(ri) && rrd_flag_check(ri, RRD_FLAG_LIVE_RETENTION)) |
| 425 | return; |
| 426 | |
| 427 | if(worker_jobs) |
| 428 | worker_is_busy(WORKER_JOB_PP_INSTANCE); |
| 429 | |
| 430 | time_t min_first_time_t = LONG_MAX, max_last_time_t = 0; |
| 431 | size_t metrics_active = 0, metrics_deleted = 0, metrics_no_tier0 = 0; |
| 432 | bool live_retention = true, currently_collected = false; |
| 433 | if(dictionary_entries(ri->rrdmetrics) > 0) { |
| 434 | RRDMETRIC *rm; |
| 435 | dfe_start_read((DICTIONARY *)ri->rrdmetrics, rm) { |
| 436 | if(unlikely(worker_jobs && !service_running(SERVICE_CONTEXT))) break; |
| 437 | |
| 438 | RRD_FLAGS reason_to_pass = reason; |
| 439 | if(rrd_flag_check(ri, RRD_FLAG_UPDATE_REASON_UPDATE_RETENTION)) |
| 440 | reason_to_pass |= RRD_FLAG_UPDATE_REASON_UPDATE_RETENTION; |
| 441 | |
| 442 | rrdmetric_process_updates(rm, force, reason_to_pass, worker_jobs); |
| 443 | |
| 444 | if(unlikely(!rrd_flag_check(rm, RRD_FLAG_LIVE_RETENTION))) |
| 445 | live_retention = false; |
| 446 | |
| 447 | if(unlikely(rrd_flag_check(rm, RRD_FLAG_NO_TIER0_RETENTION))) |
| 448 | metrics_no_tier0++; |
| 449 | |
| 450 | if (unlikely((rrdmetric_should_be_deleted(rm)))) { |
| 451 | metrics_deleted++; |
| 452 | continue; |
| 453 | } |
| 454 | |
| 455 | if(!currently_collected && rrd_flag_is_collected(rm) && rm->first_time_s) |
| 456 | currently_collected = true; |
| 457 | |
| 458 | metrics_active++; |
| 459 | |
| 460 | if (rm->first_time_s && rm->first_time_s < min_first_time_t) |
| 461 | min_first_time_t = rm->first_time_s; |
| 462 | |
| 463 | if (rm->last_time_s && rm->last_time_s > max_last_time_t) |
| 464 | max_last_time_t = rm->last_time_s; |
| 465 | } |
| 466 | dfe_done(rm); |
| 467 | } |
| 468 | |
| 469 | if(metrics_no_tier0 && metrics_no_tier0 == metrics_active) |
| 470 | rrd_flag_set(ri, RRD_FLAG_NO_TIER0_RETENTION); |
| 471 | else |
| 472 | rrd_flag_clear(ri, RRD_FLAG_NO_TIER0_RETENTION); |
| 473 | |
| 474 | if(unlikely(live_retention && !rrd_flag_check(ri, RRD_FLAG_LIVE_RETENTION))) |
| 475 | rrd_flag_set(ri, RRD_FLAG_LIVE_RETENTION); |
| 476 | else if(unlikely(!live_retention && rrd_flag_check(ri, RRD_FLAG_LIVE_RETENTION))) |
| 477 | rrd_flag_clear(ri, RRD_FLAG_LIVE_RETENTION); |
| 478 | |
| 479 | if(unlikely(!metrics_active)) { |
| 480 | // no metrics available |
| 481 | |
| 482 | if(ri->first_time_s) { |
| 483 | ri->first_time_s = 0; |
| 484 | rrd_flag_set_updated(ri, RRD_FLAG_UPDATE_REASON_CHANGED_FIRST_TIME_T); |
| 485 | } |
| 486 | |
| 487 | if(ri->last_time_s) { |
| 488 | ri->last_time_s = 0; |
| 489 | rrd_flag_set_updated(ri, RRD_FLAG_UPDATE_REASON_CHANGED_LAST_TIME_T); |
| 490 | } |
| 491 | |
| 492 | rrdinstance_set_deleted(ri, RRD_FLAG_UPDATE_REASON_ZERO_RETENTION); |
| 493 | } |
| 494 | else { |
| 495 | // we have active metrics... |
| 496 | |
| 497 | if (unlikely(min_first_time_t == LONG_MAX)) |
| 498 | min_first_time_t = 0; |
| 499 | |
| 500 | if (unlikely(min_first_time_t == 0 || max_last_time_t == 0)) { |
| 501 | if(ri->first_time_s) { |
| 502 | ri->first_time_s = 0; |
| 503 | rrd_flag_set_updated(ri, RRD_FLAG_UPDATE_REASON_CHANGED_FIRST_TIME_T); |
| 504 | } |
| 505 | |
| 506 | if(ri->last_time_s) { |
| 507 | ri->last_time_s = 0; |
| 508 | rrd_flag_set_updated(ri, RRD_FLAG_UPDATE_REASON_CHANGED_LAST_TIME_T); |
| 509 | } |
| 510 | |
| 511 | if(likely(live_retention)) |
| 512 | rrdinstance_set_deleted(ri, RRD_FLAG_UPDATE_REASON_ZERO_RETENTION); |
| 513 | } |
| 514 | else { |
| 515 | rrd_flag_clear(ri, RRD_FLAG_UPDATE_REASON_ZERO_RETENTION); |
| 516 | |
| 517 | if (unlikely(ri->first_time_s != min_first_time_t)) { |
| 518 | ri->first_time_s = min_first_time_t; |
| 519 | rrd_flag_set_updated(ri, RRD_FLAG_UPDATE_REASON_CHANGED_FIRST_TIME_T); |
| 520 | } |
| 521 | |
| 522 | if (unlikely(ri->last_time_s != max_last_time_t)) { |
| 523 | ri->last_time_s = max_last_time_t; |
| 524 | rrd_flag_set_updated(ri, RRD_FLAG_UPDATE_REASON_CHANGED_LAST_TIME_T); |
| 525 | } |
| 526 | |
| 527 | if(likely(currently_collected)) |
| 528 | rrdinstance_set_collected(ri); |
| 529 | else |
| 530 | rrdinstance_set_archived(ri); |
| 531 | } |
| 532 | } |
| 533 | |
| 534 | rrd_flag_unset_updated(ri); |
| 535 | } |
| 536 | |
| 537 | static bool rrdinstance_forcefully_clear_retention(RRDCONTEXT *rc, size_t count, const char *descr) { |
| 538 | if(!count) return false; |
| 539 | |
| 540 | RRDHOST *host = rc->rrdhost; |
| 541 | |
| 542 | time_t from_s = LONG_MAX; |
| 543 | time_t to_s = 0; |
| 544 | |
| 545 | size_t instances_deleted = 0; |
| 546 | size_t metrics_deleted = 0; |
| 547 | RRDINSTANCE *ri; |
| 548 | dfe_start_read(rc->rrdinstances, ri) { |
| 549 | if(!rrd_flag_check(ri, RRD_FLAG_NO_TIER0_RETENTION) || rrd_flag_is_collected(ri) || ri->rrdset) |
| 550 | continue; |
| 551 | |
| 552 | size_t metrics_cleared = 0; |
| 553 | RRDMETRIC *rm; |
| 554 | dfe_start_read(ri->rrdmetrics, rm) { |
| 555 | if(!rrd_flag_check(rm, RRD_FLAG_NO_TIER0_RETENTION) || rrd_flag_is_collected(rm) || rrdmetric_rrddim_atomic_load(rm)) |
| 556 | continue; |
| 557 | |
| 558 | rrdmetric_update_retention(rm); |
| 559 | |
| 560 | if(rm->first_time_s < from_s) |
| 561 | from_s = rm->first_time_s; |
| 562 | |
| 563 | if(rm->last_time_s > to_s) |
| 564 | to_s = rm->last_time_s; |
| 565 | |
| 566 | for (size_t tier = 0; tier < nd_profile.storage_tiers; tier++) { |
| 567 | STORAGE_ENGINE *eng = host->db[tier].eng; |
| 568 | eng->api.metric_retention_delete_by_id(host->db[tier].si, rm->uuid); |
| 569 | } |
| 570 | |
| 571 | metrics_cleared++; |
| 572 | metrics_deleted++; |
| 573 | rrdmetric_update_retention(rm); |
| 574 | rrdmetric_trigger_updates(rm, __FUNCTION__ ); |
| 575 | } |
| 576 | dfe_done(rm); |
| 577 | |
| 578 | if(metrics_cleared) { |
| 579 | rrdinstance_trigger_updates(ri, __FUNCTION__ ); |
| 580 | instances_deleted++; |
| 581 | |
| 582 | if(--count == 0) |
| 583 | break; |
| 584 | } |
| 585 | } |
| 586 | dfe_done(ri); |
| 587 | |
| 588 | if(metrics_deleted) { |
| 589 | char from_txt[128], to_txt[128]; |
| 590 | |
| 591 | if(!from_s || from_s == LONG_MAX) |
| 592 | snprintfz(from_txt, sizeof(from_txt), "%s", "NONE"); |
| 593 | else |
| 594 | rfc3339_datetime_ut(from_txt, sizeof(from_txt), from_s * USEC_PER_SEC, 0, true); |
| 595 | |
| 596 | if(!to_s) |
| 597 | snprintfz(to_txt, sizeof(to_txt), "%s", "NONE"); |
| 598 | else |
| 599 | rfc3339_datetime_ut(to_txt, sizeof(to_txt), to_s * USEC_PER_SEC, 0, true); |
| 600 | |
| 601 | ND_LOG_STACK lgs[] = { |
| 602 | ND_LOG_FIELD_TXT(NDF_MODULE, "extreme cardinality protection"), |
| 603 | ND_LOG_FIELD_STR(NDF_NIDL_NODE, rc->rrdhost->hostname), |
| 604 | ND_LOG_FIELD_STR(NDF_NIDL_CONTEXT, rc->id), |
| 605 | ND_LOG_FIELD_UUID(NDF_MESSAGE_ID, &extreme_cardinality_msgid), |
| 606 | ND_LOG_FIELD_END(), |
| 607 | }; |
| 608 | ND_LOG_STACK_PUSH(lgs); |
| 609 | |
| 610 | nd_log(NDLS_DAEMON, NDLP_NOTICE, |
| 611 | "EXTREME CARDINALITY PROTECTION: host '%s', context '%s', %s: " |
| 612 | "forcefully cleared the retention of %zu metrics and %zu instances, " |
| 613 | "having non-tier0 retention from %s to %s.", |
| 614 | rrdhost_hostname(rc->rrdhost), |
| 615 | string2str(rc->id), |
| 616 | descr, |
| 617 | metrics_deleted, instances_deleted, |
| 618 | from_txt, to_txt); |
| 619 | |
| 620 | return true; |
| 621 | } |
| 622 | |
| 623 | return false; |
| 624 | } |
| 625 | |
| 626 | bool rrdcontext_post_process_updates(RRDCONTEXT *rc, bool force, RRD_FLAGS reason, bool worker_jobs) { |
| 627 | bool ret = false; |
| 628 | |
| 629 | if(reason != RRD_FLAG_NONE) |
| 630 | rrd_flag_set_updated(rc, reason); |
| 631 | |
| 632 | if(worker_jobs) |
| 633 | worker_is_busy(WORKER_JOB_PP_CONTEXT); |
| 634 | |
| 635 | size_t min_priority_collected = LONG_MAX; |
| 636 | size_t min_priority_not_collected = LONG_MAX; |
| 637 | size_t min_priority = LONG_MAX; |
| 638 | time_t min_first_time_t = LONG_MAX, max_last_time_t = 0; |
| 639 | size_t instances_active = 0, instances_deleted = 0, instances_no_tier0 = 0; |
| 640 | bool live_retention = true, currently_collected = false, hidden = true; |
| 641 | if(dictionary_entries(rc->rrdinstances) > 0) { |
| 642 | RRDINSTANCE *ri; |
| 643 | dfe_start_reentrant(rc->rrdinstances, ri) { |
| 644 | if(unlikely(worker_jobs && !service_running(SERVICE_CONTEXT))) break; |
| 645 | |
| 646 | RRD_FLAGS reason_to_pass = reason; |
| 647 | if(rrd_flag_check(rc, RRD_FLAG_UPDATE_REASON_UPDATE_RETENTION)) |
| 648 | reason_to_pass |= RRD_FLAG_UPDATE_REASON_UPDATE_RETENTION; |
| 649 | |
| 650 | rrdinstance_post_process_updates(ri, force, reason_to_pass, worker_jobs); |
| 651 | |
| 652 | if(unlikely(hidden && !rrd_flag_check(ri, RRD_FLAG_HIDDEN))) |
| 653 | hidden = false; |
| 654 | |
| 655 | if(unlikely(live_retention && !rrd_flag_check(ri, RRD_FLAG_LIVE_RETENTION))) |
| 656 | live_retention = false; |
| 657 | |
| 658 | if (unlikely(rrdinstance_should_be_deleted(ri))) { |
| 659 | instances_deleted++; |
| 660 | continue; |
| 661 | } |
| 662 | |
| 663 | if(unlikely(rrd_flag_check(ri, RRD_FLAG_NO_TIER0_RETENTION))) |
| 664 | instances_no_tier0++; |
| 665 | |
| 666 | bool ri_collected = rrd_flag_is_collected(ri); |
| 667 | |
| 668 | if(ri_collected && !rrd_flag_check(ri, RRD_FLAG_MERGED_COLLECTED_RI_TO_RC)) { |
| 669 | rrdcontext_update_from_collected_rrdinstance(ri); |
| 670 | rrd_flag_set(ri, RRD_FLAG_MERGED_COLLECTED_RI_TO_RC); |
| 671 | } |
| 672 | |
| 673 | if(unlikely(!currently_collected && rrd_flag_is_collected(ri) && ri->first_time_s)) |
| 674 | currently_collected = true; |
| 675 | |
| 676 | internal_error(rc->units != ri->units, |
| 677 | "RRDCONTEXT: '%s' rrdinstance '%s' has different units, context '%s', instance '%s'", |
| 678 | string2str(rc->id), string2str(ri->id), |
| 679 | string2str(rc->units), string2str(ri->units)); |
| 680 | |
| 681 | instances_active++; |
| 682 | |
| 683 | if (ri->priority >= RRDCONTEXT_MINIMUM_ALLOWED_PRIORITY) { |
| 684 | if(rrd_flag_is_collected(ri)) { |
| 685 | if(ri->priority < min_priority_collected) |
| 686 | min_priority_collected = ri->priority; |
| 687 | } |
| 688 | else { |
| 689 | if(ri->priority < min_priority_not_collected) |
| 690 | min_priority_not_collected = ri->priority; |
| 691 | } |
| 692 | } |
| 693 | |
| 694 | if (ri->first_time_s && ri->first_time_s < min_first_time_t) |
| 695 | min_first_time_t = ri->first_time_s; |
| 696 | |
| 697 | if (ri->last_time_s && ri->last_time_s > max_last_time_t) |
| 698 | max_last_time_t = ri->last_time_s; |
| 699 | } |
| 700 | dfe_done(ri); |
| 701 | |
| 702 | if(extreme_cardinality.enabled && |
| 703 | __atomic_load_n(&extreme_cardinality.db_rotations, __ATOMIC_RELAXED) && |
| 704 | instances_active && |
| 705 | instances_no_tier0 >= extreme_cardinality.instances_count) { |
| 706 | size_t percent = (100 * instances_no_tier0 / instances_active); |
| 707 | if(percent >= extreme_cardinality.active_vs_archived_percentage) { |
| 708 | size_t to_keep = extreme_cardinality.active_vs_archived_percentage * instances_active / 100; |
| 709 | to_keep = MAX(to_keep, extreme_cardinality.instances_count); |
| 710 | size_t to_remove = instances_no_tier0 > to_keep ? instances_no_tier0 - to_keep : 0; |
| 711 | |
| 712 | if(to_remove) { |
| 713 | char buf[256]; |
| 714 | snprintfz(buf, sizeof(buf), |
| 715 | "total active instances %zu, not in tier0 %zu, ephemerality %zu%%", |
| 716 | instances_active, instances_no_tier0, percent); |
| 717 | ret = rrdinstance_forcefully_clear_retention(rc, to_remove, buf); |
| 718 | } |
| 719 | } |
| 720 | } |
| 721 | |
| 722 | if(min_priority_collected != LONG_MAX) |
| 723 | // use the collected priority |
| 724 | min_priority = min_priority_collected; |
| 725 | else |
| 726 | // use the non-collected priority |
| 727 | min_priority = min_priority_not_collected; |
| 728 | } |
| 729 | |
| 730 | { |
| 731 | bool previous_hidden = rrd_flag_check(rc, RRD_FLAG_HIDDEN); |
| 732 | if (hidden != previous_hidden) { |
| 733 | if (hidden && !rrd_flag_check(rc, RRD_FLAG_HIDDEN)) |
| 734 | rrd_flag_set(rc, RRD_FLAG_HIDDEN); |
| 735 | else if (!hidden && rrd_flag_check(rc, RRD_FLAG_HIDDEN)) |
| 736 | rrd_flag_clear(rc, RRD_FLAG_HIDDEN); |
| 737 | } |
| 738 | |
| 739 | bool previous_live_retention = rrd_flag_check(rc, RRD_FLAG_LIVE_RETENTION); |
| 740 | if (live_retention != previous_live_retention) { |
| 741 | if (live_retention && !rrd_flag_check(rc, RRD_FLAG_LIVE_RETENTION)) |
| 742 | rrd_flag_set(rc, RRD_FLAG_LIVE_RETENTION); |
| 743 | else if (!live_retention && rrd_flag_check(rc, RRD_FLAG_LIVE_RETENTION)) |
| 744 | rrd_flag_clear(rc, RRD_FLAG_LIVE_RETENTION); |
| 745 | } |
| 746 | } |
| 747 | |
| 748 | rrdcontext_lock(rc); |
| 749 | rc->pp.executions++; |
| 750 | |
| 751 | if(unlikely(!instances_active)) { |
| 752 | // we had some instances, but they are gone now... |
| 753 | |
| 754 | if(rc->first_time_s) { |
| 755 | rc->first_time_s = 0; |
| 756 | rrd_flag_set_updated(rc, RRD_FLAG_UPDATE_REASON_CHANGED_FIRST_TIME_T); |
| 757 | } |
| 758 | |
| 759 | if(rc->last_time_s) { |
| 760 | rc->last_time_s = 0; |
| 761 | rrd_flag_set_updated(rc, RRD_FLAG_UPDATE_REASON_CHANGED_LAST_TIME_T); |
| 762 | } |
| 763 | |
| 764 | rrdcontext_set_deleted(rc, RRD_FLAG_UPDATE_REASON_ZERO_RETENTION); |
| 765 | } |
| 766 | else { |
| 767 | // we have some active instances... |
| 768 | |
| 769 | if (unlikely(min_first_time_t == LONG_MAX)) |
| 770 | min_first_time_t = 0; |
| 771 | |
| 772 | if (unlikely(min_first_time_t == 0 && max_last_time_t == 0)) { |
| 773 | if(rc->first_time_s) { |
| 774 | rc->first_time_s = 0; |
| 775 | rrd_flag_set_updated(rc, RRD_FLAG_UPDATE_REASON_CHANGED_FIRST_TIME_T); |
| 776 | } |
| 777 | |
| 778 | if(rc->last_time_s) { |
| 779 | rc->last_time_s = 0; |
| 780 | rrd_flag_set_updated(rc, RRD_FLAG_UPDATE_REASON_CHANGED_LAST_TIME_T); |
| 781 | } |
| 782 | |
| 783 | rrdcontext_set_deleted(rc, RRD_FLAG_UPDATE_REASON_ZERO_RETENTION); |
| 784 | } |
| 785 | else { |
| 786 | rrd_flag_clear(rc, RRD_FLAG_UPDATE_REASON_ZERO_RETENTION); |
| 787 | |
| 788 | if (unlikely(rc->first_time_s != min_first_time_t)) { |
| 789 | rc->first_time_s = min_first_time_t; |
| 790 | rrd_flag_set_updated(rc, RRD_FLAG_UPDATE_REASON_CHANGED_FIRST_TIME_T); |
| 791 | } |
| 792 | |
| 793 | if (rc->last_time_s != max_last_time_t) { |
| 794 | rc->last_time_s = max_last_time_t; |
| 795 | rrd_flag_set_updated(rc, RRD_FLAG_UPDATE_REASON_CHANGED_LAST_TIME_T); |
| 796 | } |
| 797 | |
| 798 | if(likely(currently_collected)) |
| 799 | rrdcontext_set_collected(rc); |
| 800 | else |
| 801 | rrdcontext_set_archived(rc); |
| 802 | } |
| 803 | |
| 804 | if (min_priority != LONG_MAX && rc->priority != min_priority) { |
| 805 | rc->priority = min_priority; |
| 806 | rrd_flag_set_updated(rc, RRD_FLAG_UPDATE_REASON_CHANGED_METADATA); |
| 807 | } |
| 808 | } |
| 809 | |
| 810 | if(unlikely(rrd_flag_is_updated(rc))) { |
| 811 | if(check_if_cloud_version_changed_unsafe(rc, false)) { |
| 812 | rc->version = rrdcontext_get_next_version(rc); |
| 813 | rrdcontext_add_to_hub_queue(rc); |
| 814 | } |
| 815 | } |
| 816 | |
| 817 | rrd_flag_unset_updated(rc); |
| 818 | rrdcontext_unlock(rc); |
| 819 | |
| 820 | return ret; |
| 821 | } |
| 822 | |
| 823 | void rrdcontext_queue_for_post_processing(RRDCONTEXT *rc, const char *function __maybe_unused, RRD_FLAGS flags __maybe_unused) { |
| 824 | #if 0 |
| 825 | if(string_strcmp(rc->id, "system.cpu") == 0) { |
| 826 | CLEAN_BUFFER *wb = buffer_create(0, NULL); |
| 827 | buffer_json_initialize(wb, "\"", "\"", 0, true, BUFFER_JSON_OPTIONS_MINIFY); |
| 828 | buffer_json_member_add_array(wb, "flags"); |
| 829 | rrd_flags_to_buffer_json_array_items(rc->flags, wb); |
| 830 | buffer_json_array_close(wb); |
| 831 | buffer_json_member_add_array(wb, "reasons"); |
| 832 | rrd_reasons_to_buffer_json_array_items(rc->flags, wb); |
| 833 | buffer_json_array_close(wb); |
| 834 | buffer_json_finalize(wb); |
| 835 | nd_log(NDLS_DAEMON, NDLP_EMERG, "%s() context '%s', triggered: %s", |
| 836 | function, string2str(rc->id), buffer_tostring(wb)); |
| 837 | } |
| 838 | #endif |
| 839 | |
| 840 | rrdcontext_add_to_pp_queue(rc); |
| 841 | } |
| 842 | |
| 843 | void rrdcontext_initial_processing_after_loading(RRDCONTEXT *rc) { |
| 844 | rrdcontext_dequeue_from_post_processing(rc); |
| 845 | rrdcontext_post_process_updates(rc, false, RRD_FLAG_NONE, false); |
| 846 | } |
| 847 | |
| 848 | void rrdcontext_delete_after_loading(RRDHOST *host, RRDCONTEXT *rc) { |
| 849 | rrdcontext_del_from_hub_queue(rc, false); |
| 850 | rrdcontext_dequeue_from_post_processing(rc); |
| 851 | dictionary_del(host->rrdctx.contexts, string2str(rc->id)); |
| 852 | } |
| 853 | |
| 854 | // ---------------------------------------------------------------------------- |
| 855 | // dispatching contexts to cloud |
| 856 | |
| 857 | static uint64_t rrdcontext_get_next_version(RRDCONTEXT *rc) { |
| 858 | time_t now = now_realtime_sec(); |
| 859 | uint64_t version = MAX(rc->version, rc->hub.version); |
| 860 | version = MAX((uint64_t)now, version); |
| 861 | version++; |
| 862 | return version; |
| 863 | } |
| 864 | |
| 865 | void rrdcontext_message_send_unsafe(RRDCONTEXT *rc, bool snapshot __maybe_unused, void *bundle __maybe_unused) { |
| 866 | |
| 867 | // save it, so that we know the last version we sent to hub |
| 868 | rc->version = rc->hub.version = rrdcontext_get_next_version(rc); |
| 869 | rc->hub.id = string2str(rc->id); |
| 870 | rc->hub.title = string2str(rc->title); |
| 871 | rc->hub.units = string2str(rc->units); |
| 872 | rc->hub.family = string2str(rc->family); |
| 873 | rc->hub.chart_type = rrdset_type_name(rc->chart_type); |
| 874 | rc->hub.priority = rc->priority; |
| 875 | rc->hub.first_time_s = rc->first_time_s; |
| 876 | rc->hub.last_time_s = rrd_flag_is_collected(rc) ? 0 : rc->last_time_s; |
| 877 | rc->hub.deleted = rrd_flag_is_deleted(rc) ? true : false; |
| 878 | |
| 879 | struct context_updated message = { |
| 880 | .id = rc->hub.id, |
| 881 | .version = rc->hub.version, |
| 882 | .title = rc->hub.title, |
| 883 | .units = rc->hub.units, |
| 884 | .family = rc->hub.family, |
| 885 | .chart_type = rc->hub.chart_type, |
| 886 | .priority = rc->hub.priority, |
| 887 | .first_entry = rc->hub.first_time_s, |
| 888 | .last_entry = rc->hub.last_time_s, |
| 889 | .deleted = rc->hub.deleted, |
| 890 | }; |
| 891 | |
| 892 | if(likely(!rrd_flag_check(rc, RRD_FLAG_HIDDEN))) { |
| 893 | if (snapshot) { |
| 894 | if (!rc->hub.deleted) |
| 895 | contexts_snapshot_add_ctx_update(bundle, &message); |
| 896 | } |
| 897 | else |
| 898 | contexts_updated_add_ctx_update(bundle, &message); |
| 899 | } |
| 900 | |
| 901 | // store it to SQL |
| 902 | |
| 903 | if(rrd_flag_is_deleted(rc)) |
| 904 | rrdcontext_delete_from_sql_unsafe(rc); |
| 905 | |
| 906 | else { |
| 907 | if (rc->rrdhost->rrd_memory_mode != RRD_DB_MODE_DBENGINE) |
| 908 | return; |
| 909 | if (ctx_store_context(&rc->rrdhost->host_id.uuid, &rc->hub) != 0) |
| 910 | netdata_log_error( |
| 911 | "RRDCONTEXT: failed to save context '%s' version %" PRIu64 " to SQL.", rc->hub.id, rc->hub.version); |
| 912 | } |
| 913 | } |
| 914 | |
| 915 | bool check_if_cloud_version_changed_unsafe(RRDCONTEXT *rc, bool sending __maybe_unused) { |
| 916 | bool id_changed = false, |
| 917 | title_changed = false, |
| 918 | units_changed = false, |
| 919 | family_changed = false, |
| 920 | chart_type_changed = false, |
| 921 | priority_changed = false, |
| 922 | first_time_changed = false, |
| 923 | last_time_changed = false, |
| 924 | deleted_changed = false; |
| 925 | |
| 926 | RRD_FLAGS flags = rrd_flags_get(rc); |
| 927 | |
| 928 | if(unlikely(string2str(rc->id) != rc->hub.id)) |
| 929 | id_changed = true; |
| 930 | |
| 931 | if(unlikely(string2str(rc->title) != rc->hub.title)) |
| 932 | title_changed = true; |
| 933 | |
| 934 | if(unlikely(string2str(rc->units) != rc->hub.units)) |
| 935 | units_changed = true; |
| 936 | |
| 937 | if(unlikely(string2str(rc->family) != rc->hub.family)) |
| 938 | family_changed = true; |
| 939 | |
| 940 | if(unlikely(rrdset_type_name(rc->chart_type) != rc->hub.chart_type)) |
| 941 | chart_type_changed = true; |
| 942 | |
| 943 | if(unlikely(rc->priority != rc->hub.priority)) |
| 944 | priority_changed = true; |
| 945 | |
| 946 | if(unlikely((uint64_t)rc->first_time_s != rc->hub.first_time_s)) |
| 947 | first_time_changed = true; |
| 948 | |
| 949 | if(unlikely((uint64_t)((flags & RRD_FLAG_COLLECTED) ? 0 : rc->last_time_s) != rc->hub.last_time_s)) |
| 950 | last_time_changed = true; |
| 951 | |
| 952 | if(unlikely(((flags & RRD_FLAG_DELETED) ? true : false) != rc->hub.deleted)) |
| 953 | deleted_changed = true; |
| 954 | |
| 955 | if(unlikely(id_changed || title_changed || units_changed || family_changed || chart_type_changed || priority_changed || first_time_changed || last_time_changed || deleted_changed)) { |
| 956 | |
| 957 | internal_error(LOG_TRANSITIONS, |
| 958 | "RRDCONTEXT: %s NEW VERSION '%s'%s of host '%s', version %"PRIu64", title '%s'%s, units '%s'%s, family '%s'%s, chart type '%s'%s, priority %u%s, first_time_t %ld%s, last_time_t %ld%s, deleted '%s'%s, (queued for %llu ms, expected %llu ms)", |
| 959 | sending?"SENDING":"QUEUE", |
| 960 | string2str(rc->id), id_changed ? " (CHANGED)" : "", |
| 961 | rrdhost_hostname(rc->rrdhost), |
| 962 | rc->version, |
| 963 | string2str(rc->title), title_changed ? " (CHANGED)" : "", |
| 964 | string2str(rc->units), units_changed ? " (CHANGED)" : "", |
| 965 | string2str(rc->family), family_changed ? " (CHANGED)" : "", |
| 966 | rrdset_type_name(rc->chart_type), chart_type_changed ? " (CHANGED)" : "", |
| 967 | rc->priority, priority_changed ? " (CHANGED)" : "", |
| 968 | rc->first_time_s, first_time_changed ? " (CHANGED)" : "", |
| 969 | (flags & RRD_FLAG_COLLECTED) ? 0 : rc->last_time_s, last_time_changed ? " (CHANGED)" : "", |
| 970 | (flags & RRD_FLAG_DELETED) ? "true" : "false", deleted_changed ? " (CHANGED)" : "", |
| 971 | sending ? (now_realtime_usec() - rc->queue.queued_ut) / USEC_PER_MS : 0, |
| 972 | sending ? (rc->queue.scheduled_dispatch_ut - rc->queue.queued_ut) / USEC_PER_MS : 0 |
| 973 | ); |
| 974 | |
| 975 | rrdhost_update_cached_retention(rc->rrdhost, rc->first_time_s, rc->last_time_s, false); |
| 976 | |
| 977 | return true; |
| 978 | } |
| 979 | |
| 980 | if(!(flags & RRD_FLAG_COLLECTED)) |
| 981 | rrdhost_update_cached_retention(rc->rrdhost, rc->first_time_s, rc->last_time_s, false); |
| 982 | |
| 983 | return false; |
| 984 | } |
| 985 | |
| 986 | usec_t rrdcontext_calculate_queued_dispatch_time_ut(RRDCONTEXT *rc, usec_t now_ut) { |
| 987 | |
| 988 | if(likely(rc->queue.delay_calc_ut >= rc->queue.queued_ut)) |
| 989 | return rc->queue.scheduled_dispatch_ut; |
| 990 | |
| 991 | RRD_FLAGS flags = rc->queue.queued_flags; |
| 992 | |
| 993 | usec_t delay = LONG_MAX; |
| 994 | int i; |
| 995 | struct rrdcontext_reason *reason; |
| 996 | for(i = 0, reason = &rrdcontext_reasons[i]; reason->name ; reason = &rrdcontext_reasons[++i]) { |
| 997 | if(unlikely(flags & reason->flag)) { |
| 998 | if(reason->delay_ut < delay) |
| 999 | delay = reason->delay_ut; |
| 1000 | } |
| 1001 | } |
| 1002 | |
| 1003 | if(unlikely(delay == LONG_MAX)) { |
| 1004 | internal_error(true, "RRDCONTEXT: '%s', cannot find minimum delay of flags %x", string2str(rc->id), (unsigned int)flags); |
| 1005 | delay = 60 * USEC_PER_SEC; |
| 1006 | } |
| 1007 | |
| 1008 | rc->queue.delay_calc_ut = now_ut; |
| 1009 | usec_t dispatch_ut = rc->queue.scheduled_dispatch_ut = rc->queue.queued_ut + delay; |
| 1010 | return dispatch_ut; |
| 1011 | } |
| 1012 | |
| 1013 | // ---------------------------------------------------------------------------- |
| 1014 | // worker thread |
| 1015 | |
| 1016 | static void rrdcontext_main_cleanup(void *pptr) { |
| 1017 | struct netdata_static_thread *static_thread = CLEANUP_FUNCTION_GET_PTR(pptr); |
| 1018 | if(!static_thread) return; |
| 1019 | |
| 1020 | static_thread->enabled = NETDATA_MAIN_THREAD_EXITING; |
| 1021 | |
| 1022 | // custom code |
| 1023 | worker_unregister(); |
| 1024 | |
| 1025 | static_thread->enabled = NETDATA_MAIN_THREAD_EXITED; |
| 1026 | } |
| 1027 | |
| 1028 | void rrdcontext_main(void *ptr) { |
| 1029 | CLEANUP_FUNCTION_REGISTER(rrdcontext_main_cleanup) cleanup_ptr = ptr; |
| 1030 | |
| 1031 | worker_register("RRDCONTEXT"); |
| 1032 | worker_register_job_name(WORKER_JOB_HOSTS, "hosts"); |
| 1033 | worker_register_job_name(WORKER_JOB_CHECK, "dedup checks"); |
| 1034 | worker_register_job_name(WORKER_JOB_SEND, "sent contexts"); |
| 1035 | worker_register_job_name(WORKER_JOB_DEQUEUE, "deduplicated contexts"); |
| 1036 | worker_register_job_name(WORKER_JOB_RETENTION, "metrics retention"); |
| 1037 | worker_register_job_name(WORKER_JOB_QUEUED, "queued contexts"); |
| 1038 | worker_register_job_name(WORKER_JOB_CLEANUP, "cleanups"); |
| 1039 | worker_register_job_name(WORKER_JOB_CLEANUP_DELETE, "deletes"); |
| 1040 | worker_register_job_name(WORKER_JOB_PP_METRIC, "check metrics"); |
| 1041 | worker_register_job_name(WORKER_JOB_PP_INSTANCE, "check instances"); |
| 1042 | worker_register_job_name(WORKER_JOB_PP_CONTEXT, "check contexts"); |
| 1043 | |
| 1044 | worker_register_job_custom_metric(WORKER_JOB_HUB_QUEUE_SIZE, "hub queue size", "contexts", WORKER_METRIC_ABSOLUTE); |
| 1045 | worker_register_job_custom_metric(WORKER_JOB_PP_QUEUE_SIZE, "post processing queue size", "contexts", WORKER_METRIC_ABSOLUTE); |
| 1046 | |
| 1047 | heartbeat_t hb; |
| 1048 | heartbeat_init(&hb, RRDCONTEXT_WORKER_THREAD_HEARTBEAT_USEC); |
| 1049 | |
| 1050 | extreme_cardinality.enabled = inicfg_get_boolean( |
| 1051 | &netdata_config, CONFIG_SECTION_DB, "extreme cardinality protection", |
| 1052 | nd_profile.storage_tiers > 1 && default_rrd_memory_mode == RRD_DB_MODE_DBENGINE |
| 1053 | ); |
| 1054 | |
| 1055 | extreme_cardinality.instances_count = inicfg_get_number_range( |
| 1056 | &netdata_config, CONFIG_SECTION_DB, "extreme cardinality keep instances", |
| 1057 | (long long)extreme_cardinality.instances_count, 1, 1000000); |
| 1058 | |
| 1059 | extreme_cardinality.active_vs_archived_percentage = inicfg_get_number_range( |
| 1060 | &netdata_config, CONFIG_SECTION_DB, "extreme cardinality min ephemerality", |
| 1061 | (long long)extreme_cardinality.active_vs_archived_percentage, 0, 100); |
| 1062 | |
| 1063 | while (service_running(SERVICE_CONTEXT)) { |
| 1064 | worker_is_idle(); |
| 1065 | heartbeat_next(&hb); |
| 1066 | |
| 1067 | if(unlikely(!service_running(SERVICE_CONTEXT))) break; |
| 1068 | |
| 1069 | usec_t now_ut = now_realtime_usec(); |
| 1070 | |
| 1071 | usec_t deadline = __atomic_load_n(&rrdcontext_next_db_rotation_ut, __ATOMIC_RELAXED); |
| 1072 | if(deadline && now_ut > deadline) { |
| 1073 | // db_rotations is bumped by rrdcontext_count_db_rotation() from |
| 1074 | // rrdcontext_db_rotation() only -- the chart-cleanup trigger |
| 1075 | // (rrdcontext_request_full_gc) does NOT bump it, so the |
| 1076 | // extreme-cardinality guard at line ~700 still activates only |
| 1077 | // after a real dbengine rotation. |
| 1078 | rrdcontext_recalculate_retention_all_hosts(); |
| 1079 | rrdcontext_garbage_collect_for_all_hosts(); |
| 1080 | // Clear the slot only if it still holds the deadline we |
| 1081 | // processed. Two writers can race this pass: |
| 1082 | // - rrdcontext_request_full_gc() arms only when the slot is |
| 1083 | // zero, but the slot stays non-zero throughout this pass, |
| 1084 | // so any such request during the pass is coalesced into |
| 1085 | // the in-flight pass and does not arm a new deadline. |
| 1086 | // - rrdcontext_db_rotation() stores unconditionally. If it |
| 1087 | // fires during this pass, the slot now holds a fresh |
| 1088 | // deadline; the CAS here fails and that deadline drives |
| 1089 | // the next iteration. |
| 1090 | __atomic_compare_exchange_n(&rrdcontext_next_db_rotation_ut, |
| 1091 | &deadline, 0, |
| 1092 | false, |
| 1093 | __ATOMIC_RELAXED, __ATOMIC_RELAXED); |
| 1094 | |
| 1095 | // If a rrdcontext_request_full_gc() landed mid-pass (its CAS |
| 1096 | // failed because the slot was non-zero with an expired |
| 1097 | // deadline), the host whose archive motivated it may have |
| 1098 | // already been walked in this pass. Schedule a follow-up: |
| 1099 | // arm a fresh deadline iff the slot is currently zero. If |
| 1100 | // another path (e.g. dbengine rotation) just armed a new |
| 1101 | // deadline, the CAS leaves it alone. |
| 1102 | if(__atomic_exchange_n(&rrdcontext_full_gc_rerun_requested, 0, __ATOMIC_RELAXED)) { |
| 1103 | usec_t expected = 0; |
| 1104 | usec_t fresh = now_realtime_usec() + FULL_RETENTION_SCAN_DELAY_AFTER_DB_ROTATION_SECS * USEC_PER_SEC; |
| 1105 | __atomic_compare_exchange_n(&rrdcontext_next_db_rotation_ut, |
| 1106 | &expected, fresh, |
| 1107 | false, |
| 1108 | __ATOMIC_RELAXED, __ATOMIC_RELAXED); |
| 1109 | } |
| 1110 | } |
| 1111 | |
| 1112 | size_t hub_queued_contexts_for_all_hosts = 0; |
| 1113 | size_t pp_queued_contexts_for_all_hosts = 0; |
| 1114 | |
| 1115 | RRDHOST *host; |
| 1116 | dfe_start_reentrant(rrdhost_root_index, host) { |
| 1117 | if(unlikely(!service_running(SERVICE_CONTEXT))) break; |
| 1118 | |
| 1119 | // Allow timed-out deferred checkpoints to replay even if context loading is stuck. |
| 1120 | rrdcontext_hub_pending_checkpoint_replay(host); |
| 1121 | |
| 1122 | if(rrdhost_flag_check(host, RRDHOST_FLAG_PENDING_CONTEXT_LOAD)) |
| 1123 | continue; |
| 1124 | |
| 1125 | if(rrdhost_flag_check(host, RRDHOST_FLAG_RRDCONTEXT_GET_RETENTION)) { |
| 1126 | rrdcontext_recalculate_host_retention(host, RRD_FLAG_UPDATE_REASON_DISCONNECTED_CHILD, false); |
| 1127 | rrdhost_flag_clear(host, RRDHOST_FLAG_RRDCONTEXT_GET_RETENTION); |
| 1128 | } |
| 1129 | |
| 1130 | worker_is_busy(WORKER_JOB_HOSTS); |
| 1131 | |
| 1132 | pp_queued_contexts_for_all_hosts += rrdcontext_queue_entries(&host->rrdctx.pp_queue); |
| 1133 | rrdcontext_post_process_queued_contexts(host); |
| 1134 | |
| 1135 | // replay deferred checkpoint if post-processing drained the queue, or on timeout |
| 1136 | rrdcontext_hub_pending_checkpoint_replay(host); |
| 1137 | |
| 1138 | hub_queued_contexts_for_all_hosts += rrdcontext_queue_entries(&host->rrdctx.hub_queue); |
| 1139 | rrdcontext_dispatch_queued_contexts_to_hub(host, now_ut); |
| 1140 | |
| 1141 | if (host->rrdctx.contexts) |
| 1142 | dictionary_garbage_collect(host->rrdctx.contexts); |
| 1143 | } |
| 1144 | dfe_done(host); |
| 1145 | |
| 1146 | worker_set_metric(WORKER_JOB_HUB_QUEUE_SIZE, (NETDATA_DOUBLE)hub_queued_contexts_for_all_hosts); |
| 1147 | worker_set_metric(WORKER_JOB_PP_QUEUE_SIZE, (NETDATA_DOUBLE)pp_queued_contexts_for_all_hosts); |
| 1148 | } |
| 1149 | } |