| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | #include "ebpf.h" |
| 4 | #include "ebpf_oomkill.h" |
| 5 | #include "libbpf_api/ebpf_library.h" |
| 6 | |
| 7 | struct config oomkill_config = APPCONFIG_INITIALIZER; |
| 8 | |
| 9 | static ebpf_local_maps_t oomkill_maps[] = { |
| 10 | {.name = "tbl_oomkill", |
| 11 | .internal_input = NETDATA_OOMKILL_MAX_ENTRIES, |
| 12 | .user_input = 0, |
| 13 | .type = NETDATA_EBPF_MAP_STATIC, |
| 14 | .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED, |
| 15 | #ifdef LIBBPF_MAJOR_VERSION |
| 16 | .map_type = BPF_MAP_TYPE_PERCPU_HASH |
| 17 | #endif |
| 18 | }, |
| 19 | /* end */ |
| 20 | {.name = NULL, |
| 21 | .internal_input = 0, |
| 22 | .user_input = 0, |
| 23 | .type = NETDATA_EBPF_MAP_CONTROLLER, |
| 24 | .map_fd = ND_EBPF_MAP_FD_NOT_INITIALIZED, |
| 25 | #ifdef LIBBPF_MAJOR_VERSION |
| 26 | .map_type = BPF_MAP_TYPE_PERCPU_HASH |
| 27 | #endif |
| 28 | }}; |
| 29 | |
| 30 | static ebpf_tracepoint_t oomkill_tracepoints[] = { |
| 31 | {.enabled = false, .class = "oom", .event = "mark_victim"}, |
| 32 | /* end */ |
| 33 | {.enabled = false, .class = NULL, .event = NULL}}; |
| 34 | |
| 35 | static netdata_publish_syscall_t oomkill_publish_aggregated = |
| 36 | {.name = "kills", .dimension = "kills", .algorithm = "absolute", .next = NULL}; |
| 37 | |
| 38 | static void ebpf_obsolete_specific_oomkill_charts(char *type, int update_every); |
| 39 | |
| 40 | /** |
| 41 | * Obsolete services |
| 42 | * |
| 43 | * Obsolete all service charts created |
| 44 | * |
| 45 | * @param em a pointer to `struct ebpf_module` |
| 46 | */ |
| 47 | static void ebpf_obsolete_oomkill_services(ebpf_module_t *em, char *id) |
| 48 | { |
| 49 | ebpf_write_chart_obsolete( |
| 50 | id, |
| 51 | NETDATA_OOMKILL_CHART, |
| 52 | "", |
| 53 | "Systemd service OOM kills.", |
| 54 | EBPF_OOMKILL_UNIT_KILLS, |
| 55 | NETDATA_EBPF_MEMORY_GROUP, |
| 56 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 57 | NETDATA_CGROUP_OOMKILLS_CONTEXT, |
| 58 | 20191, |
| 59 | em->update_every); |
| 60 | } |
| 61 | |
| 62 | /** |
| 63 | * Obsolete cgroup chart |
| 64 | * |
| 65 | * Send obsolete for all charts created before to close. |
| 66 | * |
| 67 | * @param em a pointer to `struct ebpf_module` |
| 68 | */ |
| 69 | static inline void ebpf_obsolete_oomkill_cgroup_charts(ebpf_module_t *em) |
| 70 | { |
| 71 | netdata_mutex_lock(&mutex_cgroup_shm); |
| 72 | |
| 73 | ebpf_cgroup_target_t *ect; |
| 74 | for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { |
| 75 | if (ect->systemd) { |
| 76 | ebpf_obsolete_oomkill_services(em, ect->name); |
| 77 | |
| 78 | continue; |
| 79 | } |
| 80 | |
| 81 | ebpf_obsolete_specific_oomkill_charts(ect->name, em->update_every); |
| 82 | } |
| 83 | netdata_mutex_unlock(&mutex_cgroup_shm); |
| 84 | } |
| 85 | |
| 86 | /** |
| 87 | * Obsolete global |
| 88 | * |
| 89 | * Obsolete global charts created by thread. |
| 90 | * |
| 91 | * @param em a pointer to `struct ebpf_module` |
| 92 | */ |
| 93 | static void ebpf_obsolete_oomkill_apps(ebpf_module_t *em) |
| 94 | { |
| 95 | struct ebpf_target *w; |
| 96 | int update_every = em->update_every; |
| 97 | netdata_mutex_lock(&collect_data_mutex); |
| 98 | for (w = apps_groups_root_target; w; w = w->next) { |
| 99 | if (unlikely(!(w->charts_created & (1 << EBPF_MODULE_OOMKILL_IDX)))) |
| 100 | continue; |
| 101 | |
| 102 | ebpf_write_chart_obsolete( |
| 103 | NETDATA_APP_FAMILY, |
| 104 | w->clean_name, |
| 105 | NETDATA_OOMKILL_CHART, |
| 106 | "Processes OOM kills.", |
| 107 | EBPF_OOMKILL_UNIT_KILLS, |
| 108 | NETDATA_EBPF_MEMORY_GROUP, |
| 109 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 110 | "ebpf.app_oomkill", |
| 111 | 20072, |
| 112 | update_every); |
| 113 | |
| 114 | w->charts_created &= ~(1 << EBPF_MODULE_OOMKILL_IDX); |
| 115 | } |
| 116 | netdata_mutex_unlock(&collect_data_mutex); |
| 117 | } |
| 118 | |
| 119 | static void oomkill_cleanup(void *pptr) |
| 120 | { |
| 121 | ebpf_module_t *em = CLEANUP_FUNCTION_GET_PTR(pptr); |
| 122 | if (!em) |
| 123 | return; |
| 124 | |
| 125 | netdata_mutex_lock(&lock); |
| 126 | collect_pids &= ~(1 << EBPF_MODULE_OOMKILL_IDX); |
| 127 | netdata_mutex_unlock(&lock); |
| 128 | |
| 129 | if (ebpf_module_enabled_get(em) == NETDATA_THREAD_EBPF_FUNCTION_RUNNING && !ebpf_plugin_stop()) { |
| 130 | netdata_mutex_lock(&lock); |
| 131 | |
| 132 | if (em->cgroup_charts) { |
| 133 | ebpf_obsolete_oomkill_cgroup_charts(em); |
| 134 | } |
| 135 | |
| 136 | ebpf_obsolete_oomkill_apps(em); |
| 137 | |
| 138 | fflush(stdout); |
| 139 | netdata_mutex_unlock(&lock); |
| 140 | } |
| 141 | |
| 142 | if (!ebpf_plugin_stop() && em->functions.bpf_unload) |
| 143 | em->functions.bpf_unload(em); |
| 144 | |
| 145 | netdata_mutex_lock(&ebpf_exit_cleanup); |
| 146 | ebpf_module_enabled_set(em, NETDATA_THREAD_EBPF_STOPPED); |
| 147 | netdata_mutex_unlock(&ebpf_exit_cleanup); |
| 148 | } |
| 149 | |
| 150 | static void oomkill_write_data(int32_t *keys, uint32_t total) |
| 151 | { |
| 152 | // for each app, see if it was OOM killed. record as 1 if so otherwise 0. |
| 153 | struct ebpf_target *w; |
| 154 | uint32_t used_pid = 0; |
| 155 | netdata_mutex_lock(&collect_data_mutex); |
| 156 | for (w = apps_groups_root_target; w != NULL; w = w->next) { |
| 157 | if (ebpf_plugin_stop()) |
| 158 | break; |
| 159 | |
| 160 | if (unlikely(!(w->charts_created & (1 << EBPF_MODULE_OOMKILL_IDX)))) |
| 161 | continue; |
| 162 | |
| 163 | bool was_oomkilled = false; |
| 164 | if (total) { |
| 165 | struct ebpf_pid_on_target *pids = w->root_pid; |
| 166 | while (pids) { |
| 167 | uint32_t j; |
| 168 | for (j = 0; j < total; j++) { |
| 169 | if (pids->pid == keys[j]) { |
| 170 | used_pid++; |
| 171 | was_oomkilled = true; |
| 172 | // set to 0 so we consider it "done". |
| 173 | keys[j] = 0; |
| 174 | goto write_dim; |
| 175 | } |
| 176 | } |
| 177 | pids = pids->next; |
| 178 | } |
| 179 | } |
| 180 | write_dim: |
| 181 | ebpf_write_begin_chart(NETDATA_APP_FAMILY, w->clean_name, NETDATA_OOMKILL_CHART); |
| 182 | write_chart_dimension(oomkill_publish_aggregated.dimension, was_oomkilled); |
| 183 | ebpf_write_end_chart(); |
| 184 | } |
| 185 | |
| 186 | if (total != used_pid) { |
| 187 | // for any remaining keys for which we couldn't find a group, this could be |
| 188 | // for various reasons, but the primary one is that the PID has not yet |
| 189 | // been picked up by the process thread when parsing the proc filesystem. |
| 190 | // since it's been OOM killed, it will never be parsed in the future, so |
| 191 | // we have no choice but to dump it into `other`. |
| 192 | uint32_t rem_count = total - used_pid; |
| 193 | ebpf_write_begin_chart(NETDATA_APP_FAMILY, "other", NETDATA_OOMKILL_CHART); |
| 194 | write_chart_dimension(oomkill_publish_aggregated.dimension, rem_count); |
| 195 | ebpf_write_end_chart(); |
| 196 | } |
| 197 | |
| 198 | netdata_mutex_unlock(&collect_data_mutex); |
| 199 | } |
| 200 | |
| 201 | /** |
| 202 | * Create specific OOMkill charts |
| 203 | * |
| 204 | * Create charts for cgroup/application. |
| 205 | * |
| 206 | * @param type the chart type. |
| 207 | * @param update_every value to overwrite the update frequency set by the server. |
| 208 | */ |
| 209 | static void ebpf_create_specific_oomkill_charts(char *type, int update_every) |
| 210 | { |
| 211 | ebpf_create_chart( |
| 212 | type, |
| 213 | NETDATA_OOMKILL_CHART, |
| 214 | "Cgroup OOM kills.", |
| 215 | EBPF_OOMKILL_UNIT_KILLS, |
| 216 | NETDATA_EBPF_MEMORY_GROUP, |
| 217 | NETDATA_CGROUP_OOMKILLS_CONTEXT, |
| 218 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 219 | NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5600, |
| 220 | ebpf_create_global_dimension, |
| 221 | &oomkill_publish_aggregated, |
| 222 | 1, |
| 223 | update_every, |
| 224 | NETDATA_EBPF_MODULE_NAME_OOMKILL); |
| 225 | } |
| 226 | |
| 227 | /** |
| 228 | * Create Systemd OOMkill Charts |
| 229 | * |
| 230 | * Create charts when systemd is enabled |
| 231 | * |
| 232 | * @param update_every value to overwrite the update frequency set by the server. |
| 233 | **/ |
| 234 | static void ebpf_create_systemd_oomkill_charts(int update_every) |
| 235 | { |
| 236 | static ebpf_systemd_args_t data_oom = { |
| 237 | .title = "Systemd service OOM kills.", |
| 238 | .units = EBPF_OOMKILL_UNIT_KILLS, |
| 239 | .family = NETDATA_EBPF_MEMORY_GROUP, |
| 240 | .charttype = NETDATA_EBPF_CHART_TYPE_STACKED, |
| 241 | .order = 20191, |
| 242 | .algorithm = EBPF_CHART_ALGORITHM_INCREMENTAL, |
| 243 | .context = NETDATA_SYSTEMD_OOMKILLS_CONTEXT, |
| 244 | .module = NETDATA_EBPF_MODULE_NAME_OOMKILL, |
| 245 | .update_every = 0, |
| 246 | .suffix = NETDATA_OOMKILL_CHART, |
| 247 | .dimension = "kills"}; |
| 248 | |
| 249 | if (!data_oom.update_every) |
| 250 | data_oom.update_every = update_every; |
| 251 | |
| 252 | ebpf_cgroup_target_t *w; |
| 253 | for (w = ebpf_cgroup_pids; w; w = w->next) { |
| 254 | if (unlikely(!w->systemd || w->flags & NETDATA_EBPF_SERVICES_HAS_OOMKILL_CHART)) |
| 255 | continue; |
| 256 | |
| 257 | data_oom.id = w->name; |
| 258 | ebpf_create_charts_on_systemd(&data_oom); |
| 259 | |
| 260 | w->flags |= NETDATA_EBPF_SERVICES_HAS_OOMKILL_CHART; |
| 261 | } |
| 262 | } |
| 263 | |
| 264 | /** |
| 265 | * Send Systemd charts |
| 266 | * |
| 267 | * Send collected data to Netdata. |
| 268 | */ |
| 269 | static void ebpf_send_systemd_oomkill_charts() |
| 270 | { |
| 271 | ebpf_cgroup_target_t *ect; |
| 272 | for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { |
| 273 | if (unlikely(!(ect->flags & NETDATA_EBPF_SERVICES_HAS_OOMKILL_CHART))) { |
| 274 | continue; |
| 275 | } |
| 276 | ebpf_write_begin_chart(ect->name, NETDATA_OOMKILL_CHART, ""); |
| 277 | write_chart_dimension(oomkill_publish_aggregated.dimension, (long long)ect->oomkill); |
| 278 | ect->oomkill = 0; |
| 279 | ebpf_write_end_chart(); |
| 280 | } |
| 281 | } |
| 282 | |
| 283 | /* |
| 284 | * Send Specific OOMkill data |
| 285 | * |
| 286 | * Send data for specific cgroup/apps. |
| 287 | * |
| 288 | * @param type chart type |
| 289 | * @param value value for oomkill |
| 290 | */ |
| 291 | static void ebpf_send_specific_oomkill_data(char *type, int value) |
| 292 | { |
| 293 | ebpf_write_begin_chart(type, NETDATA_OOMKILL_CHART, ""); |
| 294 | write_chart_dimension(oomkill_publish_aggregated.dimension, (long long)value); |
| 295 | ebpf_write_end_chart(); |
| 296 | } |
| 297 | |
| 298 | /** |
| 299 | * Create specific OOMkill charts |
| 300 | * |
| 301 | * Create charts for cgroup/application. |
| 302 | * |
| 303 | * @param type the chart type. |
| 304 | * @param update_every value to overwrite the update frequency set by the server. |
| 305 | */ |
| 306 | static void ebpf_obsolete_specific_oomkill_charts(char *type, int update_every) |
| 307 | { |
| 308 | ebpf_write_chart_obsolete( |
| 309 | type, |
| 310 | NETDATA_OOMKILL_CHART, |
| 311 | "", |
| 312 | "Cgroup OOM kills.", |
| 313 | EBPF_OOMKILL_UNIT_KILLS, |
| 314 | NETDATA_EBPF_MEMORY_GROUP, |
| 315 | NETDATA_EBPF_CHART_TYPE_LINE, |
| 316 | NETDATA_CGROUP_OOMKILLS_CONTEXT, |
| 317 | NETDATA_CHART_PRIO_CGROUPS_CONTAINERS + 5600, |
| 318 | update_every); |
| 319 | } |
| 320 | |
| 321 | /** |
| 322 | * Send data to Netdata calling auxiliary functions. |
| 323 | * |
| 324 | * @param update_every value to overwrite the update frequency set by the server. |
| 325 | */ |
| 326 | void ebpf_oomkill_send_cgroup_data(int update_every) |
| 327 | { |
| 328 | netdata_mutex_lock(&mutex_cgroup_shm); |
| 329 | ebpf_cgroup_target_t *ect; |
| 330 | |
| 331 | if (ebpf_cgroup_systemd_enabled_get()) { |
| 332 | if (ebpf_send_cgroup_chart_get()) { |
| 333 | ebpf_create_systemd_oomkill_charts(update_every); |
| 334 | } |
| 335 | ebpf_send_systemd_oomkill_charts(); |
| 336 | } |
| 337 | |
| 338 | for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { |
| 339 | if (ect->systemd) |
| 340 | continue; |
| 341 | |
| 342 | if (!(ect->flags & NETDATA_EBPF_CGROUP_HAS_OOMKILL_CHART) && ect->updated) { |
| 343 | ebpf_create_specific_oomkill_charts(ect->name, update_every); |
| 344 | ect->flags |= NETDATA_EBPF_CGROUP_HAS_OOMKILL_CHART; |
| 345 | } |
| 346 | |
| 347 | if (ect->flags & NETDATA_EBPF_CGROUP_HAS_OOMKILL_CHART) { |
| 348 | if (ect->updated) { |
| 349 | ebpf_send_specific_oomkill_data(ect->name, ect->oomkill); |
| 350 | } else { |
| 351 | ebpf_obsolete_specific_oomkill_charts(ect->name, update_every); |
| 352 | ect->flags &= ~NETDATA_EBPF_CGROUP_HAS_OOMKILL_CHART; |
| 353 | } |
| 354 | } |
| 355 | } |
| 356 | |
| 357 | netdata_mutex_unlock(&mutex_cgroup_shm); |
| 358 | } |
| 359 | |
| 360 | /** |
| 361 | * Read data |
| 362 | * |
| 363 | * Read OOMKILL events from table. |
| 364 | * |
| 365 | * @param keys vector where data will be stored |
| 366 | * |
| 367 | * @return It returns the number of read elements |
| 368 | */ |
| 369 | static uint32_t oomkill_read_data(int32_t *keys) |
| 370 | { |
| 371 | // the first `i` entries of `keys` will contain the currently active PIDs |
| 372 | // in the eBPF map. |
| 373 | uint32_t i = 0; |
| 374 | |
| 375 | uint32_t curr_key = 0; |
| 376 | uint32_t key; |
| 377 | int mapfd = oomkill_maps[0].map_fd; |
| 378 | while (bpf_map_get_next_key(mapfd, &curr_key, &key) == 0) { |
| 379 | if (ebpf_plugin_stop()) |
| 380 | break; |
| 381 | |
| 382 | curr_key = key; |
| 383 | |
| 384 | keys[i] = (int32_t)key; |
| 385 | i += 1; |
| 386 | |
| 387 | if (unlikely(bpf_map_delete_elem(mapfd, &key) < 0)) { |
| 388 | netdata_log_error("key unexpectedly not available for deletion."); |
| 389 | } |
| 390 | if (i >= NETDATA_OOMKILL_MAX_ENTRIES) |
| 391 | break; |
| 392 | } |
| 393 | |
| 394 | return i; |
| 395 | } |
| 396 | |
| 397 | /** |
| 398 | * Update cgroup |
| 399 | * |
| 400 | * Update cgroup data based in |
| 401 | * |
| 402 | * @param keys vector with pids that had oomkill event |
| 403 | * @param total number of elements in keys vector. |
| 404 | */ |
| 405 | static void ebpf_update_oomkill_cgroup(int32_t *keys, uint32_t total) |
| 406 | { |
| 407 | ebpf_cgroup_target_t *ect; |
| 408 | netdata_mutex_lock(&mutex_cgroup_shm); |
| 409 | for (ect = ebpf_cgroup_pids; ect; ect = ect->next) { |
| 410 | ect->oomkill = 0; |
| 411 | struct pid_on_target2 *pids; |
| 412 | for (pids = ect->pids; pids; pids = pids->next) { |
| 413 | uint32_t j; |
| 414 | int32_t pid = pids->pid; |
| 415 | for (j = 0; j < total; j++) { |
| 416 | if (pid == keys[j]) { |
| 417 | ect->oomkill = 1; |
| 418 | break; |
| 419 | } |
| 420 | } |
| 421 | } |
| 422 | } |
| 423 | netdata_mutex_unlock(&mutex_cgroup_shm); |
| 424 | } |
| 425 | |
| 426 | /** |
| 427 | * Update OOMkill period |
| 428 | * |
| 429 | * Update oomkill period according function arguments. |
| 430 | * |
| 431 | * @param running_time current value of running_value. |
| 432 | * @param em the thread main structure. |
| 433 | * |
| 434 | * @return It returns new running_time value. |
| 435 | */ |
| 436 | static int ebpf_update_oomkill_period(int running_time, ebpf_module_t *em) |
| 437 | { |
| 438 | netdata_mutex_lock(&ebpf_exit_cleanup); |
| 439 | running_time += em->update_every; |
| 440 | em->running_time = running_time; |
| 441 | netdata_mutex_unlock(&ebpf_exit_cleanup); |
| 442 | |
| 443 | return running_time; |
| 444 | } |
| 445 | |
| 446 | /** |
| 447 | * Main loop for this collector. |
| 448 | * |
| 449 | * @param em the thread main structure. |
| 450 | */ |
| 451 | static void oomkill_collector(ebpf_module_t *em) |
| 452 | { |
| 453 | int cgroups = em->cgroup_charts; |
| 454 | int update_every = em->update_every; |
| 455 | int32_t keys[NETDATA_OOMKILL_MAX_ENTRIES]; |
| 456 | |
| 457 | // loop and read until ebpf plugin is closed. |
| 458 | int counter = update_every - 1; |
| 459 | uint32_t running_time = 0; |
| 460 | uint32_t lifetime = em->lifetime; |
| 461 | netdata_idx_t *stats = em->hash_table_stats; |
| 462 | heartbeat_t hb; |
| 463 | heartbeat_init(&hb, USEC_PER_SEC); |
| 464 | while (!ebpf_plugin_stop() && running_time < lifetime) { |
| 465 | if (ebpf_plugin_stop()) |
| 466 | break; |
| 467 | |
| 468 | (void)heartbeat_next(&hb); |
| 469 | if (ebpf_plugin_stop()) |
| 470 | break; |
| 471 | |
| 472 | if (++counter != update_every) |
| 473 | continue; |
| 474 | |
| 475 | counter = 0; |
| 476 | |
| 477 | uint32_t count = oomkill_read_data(keys); |
| 478 | |
| 479 | stats[NETDATA_CONTROLLER_PID_TABLE_ADD] += (uint64_t)count; |
| 480 | stats[NETDATA_CONTROLLER_PID_TABLE_DEL] += (uint64_t)count; |
| 481 | |
| 482 | if (cgroups && ebpf_cgroup_integration_active_get()) |
| 483 | ebpf_update_oomkill_cgroup(keys, count); |
| 484 | |
| 485 | if (ebpf_plugin_stop()) |
| 486 | break; |
| 487 | |
| 488 | netdata_apps_integration_flags_t apps = em->apps_charts; |
| 489 | netdata_mutex_lock(&lock); |
| 490 | // write everything from the ebpf map. |
| 491 | if (cgroups && ebpf_cgroup_integration_active_get()) |
| 492 | ebpf_oomkill_send_cgroup_data(update_every); |
| 493 | |
| 494 | if (apps & NETDATA_EBPF_APPS_FLAG_CHART_CREATED) |
| 495 | oomkill_write_data(keys, count); |
| 496 | |
| 497 | netdata_mutex_unlock(&lock); |
| 498 | |
| 499 | running_time = ebpf_update_oomkill_period(running_time, em); |
| 500 | } |
| 501 | } |
| 502 | |
| 503 | /** |
| 504 | * Create apps charts |
| 505 | * |
| 506 | * Call ebpf_create_chart to create the charts on apps submenu. |
| 507 | * |
| 508 | * @param em a pointer to the structure with the default values. |
| 509 | */ |
| 510 | void ebpf_oomkill_create_apps_charts(struct ebpf_module *em, void *ptr) |
| 511 | { |
| 512 | struct ebpf_target *root = ptr; |
| 513 | struct ebpf_target *w; |
| 514 | int update_every = em->update_every; |
| 515 | for (w = root; w; w = w->next) { |
| 516 | if (unlikely(!w->exposed)) |
| 517 | continue; |
| 518 | |
| 519 | ebpf_write_chart_cmd( |
| 520 | NETDATA_APP_FAMILY, |
| 521 | w->clean_name, |
| 522 | NETDATA_OOMKILL_CHART, |
| 523 | "Processes OOM kills.", |
| 524 | EBPF_OOMKILL_UNIT_KILLS, |
| 525 | NETDATA_EBPF_MEMORY_GROUP, |
| 526 | NETDATA_EBPF_CHART_TYPE_STACKED, |
| 527 | "app.ebpf_oomkill", |
| 528 | 20072, |
| 529 | update_every, |
| 530 | NETDATA_EBPF_MODULE_NAME_OOMKILL); |
| 531 | ebpf_create_chart_labels("app_group", w->clean_name, RRDLABEL_SRC_AUTO); |
| 532 | ebpf_commit_label(); |
| 533 | fprintf( |
| 534 | stdout, |
| 535 | "DIMENSION '%s' '' %s 1 1\n", |
| 536 | oomkill_publish_aggregated.dimension, |
| 537 | ebpf_algorithms[NETDATA_EBPF_ABSOLUTE_IDX]); |
| 538 | |
| 539 | w->charts_created |= 1 << EBPF_MODULE_OOMKILL_IDX; |
| 540 | } |
| 541 | |
| 542 | em->apps_charts |= NETDATA_EBPF_APPS_FLAG_CHART_CREATED; |
| 543 | } |
| 544 | |
| 545 | /** |
| 546 | * OOM kill tracking thread. |
| 547 | * |
| 548 | * @param ptr a `ebpf_module_t *`. |
| 549 | * @return always NULL. |
| 550 | */ |
| 551 | void ebpf_oomkill_thread(void *ptr) |
| 552 | { |
| 553 | ebpf_module_t *em = (ebpf_module_t *)ptr; |
| 554 | |
| 555 | CLEANUP_FUNCTION_REGISTER(oomkill_cleanup) cleanup_ptr = em; |
| 556 | |
| 557 | if (!ebpf_module_thread_has_valid_state(em)) { |
| 558 | goto endoomkill; |
| 559 | } |
| 560 | |
| 561 | em->maps = oomkill_maps; |
| 562 | |
| 563 | #define NETDATA_DEFAULT_OOM_DISABLED_MSG "Disabling OOMKILL thread, because" |
| 564 | if (unlikely(!em->apps_charts)) { |
| 565 | // When we are not running integration with apps, we won't fill necessary variables for this thread to run, so |
| 566 | // we need to disable it. |
| 567 | netdata_mutex_lock(&ebpf_exit_cleanup); |
| 568 | if (ebpf_module_enabled_get(em)) |
| 569 | netdata_log_info("%s apps integration is completely disabled.", NETDATA_DEFAULT_OOM_DISABLED_MSG); |
| 570 | netdata_mutex_unlock(&ebpf_exit_cleanup); |
| 571 | |
| 572 | goto endoomkill; |
| 573 | } else if (running_on_kernel < NETDATA_EBPF_KERNEL_4_14) { |
| 574 | netdata_mutex_lock(&ebpf_exit_cleanup); |
| 575 | if (ebpf_module_enabled_get(em)) |
| 576 | netdata_log_info("%s kernel does not have necessary tracepoints.", NETDATA_DEFAULT_OOM_DISABLED_MSG); |
| 577 | netdata_mutex_unlock(&ebpf_exit_cleanup); |
| 578 | |
| 579 | goto endoomkill; |
| 580 | } |
| 581 | |
| 582 | if (ebpf_enable_tracepoints(oomkill_tracepoints) == 0) { |
| 583 | goto endoomkill; |
| 584 | } |
| 585 | |
| 586 | #ifdef LIBBPF_MAJOR_VERSION |
| 587 | ebpf_define_map_type(em->maps, em->maps_per_core, running_on_kernel); |
| 588 | #endif |
| 589 | em->probe_links = ebpf_load_program(ebpf_plugin_dir, em, running_on_kernel, isrh, &em->objects); |
| 590 | if (!em->probe_links) { |
| 591 | goto endoomkill; |
| 592 | } |
| 593 | ebpf_mark_program_loaded(); |
| 594 | |
| 595 | netdata_mutex_lock(&lock); |
| 596 | ebpf_update_stats(&plugin_statistics, em); |
| 597 | ebpf_update_kernel_memory_with_vector(&plugin_statistics, em->maps, EBPF_ACTION_STAT_ADD); |
| 598 | netdata_mutex_unlock(&lock); |
| 599 | |
| 600 | oomkill_collector(em); |
| 601 | |
| 602 | endoomkill: |
| 603 | ebpf_update_disabled_plugin_stats(em); |
| 604 | } |