| 1 | // SPDX-License-Identifier: GPL-3.0-or-later |
| 2 | |
| 3 | #define PULSE_INTERNALS 1 |
| 4 | #include "pulse-workers.h" |
| 5 | |
| 6 | #define WORKERS_MIN_PERCENT_DEFAULT 10000.0 |
| 7 | |
| 8 | struct worker_spinlocks { |
| 9 | size_t locks; |
| 10 | size_t spins; |
| 11 | |
| 12 | RRDDIM *rd_locks; |
| 13 | RRDDIM *rd_spins; |
| 14 | }; |
| 15 | |
| 16 | DEFINE_JUDYL_TYPED(SPINLOCKS, struct worker_spinlocks *); |
| 17 | SPINLOCKS_JudyLSet ALL_SPINLOCKS = { 0 }; |
| 18 | |
| 19 | struct worker_job_type_gs { |
| 20 | STRING *name; |
| 21 | STRING *units; |
| 22 | |
| 23 | struct { |
| 24 | size_t jobs_started; |
| 25 | usec_t busy_time; |
| 26 | } data[2]; |
| 27 | |
| 28 | RRDDIM *rd_jobs_started; |
| 29 | RRDDIM *rd_busy_time; |
| 30 | RRDDIM *rd_avg_time; |
| 31 | |
| 32 | WORKER_METRIC_TYPE type; |
| 33 | NETDATA_DOUBLE min_value; |
| 34 | NETDATA_DOUBLE max_value; |
| 35 | NETDATA_DOUBLE sum_value; |
| 36 | size_t count_value; |
| 37 | |
| 38 | RRDSET *st; |
| 39 | RRDDIM *rd_min; |
| 40 | RRDDIM *rd_max; |
| 41 | RRDDIM *rd_avg; |
| 42 | }; |
| 43 | |
| 44 | struct worker_thread { |
| 45 | pid_t pid; |
| 46 | bool enabled; |
| 47 | |
| 48 | bool cpu_enabled; |
| 49 | double cpu; |
| 50 | |
| 51 | kernel_uint_t utime; |
| 52 | kernel_uint_t stime; |
| 53 | |
| 54 | kernel_uint_t utime_old; |
| 55 | kernel_uint_t stime_old; |
| 56 | |
| 57 | usec_t collected_time; |
| 58 | usec_t collected_time_old; |
| 59 | |
| 60 | size_t jobs_started; |
| 61 | usec_t busy_time; |
| 62 | |
| 63 | struct worker_thread *next; |
| 64 | struct worker_thread *prev; |
| 65 | }; |
| 66 | |
| 67 | struct worker_utilization { |
| 68 | const char *name; |
| 69 | const char *family; |
| 70 | size_t priority; |
| 71 | uint32_t flags; |
| 72 | |
| 73 | char *name_lowercase; |
| 74 | |
| 75 | struct worker_job_type_gs per_job_type[WORKER_UTILIZATION_MAX_JOB_TYPES]; |
| 76 | |
| 77 | size_t workers_max_job_id; |
| 78 | size_t workers_registered; |
| 79 | size_t workers_busy; |
| 80 | usec_t workers_total_busy_time; |
| 81 | usec_t workers_total_duration; |
| 82 | size_t workers_total_jobs_started; |
| 83 | double workers_min_busy_time; |
| 84 | double workers_max_busy_time; |
| 85 | |
| 86 | size_t workers_cpu_registered; |
| 87 | double workers_cpu_min; |
| 88 | double workers_cpu_max; |
| 89 | double workers_cpu_total; |
| 90 | |
| 91 | uint64_t memory_calls[WORKERS_MEMORY_CALL_MAX]; |
| 92 | |
| 93 | struct worker_thread *threads; |
| 94 | |
| 95 | RRDSET *st_workers_time; |
| 96 | RRDDIM *rd_workers_time_avg; |
| 97 | RRDDIM *rd_workers_time_min; |
| 98 | RRDDIM *rd_workers_time_max; |
| 99 | |
| 100 | RRDSET *st_workers_cpu; |
| 101 | RRDDIM *rd_workers_cpu_avg; |
| 102 | RRDDIM *rd_workers_cpu_min; |
| 103 | RRDDIM *rd_workers_cpu_max; |
| 104 | |
| 105 | RRDSET *st_workers_threads; |
| 106 | RRDDIM *rd_workers_threads_free; |
| 107 | RRDDIM *rd_workers_threads_busy; |
| 108 | |
| 109 | RRDSET *st_workers_jobs_per_job_type; |
| 110 | RRDSET *st_workers_time_per_job_type; |
| 111 | RRDSET *st_workers_avg_time_per_job_type; |
| 112 | |
| 113 | RRDDIM *rd_total_cpu_utilizaton; |
| 114 | |
| 115 | RRDSET *st_spinlocks_locks; |
| 116 | RRDSET *st_spinlocks_spins; |
| 117 | SPINLOCKS_JudyLSet spinlocks; |
| 118 | |
| 119 | RRDSET *st_memory_calls; |
| 120 | RRDDIM *rd_memory_calls[WORKERS_MEMORY_CALL_MAX]; |
| 121 | }; |
| 122 | |
| 123 | static inline size_t workers_chart_context_job_name_max_length(const char *worker_name_lowercase, const char *metric_type) { |
| 124 | size_t fixed_length = strlen("netdata.workers.") + strlen(worker_name_lowercase) + strlen(metric_type); |
| 125 | |
| 126 | if(fixed_length >= RRD_ID_LENGTH_MAX) |
| 127 | return 0; |
| 128 | |
| 129 | return RRD_ID_LENGTH_MAX - fixed_length - 1; |
| 130 | } |
| 131 | |
| 132 | static struct worker_utilization all_workers_utilization[] = { |
| 133 | { .name = "PULSE", .family = "workers pulse", .priority = 1000000 }, |
| 134 | { .name = "HEALTH", .family = "workers health alerts", .priority = 1000000 }, |
| 135 | { .name = "MLTRAIN", .family = "workers ML training", .priority = 1000000 }, |
| 136 | { .name = "MLDETECT", .family = "workers ML detection", .priority = 1000000 }, |
| 137 | { .name = "STREAM", .family = "workers streaming", .priority = 1000000 }, |
| 138 | { .name = "STREAMCNT", .family = "workers streaming connect", .priority = 1000000 }, |
| 139 | { .name = "DBENGINE", .family = "workers dbengine instances", .priority = 1000000 }, |
| 140 | { .name = "LIBUV", .family = "workers libuv threadpool", .priority = 1000000 }, |
| 141 | { .name = "WEB", .family = "workers web server", .priority = 1000000 }, |
| 142 | { .name = "ACLK", .family = "workers aclk", .priority = 1000000 }, |
| 143 | { .name = "ACLKSYNC", .family = "workers aclk sync", .priority = 1000000 }, |
| 144 | { .name = "METASYNC", .family = "workers metadata sync", .priority = 1000000 }, |
| 145 | { .name = "PLUGINSD", .family = "workers plugins.d", .priority = 1000000 }, |
| 146 | { .name = "STATSD", .family = "workers plugin statsd", .priority = 1000000 }, |
| 147 | { .name = "STATSDFLUSH", .family = "workers plugin statsd flush", .priority = 1000000 }, |
| 148 | { .name = "PROC", .family = "workers plugin proc", .priority = 1000000 }, |
| 149 | { .name = "WIN", .family = "workers plugin windows", .priority = 1000000 }, |
| 150 | { .name = "NETDEV", .family = "workers plugin proc netdev", .priority = 1000000 }, |
| 151 | { .name = "FREEBSD", .family = "workers plugin freebsd", .priority = 1000000 }, |
| 152 | { .name = "MACOS", .family = "workers plugin macos", .priority = 1000000 }, |
| 153 | { .name = "CGROUPS", .family = "workers plugin cgroups", .priority = 1000000 }, |
| 154 | { .name = "CGROUPSDISC", .family = "workers plugin cgroups find", .priority = 1000000 }, |
| 155 | { .name = "DISKSPACE", .family = "workers plugin diskspace", .priority = 1000000 }, |
| 156 | { .name = "TC", .family = "workers plugin tc", .priority = 1000000 }, |
| 157 | { .name = "TIMEX", .family = "workers plugin timex", .priority = 1000000 }, |
| 158 | { .name = "IDLEJITTER", .family = "workers plugin idlejitter", .priority = 1000000 }, |
| 159 | { .name = "RRDCONTEXT", .family = "workers contexts", .priority = 1000000 }, |
| 160 | { .name = "REPLICATION", .family = "workers replication sender", .priority = 1000000 }, |
| 161 | { .name = "SERVICE", .family = "workers service", .priority = 1000000 }, |
| 162 | { .name = "PROFILER", .family = "workers profile", .priority = 1000000 }, |
| 163 | { .name = "PGCEVICT", .family = "workers dbengine eviction", .priority = 1000000 }, |
| 164 | { .name = "BACKFILL", .family = "workers backfill", .priority = 1000000 }, |
| 165 | { .name = "WEBSOCKET", .family = "workers websocket", .priority = 1000000 }, |
| 166 | |
| 167 | // has to be terminated with a NULL |
| 168 | { .name = NULL, .family = NULL } |
| 169 | }; |
| 170 | |
| 171 | static void workers_total_spinlock_contention_chart(void) { |
| 172 | { |
| 173 | static RRDSET *st = NULL; |
| 174 | |
| 175 | if(unlikely(!st)) { |
| 176 | st = rrdset_create_localhost( |
| 177 | "netdata" |
| 178 | , "spinlock_total_locks" |
| 179 | , NULL |
| 180 | , "spinlocks" |
| 181 | , "netdata.spinlock_total_locks" |
| 182 | , "Netdata Total Spinlock Locks" |
| 183 | , "locks" |
| 184 | , "netdata" |
| 185 | , "pulse" |
| 186 | , 920000 |
| 187 | , localhost->rrd_update_every |
| 188 | , RRDSET_TYPE_LINE |
| 189 | ); |
| 190 | } |
| 191 | |
| 192 | Word_t idx = 0; |
| 193 | for(struct worker_spinlocks *wusp = SPINLOCKS_FIRST(&ALL_SPINLOCKS, &idx); |
| 194 | wusp; |
| 195 | wusp = SPINLOCKS_NEXT(&ALL_SPINLOCKS, &idx)) { |
| 196 | const char *func = (const char *)idx; |
| 197 | RRDDIM *rd = rrddim_find(st, func, false); |
| 198 | if(!rd) rd = rrddim_add(st, func, NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE); |
| 199 | rrddim_set_by_pointer(st, rd, (collected_number)wusp->locks); |
| 200 | } |
| 201 | |
| 202 | rrdset_done(st); |
| 203 | } |
| 204 | |
| 205 | { |
| 206 | static RRDSET *st = NULL; |
| 207 | if(unlikely(!st)) { |
| 208 | st = rrdset_create_localhost( |
| 209 | "netdata" |
| 210 | , "spinlock_total_spins" |
| 211 | , NULL |
| 212 | , "spinlocks" |
| 213 | , "netdata.spinlock_total_spins" |
| 214 | , "Netdata Total Spinlock Spins" |
| 215 | , "spins" |
| 216 | , "netdata" |
| 217 | , "pulse" |
| 218 | , 920001 |
| 219 | , localhost->rrd_update_every |
| 220 | , RRDSET_TYPE_LINE |
| 221 | ); |
| 222 | } |
| 223 | |
| 224 | Word_t idx = 0; |
| 225 | for(struct worker_spinlocks *wusp = SPINLOCKS_FIRST(&ALL_SPINLOCKS, &idx); |
| 226 | wusp; |
| 227 | wusp = SPINLOCKS_NEXT(&ALL_SPINLOCKS, &idx)) { |
| 228 | const char *func = (const char *)idx; |
| 229 | RRDDIM *rd = rrddim_find(st, func, false); |
| 230 | if(!rd) rd = rrddim_add(st, func, NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE); |
| 231 | rrddim_set_by_pointer(st, rd, (collected_number)wusp->spins); |
| 232 | } |
| 233 | |
| 234 | rrdset_done(st); |
| 235 | } |
| 236 | |
| 237 | { |
| 238 | static RRDSET *st = NULL; |
| 239 | if(unlikely(!st)) { |
| 240 | st = rrdset_create_localhost( |
| 241 | "netdata" |
| 242 | , "spinlock_total_spins_per_lock" |
| 243 | , NULL |
| 244 | , "spinlocks" |
| 245 | , "netdata.spinlock_total_spins_per_lock" |
| 246 | , "Netdata Average Spinlock Spins Per Lock" |
| 247 | , "spins" |
| 248 | , "netdata" |
| 249 | , "pulse" |
| 250 | , 920002 |
| 251 | , localhost->rrd_update_every |
| 252 | , RRDSET_TYPE_LINE |
| 253 | ); |
| 254 | } |
| 255 | |
| 256 | Word_t idx = 0; |
| 257 | for(struct worker_spinlocks *wusp = SPINLOCKS_FIRST(&ALL_SPINLOCKS, &idx); |
| 258 | wusp; |
| 259 | wusp = SPINLOCKS_NEXT(&ALL_SPINLOCKS, &idx)) { |
| 260 | const char *func = (const char *)idx; |
| 261 | RRDDIM *rd = rrddim_find(st, func, false); |
| 262 | if(!rd) rd = rrddim_add(st, func, NULL, 1, 10000, RRD_ALGORITHM_ABSOLUTE); |
| 263 | if(!wusp->locks) |
| 264 | rrddim_set_by_pointer(st, rd, 0); |
| 265 | else |
| 266 | rrddim_set_by_pointer(st, rd, (collected_number)((uint64_t)wusp->spins * 10000ULL / (uint64_t)wusp->locks)); |
| 267 | } |
| 268 | |
| 269 | rrdset_done(st); |
| 270 | } |
| 271 | } |
| 272 | |
| 273 | static void workers_total_memory_calls_chart(void) { |
| 274 | { |
| 275 | static RRDSET *st = NULL; |
| 276 | static RRDDIM *rd[WORKERS_MEMORY_CALL_MAX] = { NULL }; |
| 277 | uint64_t memory_calls[WORKERS_MEMORY_CALL_MAX] = { 0 }; |
| 278 | |
| 279 | if(unlikely(!st)) { |
| 280 | st = rrdset_create_localhost( |
| 281 | "netdata" |
| 282 | , "memory_calls_total" |
| 283 | , NULL |
| 284 | , "memory calls" |
| 285 | , "netdata.memory_calls_total" |
| 286 | , "Netdata Total Memory Calls" |
| 287 | , "calls" |
| 288 | , "netdata" |
| 289 | , "pulse" |
| 290 | , 920005 |
| 291 | , localhost->rrd_update_every |
| 292 | , RRDSET_TYPE_LINE |
| 293 | ); |
| 294 | |
| 295 | for (int j = 0; j < WORKERS_MEMORY_CALL_MAX; ++j) |
| 296 | rd[j] = rrddim_add(st, WORKERS_MEMORY_CALL_2str(j), NULL, 1, 1, RRD_ALGORITHM_INCREMENTAL); |
| 297 | } |
| 298 | |
| 299 | for(size_t i = 0; all_workers_utilization[i].name ;i++) { |
| 300 | struct worker_utilization *wu = &all_workers_utilization[i]; |
| 301 | |
| 302 | for (int j = 0; j < WORKERS_MEMORY_CALL_MAX; ++j) |
| 303 | memory_calls[j] += wu->memory_calls[j]; |
| 304 | } |
| 305 | |
| 306 | for (int j = 0; j < WORKERS_MEMORY_CALL_MAX; ++j) |
| 307 | rrddim_set_by_pointer(st, rd[j], (collected_number)memory_calls[j]); |
| 308 | |
| 309 | rrdset_done(st); |
| 310 | } |
| 311 | } |
| 312 | |
| 313 | static void workers_total_cpu_utilization_chart(void) { |
| 314 | size_t i, cpu_enabled = 0; |
| 315 | for(i = 0; all_workers_utilization[i].name ;i++) |
| 316 | if(all_workers_utilization[i].workers_cpu_registered) cpu_enabled++; |
| 317 | |
| 318 | if(!cpu_enabled) return; |
| 319 | |
| 320 | static RRDSET *st = NULL; |
| 321 | |
| 322 | if(!st) { |
| 323 | st = rrdset_create_localhost( |
| 324 | "netdata", |
| 325 | "workers_cpu", |
| 326 | NULL, |
| 327 | "workers", |
| 328 | "netdata.workers.cpu_total", |
| 329 | "Netdata Workers CPU Utilization (100% = 1 core)", |
| 330 | "%", |
| 331 | "netdata", |
| 332 | "pulse", |
| 333 | 999000, |
| 334 | localhost->rrd_update_every, |
| 335 | RRDSET_TYPE_STACKED); |
| 336 | } |
| 337 | |
| 338 | for(i = 0; all_workers_utilization[i].name ;i++) { |
| 339 | struct worker_utilization *wu = &all_workers_utilization[i]; |
| 340 | if(!wu->workers_cpu_registered) continue; |
| 341 | |
| 342 | if(!wu->rd_total_cpu_utilizaton) |
| 343 | wu->rd_total_cpu_utilizaton = rrddim_add(st, wu->name_lowercase, NULL, 1, 100, RRD_ALGORITHM_ABSOLUTE); |
| 344 | |
| 345 | rrddim_set_by_pointer(st, wu->rd_total_cpu_utilizaton, (collected_number)((double)wu->workers_cpu_total * 100.0)); |
| 346 | } |
| 347 | |
| 348 | rrdset_done(st); |
| 349 | } |
| 350 | |
| 351 | #define WORKER_CHART_DECIMAL_PRECISION 100 |
| 352 | |
| 353 | static void workers_utilization_update_chart(struct worker_utilization *wu) { |
| 354 | if(!wu->workers_registered) return; |
| 355 | |
| 356 | //fprintf(stderr, "%-12s WORKER UTILIZATION: %-3.2f%%, %zu jobs done, %zu running, on %zu workers, min %-3.02f%%, max %-3.02f%%.\n", |
| 357 | // wu->name, |
| 358 | // (double)wu->workers_total_busy_time * 100.0 / (double)wu->workers_total_duration, |
| 359 | // wu->workers_total_jobs_started, wu->workers_busy, wu->workers_registered, |
| 360 | // wu->workers_min_busy_time, wu->workers_max_busy_time); |
| 361 | |
| 362 | // ---------------------------------------------------------------------- |
| 363 | |
| 364 | if(unlikely(!wu->st_workers_time)) { |
| 365 | char name[RRD_ID_LENGTH_MAX + 1]; |
| 366 | snprintfz(name, RRD_ID_LENGTH_MAX, "workers_time_%s", wu->name_lowercase); |
| 367 | |
| 368 | char context[RRD_ID_LENGTH_MAX + 1]; |
| 369 | snprintf(context, RRD_ID_LENGTH_MAX, "netdata.workers.%s.time", wu->name_lowercase); |
| 370 | |
| 371 | wu->st_workers_time = rrdset_create_localhost( |
| 372 | "netdata" |
| 373 | , name |
| 374 | , NULL |
| 375 | , wu->family |
| 376 | , context |
| 377 | , "Netdata Workers Busy Time (100% = all workers busy)" |
| 378 | , "%" |
| 379 | , "netdata" |
| 380 | , "pulse" |
| 381 | , wu->priority |
| 382 | , localhost->rrd_update_every |
| 383 | , RRDSET_TYPE_AREA |
| 384 | ); |
| 385 | } |
| 386 | |
| 387 | // we add the min and max dimensions only when we have multiple workers |
| 388 | |
| 389 | if(unlikely(!wu->rd_workers_time_min && wu->workers_registered > 1)) |
| 390 | wu->rd_workers_time_min = rrddim_add(wu->st_workers_time, "min", NULL, 1, WORKER_CHART_DECIMAL_PRECISION, RRD_ALGORITHM_ABSOLUTE); |
| 391 | |
| 392 | if(unlikely(!wu->rd_workers_time_max && wu->workers_registered > 1)) |
| 393 | wu->rd_workers_time_max = rrddim_add(wu->st_workers_time, "max", NULL, 1, WORKER_CHART_DECIMAL_PRECISION, RRD_ALGORITHM_ABSOLUTE); |
| 394 | |
| 395 | if(unlikely(!wu->rd_workers_time_avg)) |
| 396 | wu->rd_workers_time_avg = rrddim_add(wu->st_workers_time, "average", NULL, 1, WORKER_CHART_DECIMAL_PRECISION, RRD_ALGORITHM_ABSOLUTE); |
| 397 | |
| 398 | if(unlikely(wu->workers_min_busy_time == WORKERS_MIN_PERCENT_DEFAULT)) wu->workers_min_busy_time = 0.0; |
| 399 | |
| 400 | if(wu->rd_workers_time_min) |
| 401 | rrddim_set_by_pointer(wu->st_workers_time, wu->rd_workers_time_min, (collected_number)((double)wu->workers_min_busy_time * WORKER_CHART_DECIMAL_PRECISION)); |
| 402 | |
| 403 | if(wu->rd_workers_time_max) |
| 404 | rrddim_set_by_pointer(wu->st_workers_time, wu->rd_workers_time_max, (collected_number)((double)wu->workers_max_busy_time * WORKER_CHART_DECIMAL_PRECISION)); |
| 405 | |
| 406 | if(wu->workers_total_duration == 0) |
| 407 | rrddim_set_by_pointer(wu->st_workers_time, wu->rd_workers_time_avg, 0); |
| 408 | else |
| 409 | rrddim_set_by_pointer(wu->st_workers_time, wu->rd_workers_time_avg, (collected_number)((double)wu->workers_total_busy_time * 100.0 * WORKER_CHART_DECIMAL_PRECISION / (double)wu->workers_total_duration)); |
| 410 | |
| 411 | rrdset_done(wu->st_workers_time); |
| 412 | |
| 413 | // ---------------------------------------------------------------------- |
| 414 | |
| 415 | #ifdef __linux__ |
| 416 | if(wu->workers_cpu_registered || wu->st_workers_cpu) { |
| 417 | if(unlikely(!wu->st_workers_cpu)) { |
| 418 | char name[RRD_ID_LENGTH_MAX + 1]; |
| 419 | snprintfz(name, RRD_ID_LENGTH_MAX, "workers_cpu_%s", wu->name_lowercase); |
| 420 | |
| 421 | char context[RRD_ID_LENGTH_MAX + 1]; |
| 422 | snprintf(context, RRD_ID_LENGTH_MAX, "netdata.workers.%s.cpu", wu->name_lowercase); |
| 423 | |
| 424 | wu->st_workers_cpu = rrdset_create_localhost( |
| 425 | "netdata" |
| 426 | , name |
| 427 | , NULL |
| 428 | , wu->family |
| 429 | , context |
| 430 | , "Netdata Workers CPU Utilization (100% = all workers busy)" |
| 431 | , "%" |
| 432 | , "netdata" |
| 433 | , "pulse" |
| 434 | , wu->priority + 1 |
| 435 | , localhost->rrd_update_every |
| 436 | , RRDSET_TYPE_AREA |
| 437 | ); |
| 438 | } |
| 439 | |
| 440 | if (unlikely(!wu->rd_workers_cpu_min && wu->workers_registered > 1)) |
| 441 | wu->rd_workers_cpu_min = rrddim_add(wu->st_workers_cpu, "min", NULL, 1, WORKER_CHART_DECIMAL_PRECISION, RRD_ALGORITHM_ABSOLUTE); |
| 442 | |
| 443 | if (unlikely(!wu->rd_workers_cpu_max && wu->workers_registered > 1)) |
| 444 | wu->rd_workers_cpu_max = rrddim_add(wu->st_workers_cpu, "max", NULL, 1, WORKER_CHART_DECIMAL_PRECISION, RRD_ALGORITHM_ABSOLUTE); |
| 445 | |
| 446 | if(unlikely(!wu->rd_workers_cpu_avg)) |
| 447 | wu->rd_workers_cpu_avg = rrddim_add(wu->st_workers_cpu, "average", NULL, 1, WORKER_CHART_DECIMAL_PRECISION, RRD_ALGORITHM_ABSOLUTE); |
| 448 | |
| 449 | if(unlikely(wu->workers_cpu_min == WORKERS_MIN_PERCENT_DEFAULT)) wu->workers_cpu_min = 0.0; |
| 450 | |
| 451 | if(wu->rd_workers_cpu_min) |
| 452 | rrddim_set_by_pointer(wu->st_workers_cpu, wu->rd_workers_cpu_min, (collected_number)(wu->workers_cpu_min * WORKER_CHART_DECIMAL_PRECISION)); |
| 453 | |
| 454 | if(wu->rd_workers_cpu_max) |
| 455 | rrddim_set_by_pointer(wu->st_workers_cpu, wu->rd_workers_cpu_max, (collected_number)(wu->workers_cpu_max * WORKER_CHART_DECIMAL_PRECISION)); |
| 456 | |
| 457 | if(wu->workers_cpu_registered == 0) |
| 458 | rrddim_set_by_pointer(wu->st_workers_cpu, wu->rd_workers_cpu_avg, 0); |
| 459 | else |
| 460 | rrddim_set_by_pointer(wu->st_workers_cpu, wu->rd_workers_cpu_avg, (collected_number)( wu->workers_cpu_total * WORKER_CHART_DECIMAL_PRECISION / (NETDATA_DOUBLE)wu->workers_cpu_registered )); |
| 461 | |
| 462 | rrdset_done(wu->st_workers_cpu); |
| 463 | } |
| 464 | #endif |
| 465 | |
| 466 | // ---------------------------------------------------------------------------------------------------------------- |
| 467 | |
| 468 | if(unlikely(!wu->st_workers_jobs_per_job_type)) { |
| 469 | char name[RRD_ID_LENGTH_MAX + 1]; |
| 470 | snprintfz(name, RRD_ID_LENGTH_MAX, "workers_jobs_by_type_%s", wu->name_lowercase); |
| 471 | |
| 472 | char context[RRD_ID_LENGTH_MAX + 1]; |
| 473 | snprintf(context, RRD_ID_LENGTH_MAX, "netdata.workers.%s.jobs_started_by_type", wu->name_lowercase); |
| 474 | |
| 475 | wu->st_workers_jobs_per_job_type = rrdset_create_localhost( |
| 476 | "netdata" |
| 477 | , name |
| 478 | , NULL |
| 479 | , wu->family |
| 480 | , context |
| 481 | , "Netdata Workers Jobs Started by Type" |
| 482 | , "jobs" |
| 483 | , "netdata" |
| 484 | , "pulse" |
| 485 | , wu->priority + 2 |
| 486 | , localhost->rrd_update_every |
| 487 | , RRDSET_TYPE_STACKED |
| 488 | ); |
| 489 | } |
| 490 | |
| 491 | { |
| 492 | size_t i; |
| 493 | for(i = 0; i <= wu->workers_max_job_id ;i++) { |
| 494 | if(unlikely(wu->per_job_type[i].type != WORKER_METRIC_IDLE_BUSY)) |
| 495 | continue; |
| 496 | |
| 497 | if (wu->per_job_type[i].name) { |
| 498 | |
| 499 | if(unlikely(!wu->per_job_type[i].rd_jobs_started)) |
| 500 | wu->per_job_type[i].rd_jobs_started = rrddim_add(wu->st_workers_jobs_per_job_type, string2str(wu->per_job_type[i].name), NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE); |
| 501 | |
| 502 | rrddim_set_by_pointer(wu->st_workers_jobs_per_job_type, wu->per_job_type[i].rd_jobs_started, (collected_number)(wu->per_job_type[i].data[0].jobs_started)); |
| 503 | } |
| 504 | } |
| 505 | } |
| 506 | |
| 507 | rrdset_done(wu->st_workers_jobs_per_job_type); |
| 508 | |
| 509 | // ---------------------------------------------------------------------------------------------------------------- |
| 510 | |
| 511 | if(unlikely(!wu->st_workers_time_per_job_type)) { |
| 512 | char name[RRD_ID_LENGTH_MAX + 1]; |
| 513 | snprintfz(name, RRD_ID_LENGTH_MAX, "workers_busy_time_by_type_%s", wu->name_lowercase); |
| 514 | |
| 515 | char context[RRD_ID_LENGTH_MAX + 1]; |
| 516 | snprintf(context, RRD_ID_LENGTH_MAX, "netdata.workers.%s.time_by_type", wu->name_lowercase); |
| 517 | |
| 518 | wu->st_workers_time_per_job_type = rrdset_create_localhost( |
| 519 | "netdata" |
| 520 | , name |
| 521 | , NULL |
| 522 | , wu->family |
| 523 | , context |
| 524 | , "Netdata Workers Busy Time by Type" |
| 525 | , "ms" |
| 526 | , "netdata" |
| 527 | , "pulse" |
| 528 | , wu->priority + 3 |
| 529 | , localhost->rrd_update_every |
| 530 | , RRDSET_TYPE_STACKED |
| 531 | ); |
| 532 | } |
| 533 | |
| 534 | { |
| 535 | size_t i; |
| 536 | for(i = 0; i <= wu->workers_max_job_id ;i++) { |
| 537 | if(unlikely(wu->per_job_type[i].type != WORKER_METRIC_IDLE_BUSY)) |
| 538 | continue; |
| 539 | |
| 540 | if (wu->per_job_type[i].name) { |
| 541 | |
| 542 | if(unlikely(!wu->per_job_type[i].rd_busy_time)) |
| 543 | wu->per_job_type[i].rd_busy_time = rrddim_add(wu->st_workers_time_per_job_type, string2str(wu->per_job_type[i].name), NULL, 1, USEC_PER_MS, RRD_ALGORITHM_ABSOLUTE); |
| 544 | |
| 545 | rrddim_set_by_pointer(wu->st_workers_time_per_job_type, wu->per_job_type[i].rd_busy_time, (collected_number)(wu->per_job_type[i].data[0].busy_time)); |
| 546 | } |
| 547 | } |
| 548 | } |
| 549 | |
| 550 | rrdset_done(wu->st_workers_time_per_job_type); |
| 551 | |
| 552 | // ---------------------------------------------------------------------------------------------------------------- |
| 553 | |
| 554 | if(unlikely(!wu->st_workers_avg_time_per_job_type)) { |
| 555 | char name[RRD_ID_LENGTH_MAX + 1]; |
| 556 | snprintfz(name, RRD_ID_LENGTH_MAX, "workers_avg_time_by_type_%s", wu->name_lowercase); |
| 557 | |
| 558 | char context[RRD_ID_LENGTH_MAX + 1]; |
| 559 | snprintf(context, RRD_ID_LENGTH_MAX, "netdata.workers.%s.avg_time_by_type", wu->name_lowercase); |
| 560 | |
| 561 | wu->st_workers_avg_time_per_job_type = rrdset_create_localhost( |
| 562 | "netdata" |
| 563 | , name |
| 564 | , NULL |
| 565 | , wu->family |
| 566 | , context |
| 567 | , "Netdata Workers Average Time by Type" |
| 568 | , "ms" |
| 569 | , "netdata" |
| 570 | , "pulse" |
| 571 | , wu->priority + 4 |
| 572 | , localhost->rrd_update_every |
| 573 | , RRDSET_TYPE_STACKED |
| 574 | ); |
| 575 | } |
| 576 | |
| 577 | { |
| 578 | size_t i; |
| 579 | for(i = 0; i <= wu->workers_max_job_id ;i++) { |
| 580 | if(unlikely(wu->per_job_type[i].type != WORKER_METRIC_IDLE_BUSY)) |
| 581 | continue; |
| 582 | |
| 583 | if (wu->per_job_type[i].name) { |
| 584 | |
| 585 | if(unlikely(!wu->per_job_type[i].rd_avg_time)) |
| 586 | wu->per_job_type[i].rd_avg_time = rrddim_add(wu->st_workers_avg_time_per_job_type, string2str(wu->per_job_type[i].name), NULL, 1, USEC_PER_MS, RRD_ALGORITHM_ABSOLUTE); |
| 587 | |
| 588 | ssize_t jobs_delta = (ssize_t)wu->per_job_type[i].data[0].jobs_started; |
| 589 | susec_t time_delta = (susec_t)wu->per_job_type[i].data[0].busy_time; |
| 590 | susec_t average = (jobs_delta != 0) ? time_delta / jobs_delta : 0; |
| 591 | rrddim_set_by_pointer(wu->st_workers_avg_time_per_job_type, wu->per_job_type[i].rd_avg_time, (collected_number)average); |
| 592 | } |
| 593 | } |
| 594 | } |
| 595 | |
| 596 | rrdset_done(wu->st_workers_avg_time_per_job_type); |
| 597 | |
| 598 | // ---------------------------------------------------------------------------------------------------------------- |
| 599 | |
| 600 | if(wu->st_workers_threads || wu->workers_registered > 1) { |
| 601 | if(unlikely(!wu->st_workers_threads)) { |
| 602 | char name[RRD_ID_LENGTH_MAX + 1]; |
| 603 | snprintfz(name, RRD_ID_LENGTH_MAX, "workers_threads_%s", wu->name_lowercase); |
| 604 | |
| 605 | char context[RRD_ID_LENGTH_MAX + 1]; |
| 606 | snprintf(context, RRD_ID_LENGTH_MAX, "netdata.workers.%s.threads", wu->name_lowercase); |
| 607 | |
| 608 | wu->st_workers_threads = rrdset_create_localhost( |
| 609 | "netdata" |
| 610 | , name |
| 611 | , NULL |
| 612 | , wu->family |
| 613 | , context |
| 614 | , "Netdata Workers Threads" |
| 615 | , "threads" |
| 616 | , "netdata" |
| 617 | , "pulse" |
| 618 | , wu->priority + 5 |
| 619 | , localhost->rrd_update_every |
| 620 | , RRDSET_TYPE_STACKED |
| 621 | ); |
| 622 | |
| 623 | wu->rd_workers_threads_free = rrddim_add(wu->st_workers_threads, "free", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE); |
| 624 | wu->rd_workers_threads_busy = rrddim_add(wu->st_workers_threads, "busy", NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE); |
| 625 | } |
| 626 | |
| 627 | rrddim_set_by_pointer(wu->st_workers_threads, wu->rd_workers_threads_free, (collected_number)(wu->workers_registered - wu->workers_busy)); |
| 628 | rrddim_set_by_pointer(wu->st_workers_threads, wu->rd_workers_threads_busy, (collected_number)(wu->workers_busy)); |
| 629 | rrdset_done(wu->st_workers_threads); |
| 630 | } |
| 631 | |
| 632 | // ---------------------------------------------------------------------- |
| 633 | // spinlocks |
| 634 | |
| 635 | { |
| 636 | if(unlikely(!wu->st_spinlocks_locks)) { |
| 637 | char name[RRD_ID_LENGTH_MAX + 1]; |
| 638 | snprintfz(name, RRD_ID_LENGTH_MAX, "workers_spinlock_locks_%s", wu->name_lowercase); |
| 639 | |
| 640 | char context[RRD_ID_LENGTH_MAX + 1]; |
| 641 | snprintf(context, RRD_ID_LENGTH_MAX, "netdata.workers.%s.spinlock_locks", wu->name_lowercase); |
| 642 | |
| 643 | wu->st_spinlocks_locks = rrdset_create_localhost( |
| 644 | "netdata" |
| 645 | , name |
| 646 | , NULL |
| 647 | , wu->family |
| 648 | , context |
| 649 | , "Netdata Spinlock Locks" |
| 650 | , "locks" |
| 651 | , "netdata" |
| 652 | , "pulse" |
| 653 | , wu->priority + 6 |
| 654 | , localhost->rrd_update_every |
| 655 | , RRDSET_TYPE_LINE |
| 656 | ); |
| 657 | } |
| 658 | |
| 659 | Word_t idx = 0; |
| 660 | for(struct worker_spinlocks *wusp = SPINLOCKS_FIRST(&wu->spinlocks, &idx); |
| 661 | wusp; |
| 662 | wusp = SPINLOCKS_NEXT(&wu->spinlocks, &idx)) { |
| 663 | const char *func = (const char *)idx; |
| 664 | if(!wusp->rd_locks) |
| 665 | wusp->rd_locks = rrddim_add(wu->st_spinlocks_locks, func, NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE); |
| 666 | |
| 667 | rrddim_set_by_pointer(wu->st_spinlocks_locks, wusp->rd_locks, (collected_number)wusp->locks); |
| 668 | } |
| 669 | |
| 670 | rrdset_done(wu->st_spinlocks_locks); |
| 671 | } |
| 672 | |
| 673 | { |
| 674 | if(unlikely(!wu->st_spinlocks_spins)) { |
| 675 | char name[RRD_ID_LENGTH_MAX + 1]; |
| 676 | snprintfz(name, RRD_ID_LENGTH_MAX, "workers_spinlock_spins_%s", wu->name_lowercase); |
| 677 | |
| 678 | char context[RRD_ID_LENGTH_MAX + 1]; |
| 679 | snprintf(context, RRD_ID_LENGTH_MAX, "netdata.workers.%s.spinlock_spins", wu->name_lowercase); |
| 680 | |
| 681 | wu->st_spinlocks_spins = rrdset_create_localhost( |
| 682 | "netdata" |
| 683 | , name |
| 684 | , NULL |
| 685 | , wu->family |
| 686 | , context |
| 687 | , "Netdata Spinlock Spins" |
| 688 | , "spins" |
| 689 | , "netdata" |
| 690 | , "pulse" |
| 691 | , wu->priority + 7 |
| 692 | , localhost->rrd_update_every |
| 693 | , RRDSET_TYPE_LINE |
| 694 | ); |
| 695 | } |
| 696 | |
| 697 | Word_t idx = 0; |
| 698 | for(struct worker_spinlocks *wusp = SPINLOCKS_FIRST(&wu->spinlocks, &idx); |
| 699 | wusp; |
| 700 | wusp = SPINLOCKS_NEXT(&wu->spinlocks, &idx)) { |
| 701 | const char *func = (const char *)idx; |
| 702 | if(!wusp->rd_spins) |
| 703 | wusp->rd_spins = rrddim_add(wu->st_spinlocks_spins, func, NULL, 1, 1, RRD_ALGORITHM_ABSOLUTE); |
| 704 | |
| 705 | rrddim_set_by_pointer(wu->st_spinlocks_spins, wusp->rd_spins, (collected_number)wusp->spins); |
| 706 | } |
| 707 | |
| 708 | rrdset_done(wu->st_spinlocks_spins); |
| 709 | } |
| 710 | |
| 711 | // ---------------------------------------------------------------------- |
| 712 | // memory calls |
| 713 | |
| 714 | { |
| 715 | if(unlikely(!wu->st_memory_calls)) { |
| 716 | char name[RRD_ID_LENGTH_MAX + 1]; |
| 717 | snprintfz(name, RRD_ID_LENGTH_MAX, "workers_memory_calls_%s", wu->name_lowercase); |
| 718 | |
| 719 | char context[RRD_ID_LENGTH_MAX + 1]; |
| 720 | snprintf(context, RRD_ID_LENGTH_MAX, "netdata.workers.%s.memory_calls", wu->name_lowercase); |
| 721 | |
| 722 | wu->st_memory_calls = rrdset_create_localhost( |
| 723 | "netdata" |
| 724 | , name |
| 725 | , NULL |
| 726 | , wu->family |
| 727 | , context |
| 728 | , "Netdata Memory Calls" |
| 729 | , "calls" |
| 730 | , "netdata" |
| 731 | , "pulse" |
| 732 | , wu->priority + 8 |
| 733 | , localhost->rrd_update_every |
| 734 | , RRDSET_TYPE_LINE |
| 735 | ); |
| 736 | } |
| 737 | |
| 738 | for(size_t i = 0; i < WORKERS_MEMORY_CALL_MAX; i++) { |
| 739 | if(!wu->rd_memory_calls[i]) |
| 740 | wu->rd_memory_calls[i] = rrddim_add(wu->st_memory_calls, WORKERS_MEMORY_CALL_2str(i), NULL, 1, 1, RRD_ALGORITHM_INCREMENTAL); |
| 741 | |
| 742 | rrddim_set_by_pointer(wu->st_memory_calls, wu->rd_memory_calls[i], (collected_number)wu->memory_calls[i]); |
| 743 | } |
| 744 | |
| 745 | rrdset_done(wu->st_memory_calls); |
| 746 | } |
| 747 | |
| 748 | // ---------------------------------------------------------------------- |
| 749 | // custom metric types WORKER_METRIC_ABSOLUTE |
| 750 | |
| 751 | { |
| 752 | size_t i; |
| 753 | for (i = 0; i <= wu->workers_max_job_id; i++) { |
| 754 | if(wu->per_job_type[i].type != WORKER_METRIC_ABSOLUTE) |
| 755 | continue; |
| 756 | |
| 757 | if(!wu->per_job_type[i].count_value) |
| 758 | continue; |
| 759 | |
| 760 | if(!wu->per_job_type[i].st) { |
| 761 | size_t job_name_len = string_strlen(wu->per_job_type[i].name); |
| 762 | size_t job_name_max_len = workers_chart_context_job_name_max_length(wu->name_lowercase, ".value."); |
| 763 | if(job_name_len > job_name_max_len) job_name_len = job_name_max_len; |
| 764 | |
| 765 | char job_name_sanitized[RRD_ID_LENGTH_MAX + 1]; |
| 766 | rrdset_strncpyz_name(job_name_sanitized, string2str(wu->per_job_type[i].name), job_name_len); |
| 767 | |
| 768 | char name[RRD_ID_LENGTH_MAX + 1]; |
| 769 | snprintfz(name, RRD_ID_LENGTH_MAX, "workers_%s_value_%s", wu->name_lowercase, job_name_sanitized); |
| 770 | |
| 771 | char context[RRD_ID_LENGTH_MAX + 1]; |
| 772 | size_t context_len = (size_t)snprintfz(context, RRD_ID_LENGTH_MAX + 1, "netdata.workers.%s.value.", wu->name_lowercase); |
| 773 | strcatz(context, context_len, job_name_sanitized, sizeof(context)); |
| 774 | |
| 775 | char title[1000 + 1]; |
| 776 | snprintf(title, 1000, "Netdata Workers %s value of %s", wu->name_lowercase, string2str(wu->per_job_type[i].name)); |
| 777 | |
| 778 | wu->per_job_type[i].st = rrdset_create_localhost( |
| 779 | "netdata" |
| 780 | , name |
| 781 | , NULL |
| 782 | , wu->family |
| 783 | , context |
| 784 | , title |
| 785 | , (wu->per_job_type[i].units)?string2str(wu->per_job_type[i].units):"value" |
| 786 | , "netdata" |
| 787 | , "pulse" |
| 788 | , wu->priority + 10 + i |
| 789 | , localhost->rrd_update_every |
| 790 | , RRDSET_TYPE_LINE |
| 791 | ); |
| 792 | |
| 793 | wu->per_job_type[i].rd_min = rrddim_add(wu->per_job_type[i].st, "min", NULL, 1, WORKER_CHART_DECIMAL_PRECISION, RRD_ALGORITHM_ABSOLUTE); |
| 794 | wu->per_job_type[i].rd_max = rrddim_add(wu->per_job_type[i].st, "max", NULL, 1, WORKER_CHART_DECIMAL_PRECISION, RRD_ALGORITHM_ABSOLUTE); |
| 795 | wu->per_job_type[i].rd_avg = rrddim_add(wu->per_job_type[i].st, "average", NULL, 1, WORKER_CHART_DECIMAL_PRECISION, RRD_ALGORITHM_ABSOLUTE); |
| 796 | } |
| 797 | |
| 798 | rrddim_set_by_pointer(wu->per_job_type[i].st, wu->per_job_type[i].rd_min, (collected_number)(wu->per_job_type[i].min_value * WORKER_CHART_DECIMAL_PRECISION)); |
| 799 | rrddim_set_by_pointer(wu->per_job_type[i].st, wu->per_job_type[i].rd_max, (collected_number)(wu->per_job_type[i].max_value * WORKER_CHART_DECIMAL_PRECISION)); |
| 800 | rrddim_set_by_pointer(wu->per_job_type[i].st, wu->per_job_type[i].rd_avg, (collected_number)(wu->per_job_type[i].sum_value / wu->per_job_type[i].count_value * WORKER_CHART_DECIMAL_PRECISION)); |
| 801 | |
| 802 | rrdset_done(wu->per_job_type[i].st); |
| 803 | } |
| 804 | } |
| 805 | |
| 806 | // ---------------------------------------------------------------------- |
| 807 | // custom metric types WORKER_METRIC_INCREMENTAL |
| 808 | |
| 809 | { |
| 810 | size_t i; |
| 811 | for (i = 0; i <= wu->workers_max_job_id ; i++) { |
| 812 | if(wu->per_job_type[i].type != WORKER_METRIC_INCREMENT && wu->per_job_type[i].type != WORKER_METRIC_INCREMENTAL_TOTAL) |
| 813 | continue; |
| 814 | |
| 815 | if(!wu->per_job_type[i].count_value) |
| 816 | continue; |
| 817 | |
| 818 | if(!wu->per_job_type[i].st) { |
| 819 | size_t job_name_len = string_strlen(wu->per_job_type[i].name); |
| 820 | size_t job_name_max_len = workers_chart_context_job_name_max_length(wu->name_lowercase, ".rate."); |
| 821 | if(job_name_len > job_name_max_len) job_name_len = job_name_max_len; |
| 822 | |
| 823 | char job_name_sanitized[RRD_ID_LENGTH_MAX + 1]; |
| 824 | rrdset_strncpyz_name(job_name_sanitized, string2str(wu->per_job_type[i].name), job_name_len); |
| 825 | |
| 826 | char name[RRD_ID_LENGTH_MAX + 1]; |
| 827 | snprintfz(name, RRD_ID_LENGTH_MAX, "workers_%s_rate_%s", wu->name_lowercase, job_name_sanitized); |
| 828 | |
| 829 | char context[RRD_ID_LENGTH_MAX + 1]; |
| 830 | size_t context_len = (size_t)snprintfz(context, RRD_ID_LENGTH_MAX + 1, "netdata.workers.%s.rate.", wu->name_lowercase); |
| 831 | strcatz(context, context_len, job_name_sanitized, sizeof(context)); |
| 832 | |
| 833 | char title[1000 + 1]; |
| 834 | snprintf(title, 1000, "Netdata Workers %s rate of %s", wu->name_lowercase, string2str(wu->per_job_type[i].name)); |
| 835 | |
| 836 | wu->per_job_type[i].st = rrdset_create_localhost( |
| 837 | "netdata" |
| 838 | , name |
| 839 | , NULL |
| 840 | , wu->family |
| 841 | , context |
| 842 | , title |
| 843 | , (wu->per_job_type[i].units)?string2str(wu->per_job_type[i].units):"rate" |
| 844 | , "netdata" |
| 845 | , "pulse" |
| 846 | , wu->priority + 10 + i |
| 847 | , localhost->rrd_update_every |
| 848 | , RRDSET_TYPE_LINE |
| 849 | ); |
| 850 | |
| 851 | wu->per_job_type[i].rd_min = rrddim_add(wu->per_job_type[i].st, "min", NULL, 1, WORKER_CHART_DECIMAL_PRECISION, RRD_ALGORITHM_ABSOLUTE); |
| 852 | wu->per_job_type[i].rd_max = rrddim_add(wu->per_job_type[i].st, "max", NULL, 1, WORKER_CHART_DECIMAL_PRECISION, RRD_ALGORITHM_ABSOLUTE); |
| 853 | wu->per_job_type[i].rd_avg = rrddim_add(wu->per_job_type[i].st, "average", NULL, 1, WORKER_CHART_DECIMAL_PRECISION, RRD_ALGORITHM_ABSOLUTE); |
| 854 | } |
| 855 | |
| 856 | rrddim_set_by_pointer(wu->per_job_type[i].st, wu->per_job_type[i].rd_min, (collected_number)(wu->per_job_type[i].min_value * WORKER_CHART_DECIMAL_PRECISION)); |
| 857 | rrddim_set_by_pointer(wu->per_job_type[i].st, wu->per_job_type[i].rd_max, (collected_number)(wu->per_job_type[i].max_value * WORKER_CHART_DECIMAL_PRECISION)); |
| 858 | rrddim_set_by_pointer(wu->per_job_type[i].st, wu->per_job_type[i].rd_avg, (collected_number)(wu->per_job_type[i].sum_value / wu->per_job_type[i].count_value * WORKER_CHART_DECIMAL_PRECISION)); |
| 859 | |
| 860 | rrdset_done(wu->per_job_type[i].st); |
| 861 | } |
| 862 | } |
| 863 | } |
| 864 | |
| 865 | static void workers_utilization_reset_statistics(struct worker_utilization *wu) { |
| 866 | Word_t idx = 0; |
| 867 | for(struct worker_spinlocks *wusp = SPINLOCKS_FIRST(&wu->spinlocks, &idx); |
| 868 | wusp; |
| 869 | wusp = SPINLOCKS_NEXT(&wu->spinlocks, &idx)) { |
| 870 | wusp->locks = 0; |
| 871 | wusp->spins = 0; |
| 872 | } |
| 873 | |
| 874 | wu->workers_registered = 0; |
| 875 | wu->workers_busy = 0; |
| 876 | wu->workers_total_busy_time = 0; |
| 877 | wu->workers_total_duration = 0; |
| 878 | wu->workers_total_jobs_started = 0; |
| 879 | wu->workers_min_busy_time = WORKERS_MIN_PERCENT_DEFAULT; |
| 880 | wu->workers_max_busy_time = 0; |
| 881 | |
| 882 | wu->workers_cpu_registered = 0; |
| 883 | wu->workers_cpu_min = WORKERS_MIN_PERCENT_DEFAULT; |
| 884 | wu->workers_cpu_max = 0; |
| 885 | wu->workers_cpu_total = 0; |
| 886 | |
| 887 | size_t i; |
| 888 | for(i = 0; i < WORKER_UTILIZATION_MAX_JOB_TYPES ;i++) { |
| 889 | if(unlikely(!wu->name_lowercase)) { |
| 890 | wu->name_lowercase = strdupz(wu->name); |
| 891 | char *s = wu->name_lowercase; |
| 892 | for( ; *s ; s++) *s = tolower(*s); |
| 893 | } |
| 894 | |
| 895 | // copy the usage data |
| 896 | wu->per_job_type[i].data[1].jobs_started = wu->per_job_type[i].data[0].jobs_started; |
| 897 | wu->per_job_type[i].data[1].busy_time = wu->per_job_type[i].data[0].busy_time; |
| 898 | |
| 899 | // reset them for the next collection |
| 900 | wu->per_job_type[i].data[0].jobs_started = 0; |
| 901 | wu->per_job_type[i].data[0].busy_time = 0; |
| 902 | |
| 903 | wu->per_job_type[i].min_value = NAN; |
| 904 | wu->per_job_type[i].max_value = NAN; |
| 905 | wu->per_job_type[i].sum_value = NAN; |
| 906 | wu->per_job_type[i].count_value = 0; |
| 907 | } |
| 908 | |
| 909 | struct worker_thread *wt; |
| 910 | for(wt = wu->threads; wt ; wt = wt->next) { |
| 911 | wt->enabled = false; |
| 912 | wt->cpu_enabled = false; |
| 913 | } |
| 914 | |
| 915 | memset(wu->memory_calls, 0, sizeof(wu->memory_calls)); |
| 916 | } |
| 917 | |
| 918 | #define TASK_STAT_PREFIX "/proc/self/task/" |
| 919 | #define TASK_STAT_SUFFIX "/stat" |
| 920 | |
| 921 | static int read_thread_cpu_time_from_proc_stat(pid_t pid __maybe_unused, kernel_uint_t *utime __maybe_unused, kernel_uint_t *stime __maybe_unused) { |
| 922 | #ifdef __linux__ |
| 923 | static char filename[sizeof(TASK_STAT_PREFIX) + sizeof(TASK_STAT_SUFFIX) + 20] = TASK_STAT_PREFIX; |
| 924 | static size_t start_pos = sizeof(TASK_STAT_PREFIX) - 1; |
| 925 | static procfile *ff = NULL; |
| 926 | |
| 927 | // construct the filename |
| 928 | size_t end_pos = snprintfz(&filename[start_pos], 20, "%d", pid); |
| 929 | strcpy(&filename[start_pos + end_pos], TASK_STAT_SUFFIX); |
| 930 | |
| 931 | // (re)open the procfile to the new filename |
| 932 | bool set_quotes = (ff == NULL) ? true : false; |
| 933 | ff = procfile_reopen(ff, filename, NULL, PROCFILE_FLAG_ERROR_ON_ERROR_LOG); |
| 934 | if(unlikely(!ff)) return -1; |
| 935 | |
| 936 | if(set_quotes) |
| 937 | procfile_set_open_close(ff, "(", ")"); |
| 938 | |
| 939 | // read the entire file and split it to lines and words |
| 940 | ff = procfile_readall(ff); |
| 941 | if(unlikely(!ff)) return -1; |
| 942 | |
| 943 | // parse the numbers we are interested |
| 944 | *utime = str2kernel_uint_t(procfile_lineword(ff, 0, 13)); |
| 945 | *stime = str2kernel_uint_t(procfile_lineword(ff, 0, 14)); |
| 946 | |
| 947 | // leave the file open for the next iteration |
| 948 | |
| 949 | return 0; |
| 950 | #else |
| 951 | // TODO: add here cpu time detection per thread, for FreeBSD and MacOS |
| 952 | *utime = 0; |
| 953 | *stime = 0; |
| 954 | return 1; |
| 955 | #endif |
| 956 | } |
| 957 | |
| 958 | static Pvoid_t workers_by_pid_JudyL_array = NULL; |
| 959 | |
| 960 | static void workers_threads_cleanup(struct worker_utilization *wu) { |
| 961 | struct worker_thread *t = wu->threads; |
| 962 | while(t) { |
| 963 | struct worker_thread *next = t->next; |
| 964 | |
| 965 | if(!t->enabled) { |
| 966 | JudyLDel(&workers_by_pid_JudyL_array, t->pid, PJE0); |
| 967 | DOUBLE_LINKED_LIST_REMOVE_ITEM_UNSAFE(wu->threads, t, prev, next); |
| 968 | freez(t); |
| 969 | } |
| 970 | t = next; |
| 971 | } |
| 972 | } |
| 973 | |
| 974 | static struct worker_thread *worker_thread_find(struct worker_utilization *wu __maybe_unused, pid_t pid) { |
| 975 | struct worker_thread *wt = NULL; |
| 976 | |
| 977 | Pvoid_t *PValue = JudyLGet(workers_by_pid_JudyL_array, pid, PJE0); |
| 978 | if(PValue) |
| 979 | wt = *PValue; |
| 980 | |
| 981 | return wt; |
| 982 | } |
| 983 | |
| 984 | static struct worker_thread *worker_thread_create(struct worker_utilization *wu, pid_t pid) { |
| 985 | struct worker_thread *wt; |
| 986 | |
| 987 | wt = (struct worker_thread *)callocz(1, sizeof(struct worker_thread)); |
| 988 | wt->pid = pid; |
| 989 | |
| 990 | Pvoid_t *PValue = JudyLIns(&workers_by_pid_JudyL_array, pid, PJE0); |
| 991 | *PValue = wt; |
| 992 | |
| 993 | // link it |
| 994 | DOUBLE_LINKED_LIST_APPEND_ITEM_UNSAFE(wu->threads, wt, prev, next); |
| 995 | |
| 996 | return wt; |
| 997 | } |
| 998 | |
| 999 | static struct worker_thread *worker_thread_find_or_create(struct worker_utilization *wu, pid_t pid) { |
| 1000 | struct worker_thread *wt; |
| 1001 | wt = worker_thread_find(wu, pid); |
| 1002 | if(!wt) wt = worker_thread_create(wu, pid); |
| 1003 | |
| 1004 | return wt; |
| 1005 | } |
| 1006 | |
| 1007 | static void worker_utilization_charts_callback(void *ptr |
| 1008 | , pid_t pid |
| 1009 | , const char *thread_tag __maybe_unused |
| 1010 | , size_t max_job_id |
| 1011 | , size_t utilization_usec |
| 1012 | , size_t duration_usec |
| 1013 | , size_t jobs_started |
| 1014 | , size_t is_running |
| 1015 | , STRING **job_types_names |
| 1016 | , STRING **job_types_units |
| 1017 | , WORKER_METRIC_TYPE *job_types_metric_types |
| 1018 | , size_t *job_types_jobs_started |
| 1019 | , usec_t *job_types_busy_time |
| 1020 | , NETDATA_DOUBLE *job_types_custom_metrics |
| 1021 | , const char *spinlock_functions[] |
| 1022 | , size_t *spinlock_locks |
| 1023 | , size_t *spinlock_spins |
| 1024 | , uint64_t *memory_calls |
| 1025 | ) { |
| 1026 | struct worker_utilization *wu = (struct worker_utilization *)ptr; |
| 1027 | |
| 1028 | // find the worker_thread in the list |
| 1029 | struct worker_thread *wt = worker_thread_find_or_create(wu, pid); |
| 1030 | |
| 1031 | if(utilization_usec > duration_usec) |
| 1032 | utilization_usec = duration_usec; |
| 1033 | |
| 1034 | wt->enabled = true; |
| 1035 | wt->busy_time = utilization_usec; |
| 1036 | wt->jobs_started = jobs_started; |
| 1037 | |
| 1038 | wt->utime_old = wt->utime; |
| 1039 | wt->stime_old = wt->stime; |
| 1040 | wt->collected_time_old = wt->collected_time; |
| 1041 | |
| 1042 | if(max_job_id > wu->workers_max_job_id) |
| 1043 | wu->workers_max_job_id = max_job_id; |
| 1044 | |
| 1045 | wu->workers_total_busy_time += utilization_usec; |
| 1046 | wu->workers_total_duration += duration_usec; |
| 1047 | wu->workers_total_jobs_started += jobs_started; |
| 1048 | wu->workers_busy += is_running; |
| 1049 | wu->workers_registered++; |
| 1050 | |
| 1051 | double util = (double)utilization_usec * 100.0 / (double)duration_usec; |
| 1052 | if(util > wu->workers_max_busy_time) |
| 1053 | wu->workers_max_busy_time = util; |
| 1054 | |
| 1055 | if(util < wu->workers_min_busy_time) |
| 1056 | wu->workers_min_busy_time = util; |
| 1057 | |
| 1058 | // accumulate per job type statistics |
| 1059 | for(size_t i = 0; i <= max_job_id ;i++) { |
| 1060 | if(!wu->per_job_type[i].name && job_types_names[i]) |
| 1061 | wu->per_job_type[i].name = string_dup(job_types_names[i]); |
| 1062 | |
| 1063 | if(!wu->per_job_type[i].units && job_types_units[i]) |
| 1064 | wu->per_job_type[i].units = string_dup(job_types_units[i]); |
| 1065 | |
| 1066 | wu->per_job_type[i].type = job_types_metric_types[i]; |
| 1067 | |
| 1068 | wu->per_job_type[i].data[0].jobs_started += job_types_jobs_started[i]; |
| 1069 | wu->per_job_type[i].data[0].busy_time += job_types_busy_time[i]; |
| 1070 | |
| 1071 | NETDATA_DOUBLE value = job_types_custom_metrics[i]; |
| 1072 | if(netdata_double_isnumber(value)) { |
| 1073 | if(!wu->per_job_type[i].count_value) { |
| 1074 | wu->per_job_type[i].count_value = 1; |
| 1075 | wu->per_job_type[i].min_value = value; |
| 1076 | wu->per_job_type[i].max_value = value; |
| 1077 | wu->per_job_type[i].sum_value = value; |
| 1078 | } |
| 1079 | else { |
| 1080 | wu->per_job_type[i].count_value++; |
| 1081 | wu->per_job_type[i].sum_value += value; |
| 1082 | if(value < wu->per_job_type[i].min_value) wu->per_job_type[i].min_value = value; |
| 1083 | if(value > wu->per_job_type[i].max_value) wu->per_job_type[i].max_value = value; |
| 1084 | } |
| 1085 | } |
| 1086 | } |
| 1087 | |
| 1088 | // find its CPU utilization |
| 1089 | if((!read_thread_cpu_time_from_proc_stat(pid, &wt->utime, &wt->stime))) { |
| 1090 | wt->collected_time = now_realtime_usec(); |
| 1091 | usec_t delta = wt->collected_time - wt->collected_time_old; |
| 1092 | |
| 1093 | double utime = (double)(wt->utime - wt->utime_old) / (double)system_hz * 100.0 * (double)USEC_PER_SEC / (double)delta; |
| 1094 | double stime = (double)(wt->stime - wt->stime_old) / (double)system_hz * 100.0 * (double)USEC_PER_SEC / (double)delta; |
| 1095 | double cpu = utime + stime; |
| 1096 | wt->cpu = cpu; |
| 1097 | wt->cpu_enabled = true; |
| 1098 | |
| 1099 | wu->workers_cpu_total += cpu; |
| 1100 | if(cpu < wu->workers_cpu_min) wu->workers_cpu_min = cpu; |
| 1101 | if(cpu > wu->workers_cpu_max) wu->workers_cpu_max = cpu; |
| 1102 | } |
| 1103 | wu->workers_cpu_registered += (wt->cpu_enabled) ? 1 : 0; |
| 1104 | |
| 1105 | // ---------------------------------------------------------------------------------------------------------------- |
| 1106 | // spinlock contention |
| 1107 | |
| 1108 | // spinlocks |
| 1109 | for(size_t i = 0; i < WORKER_SPINLOCK_CONTENTION_FUNCTIONS && spinlock_functions[i] ;i++) { |
| 1110 | struct worker_spinlocks *wusp = SPINLOCKS_GET(&wu->spinlocks, (Word_t)spinlock_functions[i]); |
| 1111 | if(!wusp) { |
| 1112 | wusp = callocz(1, sizeof(*wusp)); |
| 1113 | SPINLOCKS_SET(&wu->spinlocks, (Word_t)spinlock_functions[i], wusp); |
| 1114 | } |
| 1115 | wusp->locks += spinlock_locks[i]; |
| 1116 | wusp->spins += spinlock_spins[i]; |
| 1117 | |
| 1118 | wusp = SPINLOCKS_GET(&ALL_SPINLOCKS, (Word_t)spinlock_functions[i]); |
| 1119 | if(!wusp) { |
| 1120 | wusp = callocz(1, sizeof(*wusp)); |
| 1121 | SPINLOCKS_SET(&ALL_SPINLOCKS, (Word_t)spinlock_functions[i], wusp); |
| 1122 | } |
| 1123 | wusp->locks += spinlock_locks[i]; |
| 1124 | wusp->spins += spinlock_spins[i]; |
| 1125 | } |
| 1126 | |
| 1127 | // ---------------------------------------------------------------------------------------------------------------- |
| 1128 | // memory calls |
| 1129 | |
| 1130 | for(size_t i = 0; i < WORKERS_MEMORY_CALL_MAX ;i++) |
| 1131 | wu->memory_calls[i] += memory_calls[i]; |
| 1132 | } |
| 1133 | |
| 1134 | static void spinlocks_free_callback(Word_t key __maybe_unused, struct worker_spinlocks *wusp, void *data __maybe_unused) { |
| 1135 | freez(wusp); |
| 1136 | } |
| 1137 | |
| 1138 | void pulse_workers_cleanup(void) { |
| 1139 | int i, j; |
| 1140 | for(i = 0; all_workers_utilization[i].name ;i++) { |
| 1141 | struct worker_utilization *wu = &all_workers_utilization[i]; |
| 1142 | |
| 1143 | if(wu->name_lowercase) { |
| 1144 | freez(wu->name_lowercase); |
| 1145 | wu->name_lowercase = NULL; |
| 1146 | } |
| 1147 | |
| 1148 | for(j = 0; j < WORKER_UTILIZATION_MAX_JOB_TYPES ;j++) { |
| 1149 | string_freez(wu->per_job_type[j].name); |
| 1150 | wu->per_job_type[j].name = NULL; |
| 1151 | |
| 1152 | string_freez(wu->per_job_type[j].units); |
| 1153 | wu->per_job_type[j].units = NULL; |
| 1154 | } |
| 1155 | |
| 1156 | // Free the spinlocks Judy array for this worker |
| 1157 | SPINLOCKS_FREE(&wu->spinlocks, spinlocks_free_callback, NULL); |
| 1158 | |
| 1159 | // mark all threads as not enabled |
| 1160 | struct worker_thread *t; |
| 1161 | for(t = wu->threads; t ; t = t->next) |
| 1162 | t->enabled = false; |
| 1163 | |
| 1164 | // let the cleanup job free them |
| 1165 | workers_threads_cleanup(wu); |
| 1166 | } |
| 1167 | |
| 1168 | // Clean up the global spinlocks array |
| 1169 | SPINLOCKS_FREE(&ALL_SPINLOCKS, spinlocks_free_callback, NULL); |
| 1170 | } |
| 1171 | |
| 1172 | void pulse_workers_do(bool extended) { |
| 1173 | if(!extended) return; |
| 1174 | |
| 1175 | static size_t iterations = 0; |
| 1176 | iterations++; |
| 1177 | |
| 1178 | Word_t idx = 0; |
| 1179 | for(struct worker_spinlocks *wusp = SPINLOCKS_FIRST(&ALL_SPINLOCKS, &idx); |
| 1180 | wusp; |
| 1181 | wusp = SPINLOCKS_NEXT(&ALL_SPINLOCKS, &idx)) { |
| 1182 | wusp->locks = 0; |
| 1183 | wusp->spins = 0; |
| 1184 | } |
| 1185 | |
| 1186 | for(int i = 0; all_workers_utilization[i].name ;i++) { |
| 1187 | workers_utilization_reset_statistics(&all_workers_utilization[i]); |
| 1188 | |
| 1189 | workers_foreach(all_workers_utilization[i].name, worker_utilization_charts_callback, &all_workers_utilization[i]); |
| 1190 | |
| 1191 | // skip the first iteration, so that we don't accumulate startup utilization to our charts |
| 1192 | if(likely(iterations > 1)) |
| 1193 | workers_utilization_update_chart(&all_workers_utilization[i]); |
| 1194 | |
| 1195 | workers_threads_cleanup(&all_workers_utilization[i]); |
| 1196 | } |
| 1197 | |
| 1198 | workers_total_cpu_utilization_chart(); |
| 1199 | workers_total_spinlock_contention_chart(); |
| 1200 | workers_total_memory_calls_chart(); |
| 1201 | } |