| 1 | #include <stdio.h> |
| 2 | #include <string.h> |
| 3 | #include <sys/time.h> |
| 4 | #include <sys/types.h> |
| 5 | #include <sys/stat.h> |
| 6 | #include <dirent.h> |
| 7 | #include <libgen.h> |
| 8 | #include <math.h> |
| 9 | #include <assert.h> |
| 10 | |
| 11 | #include "fio.h" |
| 12 | |
| 13 | static struct itimerval itimer; |
| 14 | static struct list_head disk_list = LIST_HEAD_INIT(disk_list); |
| 15 | static dev_t last_dev; |
| 16 | |
| 17 | /* |
| 18 | * Cheesy number->string conversion, complete with carry rounding error. |
| 19 | */ |
| 20 | static char *num2str(unsigned long num, int maxlen, int base) |
| 21 | { |
| 22 | /* |
| 23 | * could be passed in for 10^3 base, but every caller expects |
| 24 | * 2^10 base right now. |
| 25 | */ |
| 26 | const unsigned int thousand = 1024; |
| 27 | char postfix[] = { 'K', 'M', 'G', 'P', 'E' }; |
| 28 | char *buf; |
| 29 | int i; |
| 30 | |
| 31 | buf = malloc(128); |
| 32 | |
| 33 | for (i = 0; base > 1; i++) |
| 34 | base /= thousand; |
| 35 | |
| 36 | do { |
| 37 | int len, carry = 0; |
| 38 | |
| 39 | len = sprintf(buf, "%'lu", num); |
| 40 | if (len <= maxlen) { |
| 41 | buf[len] = postfix[i]; |
| 42 | buf[len + 1] = '\0'; |
| 43 | return buf; |
| 44 | } |
| 45 | |
| 46 | if ((num % thousand) >= (thousand / 2)) |
| 47 | carry = 1; |
| 48 | |
| 49 | num /= thousand; |
| 50 | num += carry; |
| 51 | i++; |
| 52 | } while (i <= 5); |
| 53 | |
| 54 | return buf; |
| 55 | } |
| 56 | |
| 57 | static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus) |
| 58 | { |
| 59 | unsigned in_flight; |
| 60 | char line[256]; |
| 61 | FILE *f; |
| 62 | char *p; |
| 63 | |
| 64 | f = fopen(du->path, "r"); |
| 65 | if (!f) |
| 66 | return 1; |
| 67 | |
| 68 | p = fgets(line, sizeof(line), f); |
| 69 | if (!p) { |
| 70 | fclose(f); |
| 71 | return 1; |
| 72 | } |
| 73 | |
| 74 | if (sscanf(p, "%u %u %llu %u %u %u %llu %u %u %u %u\n", &dus->ios[0], &dus->merges[0], &dus->sectors[0], &dus->ticks[0], &dus->ios[1], &dus->merges[1], &dus->sectors[1], &dus->ticks[1], &in_flight, &dus->io_ticks, &dus->time_in_queue) != 11) { |
| 75 | fclose(f); |
| 76 | return 1; |
| 77 | } |
| 78 | |
| 79 | fclose(f); |
| 80 | return 0; |
| 81 | } |
| 82 | |
| 83 | static void update_io_tick_disk(struct disk_util *du) |
| 84 | { |
| 85 | struct disk_util_stat __dus, *dus, *ldus; |
| 86 | struct timeval t; |
| 87 | |
| 88 | if (get_io_ticks(du, &__dus)) |
| 89 | return; |
| 90 | |
| 91 | dus = &du->dus; |
| 92 | ldus = &du->last_dus; |
| 93 | |
| 94 | dus->sectors[0] += (__dus.sectors[0] - ldus->sectors[0]); |
| 95 | dus->sectors[1] += (__dus.sectors[1] - ldus->sectors[1]); |
| 96 | dus->ios[0] += (__dus.ios[0] - ldus->ios[0]); |
| 97 | dus->ios[1] += (__dus.ios[1] - ldus->ios[1]); |
| 98 | dus->merges[0] += (__dus.merges[0] - ldus->merges[0]); |
| 99 | dus->merges[1] += (__dus.merges[1] - ldus->merges[1]); |
| 100 | dus->ticks[0] += (__dus.ticks[0] - ldus->ticks[0]); |
| 101 | dus->ticks[1] += (__dus.ticks[1] - ldus->ticks[1]); |
| 102 | dus->io_ticks += (__dus.io_ticks - ldus->io_ticks); |
| 103 | dus->time_in_queue += (__dus.time_in_queue - ldus->time_in_queue); |
| 104 | |
| 105 | fio_gettime(&t, NULL); |
| 106 | du->msec += mtime_since(&du->time, &t); |
| 107 | memcpy(&du->time, &t, sizeof(t)); |
| 108 | memcpy(ldus, &__dus, sizeof(__dus)); |
| 109 | } |
| 110 | |
| 111 | void update_io_ticks(void) |
| 112 | { |
| 113 | struct list_head *entry; |
| 114 | struct disk_util *du; |
| 115 | |
| 116 | list_for_each(entry, &disk_list) { |
| 117 | du = list_entry(entry, struct disk_util, list); |
| 118 | update_io_tick_disk(du); |
| 119 | } |
| 120 | } |
| 121 | |
| 122 | static int disk_util_exists(dev_t dev) |
| 123 | { |
| 124 | struct list_head *entry; |
| 125 | struct disk_util *du; |
| 126 | |
| 127 | list_for_each(entry, &disk_list) { |
| 128 | du = list_entry(entry, struct disk_util, list); |
| 129 | |
| 130 | if (du->dev == dev) |
| 131 | return 1; |
| 132 | } |
| 133 | |
| 134 | return 0; |
| 135 | } |
| 136 | |
| 137 | static void disk_util_add(dev_t dev, char *path) |
| 138 | { |
| 139 | struct disk_util *du, *__du; |
| 140 | struct list_head *entry; |
| 141 | |
| 142 | du = malloc(sizeof(*du)); |
| 143 | memset(du, 0, sizeof(*du)); |
| 144 | INIT_LIST_HEAD(&du->list); |
| 145 | sprintf(du->path, "%s/stat", path); |
| 146 | du->name = strdup(basename(path)); |
| 147 | du->dev = dev; |
| 148 | |
| 149 | list_for_each(entry, &disk_list) { |
| 150 | __du = list_entry(entry, struct disk_util, list); |
| 151 | |
| 152 | if (!strcmp(du->name, __du->name)) { |
| 153 | free(du->name); |
| 154 | free(du); |
| 155 | return; |
| 156 | } |
| 157 | } |
| 158 | |
| 159 | fio_gettime(&du->time, NULL); |
| 160 | get_io_ticks(du, &du->last_dus); |
| 161 | |
| 162 | list_add_tail(&du->list, &disk_list); |
| 163 | } |
| 164 | |
| 165 | static int check_dev_match(dev_t dev, char *path) |
| 166 | { |
| 167 | unsigned int major, minor; |
| 168 | char line[256], *p; |
| 169 | FILE *f; |
| 170 | |
| 171 | f = fopen(path, "r"); |
| 172 | if (!f) { |
| 173 | perror("open path"); |
| 174 | return 1; |
| 175 | } |
| 176 | |
| 177 | p = fgets(line, sizeof(line), f); |
| 178 | if (!p) { |
| 179 | fclose(f); |
| 180 | return 1; |
| 181 | } |
| 182 | |
| 183 | if (sscanf(p, "%u:%u", &major, &minor) != 2) { |
| 184 | fclose(f); |
| 185 | return 1; |
| 186 | } |
| 187 | |
| 188 | if (((major << 8) | minor) == dev) { |
| 189 | fclose(f); |
| 190 | return 0; |
| 191 | } |
| 192 | |
| 193 | fclose(f); |
| 194 | return 1; |
| 195 | } |
| 196 | |
| 197 | static int find_block_dir(dev_t dev, char *path) |
| 198 | { |
| 199 | struct dirent *dir; |
| 200 | struct stat st; |
| 201 | int found = 0; |
| 202 | DIR *D; |
| 203 | |
| 204 | D = opendir(path); |
| 205 | if (!D) |
| 206 | return 0; |
| 207 | |
| 208 | while ((dir = readdir(D)) != NULL) { |
| 209 | char full_path[256]; |
| 210 | |
| 211 | if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, "..")) |
| 212 | continue; |
| 213 | |
| 214 | sprintf(full_path, "%s/%s", path, dir->d_name); |
| 215 | |
| 216 | if (!strcmp(dir->d_name, "dev")) { |
| 217 | if (!check_dev_match(dev, full_path)) { |
| 218 | found = 1; |
| 219 | break; |
| 220 | } |
| 221 | } |
| 222 | |
| 223 | if (lstat(full_path, &st) == -1) { |
| 224 | perror("stat"); |
| 225 | break; |
| 226 | } |
| 227 | |
| 228 | if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode)) |
| 229 | continue; |
| 230 | |
| 231 | found = find_block_dir(dev, full_path); |
| 232 | if (found) { |
| 233 | strcpy(path, full_path); |
| 234 | break; |
| 235 | } |
| 236 | } |
| 237 | |
| 238 | closedir(D); |
| 239 | return found; |
| 240 | } |
| 241 | |
| 242 | void init_disk_util(struct thread_data *td) |
| 243 | { |
| 244 | struct fio_file *f; |
| 245 | struct stat st; |
| 246 | char foo[PATH_MAX], tmp[PATH_MAX]; |
| 247 | dev_t dev; |
| 248 | char *p; |
| 249 | |
| 250 | if (!td->do_disk_util || (td->io_ops->flags & FIO_DISKLESSIO)) |
| 251 | return; |
| 252 | |
| 253 | /* |
| 254 | * Just use the same file, they are on the same device. |
| 255 | */ |
| 256 | f = &td->files[0]; |
| 257 | if (!stat(f->file_name, &st)) { |
| 258 | if (S_ISBLK(st.st_mode)) |
| 259 | dev = st.st_rdev; |
| 260 | else |
| 261 | dev = st.st_dev; |
| 262 | } else { |
| 263 | /* |
| 264 | * must be a file, open "." in that path |
| 265 | */ |
| 266 | strncpy(foo, f->file_name, PATH_MAX - 1); |
| 267 | p = dirname(foo); |
| 268 | if (stat(p, &st)) { |
| 269 | perror("disk util stat"); |
| 270 | return; |
| 271 | } |
| 272 | |
| 273 | dev = st.st_dev; |
| 274 | } |
| 275 | |
| 276 | if (disk_util_exists(dev)) |
| 277 | return; |
| 278 | |
| 279 | /* |
| 280 | * for an fs without a device, we will repeatedly stat through |
| 281 | * sysfs which can take oodles of time for thousands of files. so |
| 282 | * cache the last lookup and compare with that before going through |
| 283 | * everything again. |
| 284 | */ |
| 285 | if (dev == last_dev) |
| 286 | return; |
| 287 | |
| 288 | last_dev = dev; |
| 289 | |
| 290 | sprintf(foo, "/sys/block"); |
| 291 | if (!find_block_dir(dev, foo)) |
| 292 | return; |
| 293 | |
| 294 | /* |
| 295 | * If there's a ../queue/ directory there, we are inside a partition. |
| 296 | * Check if that is the case and jump back. For loop/md/dm etc we |
| 297 | * are already in the right spot. |
| 298 | */ |
| 299 | sprintf(tmp, "%s/../queue", foo); |
| 300 | if (!stat(tmp, &st)) { |
| 301 | p = dirname(foo); |
| 302 | sprintf(tmp, "%s/queue", p); |
| 303 | if (stat(tmp, &st)) { |
| 304 | log_err("unknown sysfs layout\n"); |
| 305 | return; |
| 306 | } |
| 307 | strncpy(tmp, p, PATH_MAX - 1); |
| 308 | sprintf(foo, "%s", tmp); |
| 309 | } |
| 310 | |
| 311 | if (td->ioscheduler) |
| 312 | td->sysfs_root = strdup(foo); |
| 313 | |
| 314 | disk_util_add(dev, foo); |
| 315 | } |
| 316 | |
| 317 | void disk_util_timer_arm(void) |
| 318 | { |
| 319 | itimer.it_value.tv_sec = 0; |
| 320 | itimer.it_value.tv_usec = DISK_UTIL_MSEC * 1000; |
| 321 | setitimer(ITIMER_REAL, &itimer, NULL); |
| 322 | } |
| 323 | |
| 324 | void update_rusage_stat(struct thread_data *td) |
| 325 | { |
| 326 | struct thread_stat *ts = &td->ts; |
| 327 | |
| 328 | getrusage(RUSAGE_SELF, &ts->ru_end); |
| 329 | |
| 330 | ts->usr_time += mtime_since(&ts->ru_start.ru_utime, &ts->ru_end.ru_utime); |
| 331 | ts->sys_time += mtime_since(&ts->ru_start.ru_stime, &ts->ru_end.ru_stime); |
| 332 | ts->ctx += ts->ru_end.ru_nvcsw + ts->ru_end.ru_nivcsw - (ts->ru_start.ru_nvcsw + ts->ru_start.ru_nivcsw); |
| 333 | |
| 334 | memcpy(&ts->ru_start, &ts->ru_end, sizeof(ts->ru_end)); |
| 335 | } |
| 336 | |
| 337 | static int calc_lat(struct io_stat *is, unsigned long *min, unsigned long *max, |
| 338 | double *mean, double *dev) |
| 339 | { |
| 340 | double n = is->samples; |
| 341 | |
| 342 | if (is->samples == 0) |
| 343 | return 0; |
| 344 | |
| 345 | *min = is->min_val; |
| 346 | *max = is->max_val; |
| 347 | |
| 348 | n = (double) is->samples; |
| 349 | *mean = is->mean; |
| 350 | |
| 351 | if (n > 1.0) |
| 352 | *dev = sqrt(is->S / (n - 1.0)); |
| 353 | else |
| 354 | *dev = -1.0; |
| 355 | |
| 356 | return 1; |
| 357 | } |
| 358 | |
| 359 | static void show_group_stats(struct group_run_stats *rs, int id) |
| 360 | { |
| 361 | char *p1, *p2, *p3, *p4; |
| 362 | const char *ddir_str[] = { " READ", " WRITE" }; |
| 363 | int i; |
| 364 | |
| 365 | fprintf(f_out, "\nRun status group %d (all jobs):\n", id); |
| 366 | |
| 367 | for (i = 0; i <= DDIR_WRITE; i++) { |
| 368 | if (!rs->max_run[i]) |
| 369 | continue; |
| 370 | |
| 371 | p1 = num2str(rs->io_kb[i], 6, 1); |
| 372 | p2 = num2str(rs->agg[i], 6, 1); |
| 373 | p3 = num2str(rs->min_bw[i], 6, 1); |
| 374 | p4 = num2str(rs->max_bw[i], 6, 1); |
| 375 | |
| 376 | fprintf(f_out, "%s: io=%siB, aggrb=%siB/s, minb=%siB/s, maxb=%siB/s, mint=%llumsec, maxt=%llumsec\n", ddir_str[i], p1, p2, p3, p4, rs->min_run[0], rs->max_run[0]); |
| 377 | |
| 378 | free(p1); |
| 379 | free(p2); |
| 380 | free(p3); |
| 381 | free(p4); |
| 382 | } |
| 383 | } |
| 384 | |
| 385 | static void show_disk_util(void) |
| 386 | { |
| 387 | struct disk_util_stat *dus; |
| 388 | struct list_head *entry, *next; |
| 389 | struct disk_util *du; |
| 390 | double util; |
| 391 | |
| 392 | fprintf(f_out, "\nDisk stats (read/write):\n"); |
| 393 | |
| 394 | list_for_each(entry, &disk_list) { |
| 395 | du = list_entry(entry, struct disk_util, list); |
| 396 | dus = &du->dus; |
| 397 | |
| 398 | util = (double) 100 * du->dus.io_ticks / (double) du->msec; |
| 399 | if (util > 100.0) |
| 400 | util = 100.0; |
| 401 | |
| 402 | fprintf(f_out, " %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, in_queue=%u, util=%3.2f%%\n", du->name, dus->ios[0], dus->ios[1], dus->merges[0], dus->merges[1], dus->ticks[0], dus->ticks[1], dus->time_in_queue, util); |
| 403 | } |
| 404 | |
| 405 | /* |
| 406 | * now free the list |
| 407 | */ |
| 408 | list_for_each_safe(entry, next, &disk_list) { |
| 409 | list_del(entry); |
| 410 | du = list_entry(entry, struct disk_util, list); |
| 411 | free(du->name); |
| 412 | free(du); |
| 413 | } |
| 414 | } |
| 415 | |
| 416 | static void show_ddir_status(struct group_run_stats *rs, struct thread_stat *ts, |
| 417 | int ddir) |
| 418 | { |
| 419 | const char *ddir_str[] = { "read ", "write" }; |
| 420 | unsigned long min, max; |
| 421 | unsigned long long bw; |
| 422 | double mean, dev; |
| 423 | char *io_p, *bw_p; |
| 424 | |
| 425 | if (!ts->runtime[ddir]) |
| 426 | return; |
| 427 | |
| 428 | bw = ts->io_bytes[ddir] / ts->runtime[ddir]; |
| 429 | io_p = num2str(ts->io_bytes[ddir] >> 10, 6, 1); |
| 430 | bw_p = num2str(bw, 6, 1); |
| 431 | |
| 432 | fprintf(f_out, " %s: io=%siB, bw=%siB/s, runt=%6lumsec\n", ddir_str[ddir], io_p, bw_p, ts->runtime[ddir]); |
| 433 | |
| 434 | free(io_p); |
| 435 | free(bw_p); |
| 436 | |
| 437 | if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev)) |
| 438 | fprintf(f_out, " slat (msec): min=%5lu, max=%5lu, avg=%5.02f, stdev=%5.02f\n", min, max, mean, dev); |
| 439 | |
| 440 | if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev)) |
| 441 | fprintf(f_out, " clat (msec): min=%5lu, max=%5lu, avg=%5.02f, stdev=%5.02f\n", min, max, mean, dev); |
| 442 | |
| 443 | if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) { |
| 444 | double p_of_agg; |
| 445 | |
| 446 | p_of_agg = mean * 100 / (double) rs->agg[ddir]; |
| 447 | fprintf(f_out, " bw (KiB/s) : min=%5lu, max=%5lu, per=%3.2f%%, avg=%5.02f, stdev=%5.02f\n", min, max, p_of_agg, mean, dev); |
| 448 | } |
| 449 | } |
| 450 | |
| 451 | static void show_thread_status(struct thread_stat *ts, |
| 452 | struct group_run_stats *rs) |
| 453 | { |
| 454 | double usr_cpu, sys_cpu; |
| 455 | unsigned long runtime; |
| 456 | double io_u_dist[FIO_IO_U_MAP_NR]; |
| 457 | double io_u_lat[FIO_IO_U_LAT_NR]; |
| 458 | int i; |
| 459 | |
| 460 | if (!(ts->io_bytes[0] + ts->io_bytes[1])) |
| 461 | return; |
| 462 | |
| 463 | if (!ts->error) |
| 464 | fprintf(f_out, "%s: (groupid=%d): err=%2d: pid=%d\n", ts->name, ts->groupid, ts->error, ts->pid); |
| 465 | else |
| 466 | fprintf(f_out, "%s: (groupid=%d): err=%2d (%s): pid=%d\n", ts->name, ts->groupid, ts->error, ts->verror, ts->pid); |
| 467 | |
| 468 | if (ts->io_bytes[DDIR_READ]) |
| 469 | show_ddir_status(rs, ts, DDIR_READ); |
| 470 | if (ts->io_bytes[DDIR_WRITE]) |
| 471 | show_ddir_status(rs, ts, DDIR_WRITE); |
| 472 | |
| 473 | runtime = ts->total_run_time; |
| 474 | if (runtime) { |
| 475 | double runt = (double) runtime; |
| 476 | |
| 477 | usr_cpu = (double) ts->usr_time * 100 / runt; |
| 478 | sys_cpu = (double) ts->sys_time * 100 / runt; |
| 479 | } else { |
| 480 | usr_cpu = 0; |
| 481 | sys_cpu = 0; |
| 482 | } |
| 483 | |
| 484 | fprintf(f_out, " cpu : usr=%3.2f%%, sys=%3.2f%%, ctx=%lu\n", usr_cpu, sys_cpu, ts->ctx); |
| 485 | |
| 486 | /* |
| 487 | * Do depth distribution calculations |
| 488 | */ |
| 489 | for (i = 0; i < FIO_IO_U_MAP_NR; i++) { |
| 490 | io_u_dist[i] = (double) ts->io_u_map[i] / (double) ts->total_io_u; |
| 491 | io_u_dist[i] *= 100.0; |
| 492 | } |
| 493 | |
| 494 | fprintf(f_out, " IO depths : 1=%3.1f%%, 2=%3.1f%%, 4=%3.1f%%, 8=%3.1f%%, 16=%3.1f%%, 32=%3.1f%%, >=64=%3.1f%%\n", io_u_dist[0], io_u_dist[1], io_u_dist[2], io_u_dist[3], io_u_dist[4], io_u_dist[5], io_u_dist[6]); |
| 495 | |
| 496 | /* |
| 497 | * Do latency distribution calculations |
| 498 | */ |
| 499 | for (i = 0; i < FIO_IO_U_LAT_NR; i++) { |
| 500 | io_u_lat[i] = (double) ts->io_u_lat[i] / (double) ts->total_io_u; |
| 501 | io_u_lat[i] *= 100.0; |
| 502 | } |
| 503 | |
| 504 | fprintf(f_out, " lat (msec): 2=%3.1f%%, 4=%3.1f%%, 10=%3.1f%%, 20=%3.1f%%, 50=%3.1f%%, 100=%3.1f%%\n", io_u_lat[0], io_u_lat[1], io_u_lat[2], io_u_lat[3], io_u_lat[4], io_u_lat[5]); |
| 505 | fprintf(f_out, " lat (msec): 250=%3.1f%%, 500=%3.1f%%, 750=%3.1f%%, 1000=%3.1f%%, >=2000=%3.1f%%\n", io_u_lat[6], io_u_lat[7], io_u_lat[8], io_u_lat[9], io_u_lat[10]); |
| 506 | |
| 507 | if (ts->description) |
| 508 | fprintf(f_out, "%s\n", ts->description); |
| 509 | } |
| 510 | |
| 511 | static void show_ddir_status_terse(struct thread_stat *ts, |
| 512 | struct group_run_stats *rs, int ddir) |
| 513 | { |
| 514 | unsigned long min, max; |
| 515 | unsigned long long bw; |
| 516 | double mean, dev; |
| 517 | |
| 518 | bw = 0; |
| 519 | if (ts->runtime[ddir]) |
| 520 | bw = ts->io_bytes[ddir] / ts->runtime[ddir]; |
| 521 | |
| 522 | fprintf(f_out, ",%llu,%llu,%lu", ts->io_bytes[ddir] >> 10, bw, ts->runtime[ddir]); |
| 523 | |
| 524 | if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev)) |
| 525 | fprintf(f_out, ",%lu,%lu,%f,%f", min, max, mean, dev); |
| 526 | else |
| 527 | fprintf(f_out, ",%lu,%lu,%f,%f", 0UL, 0UL, 0.0, 0.0); |
| 528 | |
| 529 | if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev)) |
| 530 | fprintf(f_out, ",%lu,%lu,%f,%f", min, max, mean, dev); |
| 531 | else |
| 532 | fprintf(f_out, ",%lu,%lu,%f,%f", 0UL, 0UL, 0.0, 0.0); |
| 533 | |
| 534 | if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) { |
| 535 | double p_of_agg; |
| 536 | |
| 537 | p_of_agg = mean * 100 / (double) rs->agg[ddir]; |
| 538 | fprintf(f_out, ",%lu,%lu,%f%%,%f,%f", min, max, p_of_agg, mean, dev); |
| 539 | } else |
| 540 | fprintf(f_out, ",%lu,%lu,%f%%,%f,%f", 0UL, 0UL, 0.0, 0.0, 0.0); |
| 541 | } |
| 542 | |
| 543 | |
| 544 | static void show_thread_status_terse(struct thread_stat *ts, |
| 545 | struct group_run_stats *rs) |
| 546 | { |
| 547 | double usr_cpu, sys_cpu; |
| 548 | |
| 549 | fprintf(f_out, "%s,%d,%d", ts->name, ts->groupid, ts->error); |
| 550 | |
| 551 | show_ddir_status_terse(ts, rs, 0); |
| 552 | show_ddir_status_terse(ts, rs, 1); |
| 553 | |
| 554 | if (ts->total_run_time) { |
| 555 | double runt = (double) ts->total_run_time; |
| 556 | |
| 557 | usr_cpu = (double) ts->usr_time * 100 / runt; |
| 558 | sys_cpu = (double) ts->sys_time * 100 / runt; |
| 559 | } else { |
| 560 | usr_cpu = 0; |
| 561 | sys_cpu = 0; |
| 562 | } |
| 563 | |
| 564 | fprintf(f_out, ",%f%%,%f%%,%lu\n", usr_cpu, sys_cpu, ts->ctx); |
| 565 | } |
| 566 | |
| 567 | static void __sum_stat(struct io_stat *dst, struct io_stat *src, int nr) |
| 568 | { |
| 569 | double mean, S; |
| 570 | |
| 571 | dst->min_val = min(dst->min_val, src->min_val); |
| 572 | dst->max_val = max(dst->max_val, src->max_val); |
| 573 | dst->samples += src->samples; |
| 574 | |
| 575 | /* |
| 576 | * Needs a new method for calculating stddev, we cannot just |
| 577 | * average them we do below for nr > 1 |
| 578 | */ |
| 579 | if (nr == 1) { |
| 580 | mean = src->mean; |
| 581 | S = src->S; |
| 582 | } else { |
| 583 | mean = ((src->mean * (double) nr) + dst->mean) / ((double) nr + 1.0); |
| 584 | S = ((src->S * (double) nr) + dst->S) / ((double) nr + 1.0); |
| 585 | } |
| 586 | |
| 587 | dst->mean = mean; |
| 588 | dst->S = S; |
| 589 | } |
| 590 | |
| 591 | static void sum_stat(struct io_stat *dst, struct io_stat *src, int nr) |
| 592 | { |
| 593 | __sum_stat(&dst[DDIR_READ], &src[DDIR_READ], nr); |
| 594 | __sum_stat(&dst[DDIR_WRITE], &src[DDIR_WRITE], nr); |
| 595 | } |
| 596 | |
| 597 | void show_run_stats(void) |
| 598 | { |
| 599 | struct group_run_stats *runstats, *rs; |
| 600 | struct thread_data *td; |
| 601 | struct thread_stat *threadstats, *ts; |
| 602 | int i, j, k, nr_ts, last_ts, members; |
| 603 | |
| 604 | runstats = malloc(sizeof(struct group_run_stats) * (groupid + 1)); |
| 605 | |
| 606 | for (i = 0; i < groupid + 1; i++) { |
| 607 | rs = &runstats[i]; |
| 608 | |
| 609 | memset(rs, 0, sizeof(*rs)); |
| 610 | rs->min_bw[0] = rs->min_run[0] = ~0UL; |
| 611 | rs->min_bw[1] = rs->min_run[1] = ~0UL; |
| 612 | } |
| 613 | |
| 614 | /* |
| 615 | * find out how many threads stats we need. if group reporting isn't |
| 616 | * enabled, it's one-per-td. |
| 617 | */ |
| 618 | nr_ts = 0; |
| 619 | last_ts = -1; |
| 620 | for_each_td(td, i) { |
| 621 | if (!td->group_reporting) { |
| 622 | nr_ts++; |
| 623 | continue; |
| 624 | } |
| 625 | if (last_ts == td->groupid) |
| 626 | continue; |
| 627 | |
| 628 | last_ts = td->groupid; |
| 629 | nr_ts++; |
| 630 | } |
| 631 | |
| 632 | threadstats = malloc(nr_ts * sizeof(struct thread_stat)); |
| 633 | |
| 634 | for (i = 0; i < nr_ts; i++) { |
| 635 | ts = &threadstats[i]; |
| 636 | |
| 637 | memset(ts, 0, sizeof(*ts)); |
| 638 | ts->clat_stat[0].min_val = -1UL; |
| 639 | ts->clat_stat[1].min_val = -1UL; |
| 640 | ts->slat_stat[0].min_val = -1UL; |
| 641 | ts->slat_stat[1].min_val = -1UL; |
| 642 | ts->bw_stat[0].min_val = -1UL; |
| 643 | ts->bw_stat[1].min_val = -1UL; |
| 644 | } |
| 645 | |
| 646 | j = 0; |
| 647 | last_ts = -1; |
| 648 | members = 0; |
| 649 | for_each_td(td, i) { |
| 650 | ts = &threadstats[j]; |
| 651 | |
| 652 | members++; |
| 653 | |
| 654 | if (!ts->groupid) { |
| 655 | ts->name = td->name; |
| 656 | ts->description = td->description; |
| 657 | ts->error = td->error; |
| 658 | ts->groupid = td->groupid; |
| 659 | ts->pid = td->pid; |
| 660 | ts->verror = td->verror; |
| 661 | } |
| 662 | |
| 663 | sum_stat(ts->clat_stat, td->ts.clat_stat, members); |
| 664 | sum_stat(ts->slat_stat, td->ts.slat_stat, members); |
| 665 | sum_stat(ts->bw_stat, td->ts.bw_stat, members); |
| 666 | |
| 667 | ts->stat_io_bytes[0] += td->ts.stat_io_bytes[0]; |
| 668 | ts->stat_io_bytes[1] += td->ts.stat_io_bytes[1]; |
| 669 | |
| 670 | ts->usr_time += td->ts.usr_time; |
| 671 | ts->sys_time += td->ts.sys_time; |
| 672 | ts->ctx += td->ts.ctx; |
| 673 | |
| 674 | for (k = 0; k < FIO_IO_U_MAP_NR; k++) |
| 675 | ts->io_u_map[k] += td->ts.io_u_map[k]; |
| 676 | for (k = 0; k < FIO_IO_U_LAT_NR; k++) |
| 677 | ts->io_u_lat[k] += td->ts.io_u_lat[k]; |
| 678 | |
| 679 | ts->total_io_u += td->ts.total_io_u; |
| 680 | ts->io_bytes[0] += td->ts.io_bytes[0]; |
| 681 | ts->io_bytes[1] += td->ts.io_bytes[1]; |
| 682 | |
| 683 | if (ts->runtime[0] < td->ts.runtime[0]) |
| 684 | ts->runtime[0] = td->ts.runtime[0]; |
| 685 | if (ts->runtime[1] < td->ts.runtime[1]) |
| 686 | ts->runtime[1] = td->ts.runtime[1]; |
| 687 | |
| 688 | ts->total_run_time += td->ts.total_run_time; |
| 689 | |
| 690 | if (!td->group_reporting) { |
| 691 | members = 0; |
| 692 | j++; |
| 693 | continue; |
| 694 | } |
| 695 | if (last_ts == td->groupid) |
| 696 | continue; |
| 697 | |
| 698 | if (last_ts != -1) { |
| 699 | members = 0; |
| 700 | j++; |
| 701 | } |
| 702 | |
| 703 | last_ts = td->groupid; |
| 704 | } |
| 705 | |
| 706 | for (i = 0; i < nr_ts; i++) { |
| 707 | unsigned long long rbw, wbw; |
| 708 | |
| 709 | ts = &threadstats[i]; |
| 710 | rs = &runstats[ts->groupid]; |
| 711 | |
| 712 | if (ts->runtime[0] < rs->min_run[0] || !rs->min_run[0]) |
| 713 | rs->min_run[0] = ts->runtime[0]; |
| 714 | if (ts->runtime[0] > rs->max_run[0]) |
| 715 | rs->max_run[0] = ts->runtime[0]; |
| 716 | if (ts->runtime[1] < rs->min_run[1] || !rs->min_run[1]) |
| 717 | rs->min_run[1] = ts->runtime[1]; |
| 718 | if (ts->runtime[1] > rs->max_run[1]) |
| 719 | rs->max_run[1] = ts->runtime[1]; |
| 720 | |
| 721 | rbw = wbw = 0; |
| 722 | if (ts->runtime[0]) |
| 723 | rbw = td->io_bytes[0] / (unsigned long long) ts->runtime[0]; |
| 724 | if (ts->runtime[1]) |
| 725 | wbw = td->io_bytes[1] / (unsigned long long) ts->runtime[1]; |
| 726 | |
| 727 | if (rbw < rs->min_bw[0]) |
| 728 | rs->min_bw[0] = rbw; |
| 729 | if (wbw < rs->min_bw[1]) |
| 730 | rs->min_bw[1] = wbw; |
| 731 | if (rbw > rs->max_bw[0]) |
| 732 | rs->max_bw[0] = rbw; |
| 733 | if (wbw > rs->max_bw[1]) |
| 734 | rs->max_bw[1] = wbw; |
| 735 | |
| 736 | rs->io_kb[0] += ts->io_bytes[0] >> 10; |
| 737 | rs->io_kb[1] += ts->io_bytes[1] >> 10; |
| 738 | } |
| 739 | |
| 740 | for (i = 0; i < groupid + 1; i++) { |
| 741 | rs = &runstats[i]; |
| 742 | |
| 743 | if (rs->max_run[0]) |
| 744 | rs->agg[0] = (rs->io_kb[0]*1024) / rs->max_run[0]; |
| 745 | if (rs->max_run[1]) |
| 746 | rs->agg[1] = (rs->io_kb[1]*1024) / rs->max_run[1]; |
| 747 | } |
| 748 | |
| 749 | /* |
| 750 | * don't overwrite last signal output |
| 751 | */ |
| 752 | if (!terse_output) |
| 753 | printf("\n"); |
| 754 | |
| 755 | for (i = 0; i < nr_ts; i++) { |
| 756 | ts = &threadstats[i]; |
| 757 | rs = &runstats[ts->groupid]; |
| 758 | |
| 759 | if (terse_output) |
| 760 | show_thread_status_terse(ts, rs); |
| 761 | else |
| 762 | show_thread_status(ts, rs); |
| 763 | } |
| 764 | |
| 765 | if (!terse_output) { |
| 766 | for (i = 0; i < groupid + 1; i++) |
| 767 | show_group_stats(&runstats[i], i); |
| 768 | |
| 769 | show_disk_util(); |
| 770 | } |
| 771 | |
| 772 | free(runstats); |
| 773 | free(threadstats); |
| 774 | } |
| 775 | |
| 776 | static inline void add_stat_sample(struct io_stat *is, unsigned long data) |
| 777 | { |
| 778 | double val = data; |
| 779 | double delta, n; |
| 780 | |
| 781 | if (data > is->max_val) |
| 782 | is->max_val = data; |
| 783 | if (data < is->min_val) |
| 784 | is->min_val = data; |
| 785 | |
| 786 | delta = val - is->mean; |
| 787 | n = is->samples + 1.0; |
| 788 | is->mean += delta / n; |
| 789 | is->S += delta * (val - is->mean); |
| 790 | |
| 791 | is->samples++; |
| 792 | } |
| 793 | |
| 794 | static void __add_log_sample(struct io_log *iolog, unsigned long val, |
| 795 | enum fio_ddir ddir, unsigned long time) |
| 796 | { |
| 797 | if (iolog->nr_samples == iolog->max_samples) { |
| 798 | int new_size = sizeof(struct io_sample) * iolog->max_samples*2; |
| 799 | |
| 800 | iolog->log = realloc(iolog->log, new_size); |
| 801 | iolog->max_samples <<= 1; |
| 802 | } |
| 803 | |
| 804 | iolog->log[iolog->nr_samples].val = val; |
| 805 | iolog->log[iolog->nr_samples].time = time; |
| 806 | iolog->log[iolog->nr_samples].ddir = ddir; |
| 807 | iolog->nr_samples++; |
| 808 | } |
| 809 | |
| 810 | static void add_log_sample(struct thread_data *td, struct io_log *iolog, |
| 811 | unsigned long val, enum fio_ddir ddir) |
| 812 | { |
| 813 | __add_log_sample(iolog, val, ddir, mtime_since_now(&td->epoch)); |
| 814 | } |
| 815 | |
| 816 | void add_agg_sample(unsigned long val, enum fio_ddir ddir) |
| 817 | { |
| 818 | struct io_log *iolog = agg_io_log[ddir]; |
| 819 | |
| 820 | __add_log_sample(iolog, val, ddir, mtime_since_genesis()); |
| 821 | } |
| 822 | |
| 823 | void add_clat_sample(struct thread_data *td, enum fio_ddir ddir, |
| 824 | unsigned long msec) |
| 825 | { |
| 826 | struct thread_stat *ts = &td->ts; |
| 827 | |
| 828 | add_stat_sample(&ts->clat_stat[ddir], msec); |
| 829 | |
| 830 | if (ts->clat_log) |
| 831 | add_log_sample(td, ts->clat_log, msec, ddir); |
| 832 | } |
| 833 | |
| 834 | void add_slat_sample(struct thread_data *td, enum fio_ddir ddir, |
| 835 | unsigned long msec) |
| 836 | { |
| 837 | struct thread_stat *ts = &td->ts; |
| 838 | |
| 839 | add_stat_sample(&ts->slat_stat[ddir], msec); |
| 840 | |
| 841 | if (ts->slat_log) |
| 842 | add_log_sample(td, ts->slat_log, msec, ddir); |
| 843 | } |
| 844 | |
| 845 | void add_bw_sample(struct thread_data *td, enum fio_ddir ddir, |
| 846 | struct timeval *t) |
| 847 | { |
| 848 | struct thread_stat *ts = &td->ts; |
| 849 | unsigned long spent = mtime_since(&ts->stat_sample_time[ddir], t); |
| 850 | unsigned long rate; |
| 851 | |
| 852 | if (spent < td->bw_avg_time) |
| 853 | return; |
| 854 | |
| 855 | rate = (td->this_io_bytes[ddir] - ts->stat_io_bytes[ddir]) / spent; |
| 856 | add_stat_sample(&ts->bw_stat[ddir], rate); |
| 857 | |
| 858 | if (ts->bw_log) |
| 859 | add_log_sample(td, ts->bw_log, rate, ddir); |
| 860 | |
| 861 | fio_gettime(&ts->stat_sample_time[ddir], NULL); |
| 862 | ts->stat_io_bytes[ddir] = td->this_io_bytes[ddir]; |
| 863 | } |