6 #ifdef CONFIG_VALGRIND_DEV
7 #include <valgrind/drd.h>
9 #define DRD_IGNORE_VAR(x) do { } while (0)
15 static char __run_str[REAL_MAX_JOBS + 1];
16 static char run_str[__THREAD_RUNSTR_SZ(REAL_MAX_JOBS) + 1];
18 static void update_condensed_str(char *rstr, char *run_str_condensed)
24 *run_str_condensed++ = *rstr++;
25 while (*(rstr - 1) == *rstr) {
29 run_str_condensed += sprintf(run_str_condensed, "(%u),", nr);
33 *run_str_condensed = '\0';
37 * Sets the status of the 'td' in the printed status map.
39 static void check_str_update(struct thread_data *td)
41 char c = __run_str[td->thread_number - 1];
43 switch (td->runstate) {
61 if (td->o.rwmix[DDIR_READ] == 100)
63 else if (td->o.rwmix[DDIR_WRITE] == 100)
68 if (td->o.rwmix[DDIR_READ] == 100)
70 else if (td->o.rwmix[DDIR_WRITE] == 100)
75 } else if (td_read(td)) {
80 } else if (td_write(td)) {
115 log_err("state %d\n", td->runstate);
118 __run_str[td->thread_number - 1] = c;
119 update_condensed_str(__run_str, run_str);
123 * Convert seconds to a printable string.
125 void eta_to_str(char *str, unsigned long eta_sec)
127 unsigned int d, h, m, s;
145 str += sprintf(str, "%02ud:", d);
149 str += sprintf(str, "%02uh:", h);
151 str += sprintf(str, "%02um:", m);
152 sprintf(str, "%02us", s);
156 * Best effort calculation of the estimated pending runtime of a job.
158 static unsigned long thread_eta(struct thread_data *td)
160 unsigned long long bytes_total, bytes_done;
161 unsigned long eta_sec = 0;
162 unsigned long elapsed;
165 elapsed = (mtime_since_now(&td->epoch) + 999) / 1000;
166 timeout = td->o.timeout / 1000000UL;
168 bytes_total = td->total_io_size;
170 if (td->flags & TD_F_NO_PROGRESS)
173 if (td->o.fill_device && td->o.size == -1ULL) {
174 if (!td->fill_device_size || td->fill_device_size == -1ULL)
177 bytes_total = td->fill_device_size;
180 if (td->o.zone_size && td->o.zone_skip && bytes_total) {
181 unsigned int nr_zones;
184 zone_bytes = bytes_total + td->o.zone_size + td->o.zone_skip;
185 nr_zones = (zone_bytes - 1) / (td->o.zone_size + td->o.zone_skip);
186 bytes_total -= nr_zones * td->o.zone_skip;
190 * if writing and verifying afterwards, bytes_total will be twice the
191 * size. In a mixed workload, verify phase will be the size of the
192 * first stage writes.
194 if (td->o.do_verify && td->o.verify && td_write(td)) {
196 unsigned int perc = 50;
198 if (td->o.rwmix[DDIR_WRITE])
199 perc = td->o.rwmix[DDIR_WRITE];
201 bytes_total += (bytes_total * perc) / 100;
206 if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING) {
209 bytes_done = ddir_rw_sum(td->io_bytes);
212 perc = (double) bytes_done / (double) bytes_total;
218 if (td->o.time_based) {
220 perc_t = (double) elapsed / (double) timeout;
225 * Will never hit, we can't have time_based
226 * without a timeout set.
235 eta_sec = (unsigned long) (elapsed * (1.0 / perc)) - elapsed;
239 eta_sec > (timeout + done_secs - elapsed))
240 eta_sec = timeout + done_secs - elapsed;
241 } else if (td->runstate == TD_NOT_CREATED || td->runstate == TD_CREATED
242 || td->runstate == TD_INITIALIZED
243 || td->runstate == TD_SETTING_UP
244 || td->runstate == TD_RAMP
245 || td->runstate == TD_PRE_READING) {
246 int64_t t_eta = 0, r_eta = 0;
247 unsigned long long rate_bytes;
250 * We can only guess - assume it'll run the full timeout
251 * if given, otherwise assume it'll run at the specified rate.
254 uint64_t __timeout = td->o.timeout;
255 uint64_t start_delay = td->o.start_delay;
256 uint64_t ramp_time = td->o.ramp_time;
258 t_eta = __timeout + start_delay;
259 if (!td->ramp_time_over) {
264 if ((td->runstate == TD_RAMP) && in_ramp_time(td)) {
265 unsigned long ramp_left;
267 ramp_left = mtime_since_now(&td->epoch);
268 ramp_left = (ramp_left + 999) / 1000;
269 if (ramp_left <= t_eta)
275 rate_bytes = td->o.rate[DDIR_READ];
277 rate_bytes += td->o.rate[DDIR_WRITE];
279 rate_bytes += td->o.rate[DDIR_TRIM];
282 r_eta = bytes_total / rate_bytes;
283 r_eta += (td->o.start_delay / 1000000ULL);
287 eta_sec = min(r_eta, t_eta);
296 * thread is already done or waiting for fsync
304 static void calc_rate(int unified_rw_rep, unsigned long mtime,
305 unsigned long long *io_bytes,
306 unsigned long long *prev_io_bytes, uint64_t *rate)
310 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
311 unsigned long long diff, this_rate;
313 diff = io_bytes[i] - prev_io_bytes[i];
315 this_rate = ((1000 * diff) / mtime) / 1024; /* KiB/s */
319 if (unified_rw_rep) {
321 rate[0] += this_rate;
325 prev_io_bytes[i] = io_bytes[i];
329 static void calc_iops(int unified_rw_rep, unsigned long mtime,
330 unsigned long long *io_iops,
331 unsigned long long *prev_io_iops, unsigned int *iops)
335 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
336 unsigned long long diff, this_iops;
338 diff = io_iops[i] - prev_io_iops[i];
340 this_iops = (diff * 1000) / mtime;
344 if (unified_rw_rep) {
346 iops[0] += this_iops;
350 prev_io_iops[i] = io_iops[i];
355 * Allow a little slack - if we're within 95% of the time, allow ETA.
357 bool eta_time_within_slack(unsigned int time)
359 return time > ((eta_interval_msec * 95) / 100);
363 * Print status of the jobs we know about. This includes rate estimates,
364 * ETA, thread state, etc.
366 bool calc_thread_status(struct jobs_eta *je, int force)
368 struct thread_data *td;
369 int i, unified_rw_rep;
370 uint64_t rate_time, disp_time, bw_avg_time, *eta_secs;
371 unsigned long long io_bytes[DDIR_RWDIR_CNT];
372 unsigned long long io_iops[DDIR_RWDIR_CNT];
375 static unsigned long long rate_io_bytes[DDIR_RWDIR_CNT];
376 static unsigned long long disp_io_bytes[DDIR_RWDIR_CNT];
377 static unsigned long long disp_io_iops[DDIR_RWDIR_CNT];
378 static struct timespec rate_prev_time, disp_prev_time;
381 if (!(output_format & FIO_OUTPUT_NORMAL) &&
384 if (temp_stall_ts || eta_print == FIO_ETA_NEVER)
387 if (!isatty(STDOUT_FILENO) && (eta_print != FIO_ETA_ALWAYS))
391 if (!ddir_rw_sum(rate_io_bytes))
392 fill_start_time(&rate_prev_time);
393 if (!ddir_rw_sum(disp_io_bytes))
394 fill_start_time(&disp_prev_time);
396 eta_secs = malloc(thread_number * sizeof(uint64_t));
397 memset(eta_secs, 0, thread_number * sizeof(uint64_t));
399 je->elapsed_sec = (mtime_since_genesis() + 999) / 1000;
401 io_bytes[DDIR_READ] = io_bytes[DDIR_WRITE] = io_bytes[DDIR_TRIM] = 0;
402 io_iops[DDIR_READ] = io_iops[DDIR_WRITE] = io_iops[DDIR_TRIM] = 0;
403 bw_avg_time = ULONG_MAX;
406 unified_rw_rep += td->o.unified_rw_rep;
407 if (is_power_of_2(td->o.kb_base))
409 je->unit_base = td->o.unit_base;
410 if (td->o.bw_avg_time < bw_avg_time)
411 bw_avg_time = td->o.bw_avg_time;
412 if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING
413 || td->runstate == TD_FSYNCING
414 || td->runstate == TD_PRE_READING
415 || td->runstate == TD_FINISHING) {
418 je->t_rate[0] += td->o.rate[DDIR_READ];
419 je->t_iops[0] += td->o.rate_iops[DDIR_READ];
420 je->m_rate[0] += td->o.ratemin[DDIR_READ];
421 je->m_iops[0] += td->o.rate_iops_min[DDIR_READ];
424 je->t_rate[1] += td->o.rate[DDIR_WRITE];
425 je->t_iops[1] += td->o.rate_iops[DDIR_WRITE];
426 je->m_rate[1] += td->o.ratemin[DDIR_WRITE];
427 je->m_iops[1] += td->o.rate_iops_min[DDIR_WRITE];
430 je->t_rate[2] += td->o.rate[DDIR_TRIM];
431 je->t_iops[2] += td->o.rate_iops[DDIR_TRIM];
432 je->m_rate[2] += td->o.ratemin[DDIR_TRIM];
433 je->m_iops[2] += td->o.rate_iops_min[DDIR_TRIM];
436 je->files_open += td->nr_open_files;
437 } else if (td->runstate == TD_RAMP) {
440 } else if (td->runstate == TD_SETTING_UP)
442 else if (td->runstate < TD_RUNNING)
445 if (je->elapsed_sec >= 3)
446 eta_secs[i] = thread_eta(td);
448 eta_secs[i] = INT_MAX;
450 check_str_update(td);
452 if (td->runstate > TD_SETTING_UP) {
455 for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) {
456 if (unified_rw_rep) {
457 io_bytes[0] += td->io_bytes[ddir];
458 io_iops[0] += td->io_blocks[ddir];
460 io_bytes[ddir] += td->io_bytes[ddir];
461 io_iops[ddir] += td->io_blocks[ddir];
467 if (exitall_on_terminate) {
468 je->eta_sec = INT_MAX;
470 if (eta_secs[i] < je->eta_sec)
471 je->eta_sec = eta_secs[i];
474 unsigned long eta_stone = 0;
478 if ((td->runstate == TD_NOT_CREATED) && td->o.stonewall)
479 eta_stone += eta_secs[i];
481 if (eta_secs[i] > je->eta_sec)
482 je->eta_sec = eta_secs[i];
485 je->eta_sec += eta_stone;
490 fio_gettime(&now, NULL);
491 rate_time = mtime_since(&rate_prev_time, &now);
493 if (write_bw_log && rate_time > bw_avg_time && !in_ramp_time(td)) {
494 calc_rate(unified_rw_rep, rate_time, io_bytes, rate_io_bytes,
496 memcpy(&rate_prev_time, &now, sizeof(now));
497 add_agg_sample(sample_val(je->rate[DDIR_READ]), DDIR_READ, 0);
498 add_agg_sample(sample_val(je->rate[DDIR_WRITE]), DDIR_WRITE, 0);
499 add_agg_sample(sample_val(je->rate[DDIR_TRIM]), DDIR_TRIM, 0);
502 disp_time = mtime_since(&disp_prev_time, &now);
504 if (!force && !eta_time_within_slack(disp_time))
507 calc_rate(unified_rw_rep, disp_time, io_bytes, disp_io_bytes, je->rate);
508 calc_iops(unified_rw_rep, disp_time, io_iops, disp_io_iops, je->iops);
510 memcpy(&disp_prev_time, &now, sizeof(now));
512 if (!force && !je->nr_running && !je->nr_pending)
515 je->nr_threads = thread_number;
516 update_condensed_str(__run_str, run_str);
517 memcpy(je->run_str, run_str, strlen(run_str));
521 void display_thread_status(struct jobs_eta *je)
523 static struct timespec disp_eta_new_line;
524 static int eta_new_line_init, eta_new_line_pending;
525 static int linelen_last;
527 char output[__THREAD_RUNSTR_SZ(REAL_MAX_JOBS) + 512], *p = output;
531 if (je->eta_sec != INT_MAX && je->elapsed_sec) {
532 perc = (double) je->elapsed_sec / (double) (je->elapsed_sec + je->eta_sec);
533 eta_to_str(eta_str, je->eta_sec);
536 if (eta_new_line_pending) {
537 eta_new_line_pending = 0;
539 p += sprintf(p, "\n");
542 p += sprintf(p, "Jobs: %d (f=%d)", je->nr_running, je->files_open);
544 /* rate limits, if any */
545 if (je->m_rate[0] || je->m_rate[1] || je->m_rate[2] ||
546 je->t_rate[0] || je->t_rate[1] || je->t_rate[2]) {
549 mr = num2str(je->m_rate[0] + je->m_rate[1] + je->m_rate[2],
550 je->sig_figs, 0, je->is_pow2, N2S_BYTEPERSEC);
551 tr = num2str(je->t_rate[0] + je->t_rate[1] + je->t_rate[2],
552 je->sig_figs, 0, je->is_pow2, N2S_BYTEPERSEC);
554 p += sprintf(p, ", %s-%s", mr, tr);
557 } else if (je->m_iops[0] || je->m_iops[1] || je->m_iops[2] ||
558 je->t_iops[0] || je->t_iops[1] || je->t_iops[2]) {
559 p += sprintf(p, ", %d-%d IOPS",
560 je->m_iops[0] + je->m_iops[1] + je->m_iops[2],
561 je->t_iops[0] + je->t_iops[1] + je->t_iops[2]);
564 /* current run string, % done, bandwidth, iops, eta */
565 if (je->eta_sec != INT_MAX && je->nr_running) {
567 char *iops_str[DDIR_RWDIR_CNT];
568 char *rate_str[DDIR_RWDIR_CNT];
574 if ((!je->eta_sec && !eta_good) || je->nr_ramp == je->nr_running ||
576 strcpy(perc_str, "-.-%");
580 if (je->nr_setting_up && je->nr_running)
581 mult *= (1.0 - (double) je->nr_setting_up / (double) je->nr_running);
585 sprintf(perc_str, "%3.1f%%", perc);
588 for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) {
589 rate_str[ddir] = num2str(je->rate[ddir], 4,
590 1024, je->is_pow2, je->unit_base);
591 iops_str[ddir] = num2str(je->iops[ddir], 4, 1, 0, N2S_NONE);
594 left = sizeof(output) - (p - output) - 1;
596 if (je->rate[DDIR_TRIM] || je->iops[DDIR_TRIM])
597 l = snprintf(p, left,
598 ": [%s][%s][r=%s,w=%s,t=%s][r=%s,w=%s,t=%s IOPS][eta %s]",
599 je->run_str, perc_str, rate_str[DDIR_READ],
600 rate_str[DDIR_WRITE], rate_str[DDIR_TRIM],
601 iops_str[DDIR_READ], iops_str[DDIR_WRITE],
602 iops_str[DDIR_TRIM], eta_str);
604 l = snprintf(p, left,
605 ": [%s][%s][r=%s,w=%s][r=%s,w=%s IOPS][eta %s]",
606 je->run_str, perc_str,
607 rate_str[DDIR_READ], rate_str[DDIR_WRITE],
608 iops_str[DDIR_READ], iops_str[DDIR_WRITE],
610 /* If truncation occurred adjust l so p is on the null */
614 linelen = p - output;
615 if (l >= 0 && linelen < linelen_last)
616 p += sprintf(p, "%*s", linelen_last - linelen, "");
617 linelen_last = linelen;
619 for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) {
620 free(rate_str[ddir]);
621 free(iops_str[ddir]);
626 printf("%s", output);
628 if (!eta_new_line_init) {
629 fio_gettime(&disp_eta_new_line, NULL);
630 eta_new_line_init = 1;
631 } else if (eta_new_line && mtime_since_now(&disp_eta_new_line) > eta_new_line) {
632 fio_gettime(&disp_eta_new_line, NULL);
633 eta_new_line_pending = 1;
639 struct jobs_eta *get_jobs_eta(bool force, size_t *size)
646 *size = sizeof(*je) + THREAD_RUNSTR_SZ + 8;
650 memset(je, 0, *size);
652 if (!calc_thread_status(je, force)) {
657 *size = sizeof(*je) + strlen((char *) je->run_str) + 1;
661 void print_thread_status(void)
666 je = get_jobs_eta(false, &size);
668 display_thread_status(je);
673 void print_status_init(int thr_number)
675 DRD_IGNORE_VAR(__run_str);
676 __run_str[thr_number] = 'P';
677 update_condensed_str(__run_str, run_str);