11 static char __run_str[REAL_MAX_JOBS + 1];
12 static char run_str[__THREAD_RUNSTR_SZ(REAL_MAX_JOBS)];
14 static void update_condensed_str(char *rstr, char *run_str_condensed)
20 *run_str_condensed++ = *rstr++;
21 while (*(rstr - 1) == *rstr) {
25 run_str_condensed += sprintf(run_str_condensed, "(%u),", nr);
29 *run_str_condensed = '\0';
33 * Sets the status of the 'td' in the printed status map.
35 static void check_str_update(struct thread_data *td)
37 char c = __run_str[td->thread_number - 1];
39 switch (td->runstate) {
57 if (td->o.rwmix[DDIR_READ] == 100)
59 else if (td->o.rwmix[DDIR_WRITE] == 100)
64 if (td->o.rwmix[DDIR_READ] == 100)
66 else if (td->o.rwmix[DDIR_WRITE] == 100)
71 } else if (td_read(td)) {
76 } else if (td_write(td)) {
111 log_err("state %d\n", td->runstate);
114 __run_str[td->thread_number - 1] = c;
115 update_condensed_str(__run_str, run_str);
119 * Convert seconds to a printable string.
121 void eta_to_str(char *str, unsigned long eta_sec)
123 unsigned int d, h, m, s;
141 str += sprintf(str, "%02ud:", d);
145 str += sprintf(str, "%02uh:", h);
147 str += sprintf(str, "%02um:", m);
148 str += sprintf(str, "%02us", s);
152 * Best effort calculation of the estimated pending runtime of a job.
154 static unsigned long thread_eta(struct thread_data *td)
156 unsigned long long bytes_total, bytes_done;
157 unsigned long eta_sec = 0;
158 unsigned long elapsed;
161 elapsed = (mtime_since_now(&td->epoch) + 999) / 1000;
162 timeout = td->o.timeout / 1000000UL;
164 bytes_total = td->total_io_size;
166 if (td->flags & TD_F_NO_PROGRESS)
169 if (td->o.fill_device && td->o.size == -1ULL) {
170 if (!td->fill_device_size || td->fill_device_size == -1ULL)
173 bytes_total = td->fill_device_size;
176 if (td->o.zone_size && td->o.zone_skip && bytes_total) {
177 unsigned int nr_zones;
180 zone_bytes = bytes_total + td->o.zone_size + td->o.zone_skip;
181 nr_zones = (zone_bytes - 1) / (td->o.zone_size + td->o.zone_skip);
182 bytes_total -= nr_zones * td->o.zone_skip;
186 * if writing and verifying afterwards, bytes_total will be twice the
187 * size. In a mixed workload, verify phase will be the size of the
188 * first stage writes.
190 if (td->o.do_verify && td->o.verify && td_write(td)) {
192 unsigned int perc = 50;
194 if (td->o.rwmix[DDIR_WRITE])
195 perc = td->o.rwmix[DDIR_WRITE];
197 bytes_total += (bytes_total * perc) / 100;
202 if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING) {
205 bytes_done = ddir_rw_sum(td->io_bytes);
208 perc = (double) bytes_done / (double) bytes_total;
214 if (td->o.time_based) {
216 perc_t = (double) elapsed / (double) timeout;
221 * Will never hit, we can't have time_based
222 * without a timeout set.
228 eta_sec = (unsigned long) (elapsed * (1.0 / perc)) - elapsed;
231 eta_sec > (timeout + done_secs - elapsed))
232 eta_sec = timeout + done_secs - elapsed;
233 } else if (td->runstate == TD_NOT_CREATED || td->runstate == TD_CREATED
234 || td->runstate == TD_INITIALIZED
235 || td->runstate == TD_SETTING_UP
236 || td->runstate == TD_RAMP
237 || td->runstate == TD_PRE_READING) {
238 int64_t t_eta = 0, r_eta = 0;
239 unsigned long long rate_bytes;
242 * We can only guess - assume it'll run the full timeout
243 * if given, otherwise assume it'll run at the specified rate.
246 uint64_t __timeout = td->o.timeout;
247 uint64_t start_delay = td->o.start_delay;
248 uint64_t ramp_time = td->o.ramp_time;
250 t_eta = __timeout + start_delay + ramp_time;
253 if ((td->runstate == TD_RAMP) && in_ramp_time(td)) {
254 unsigned long ramp_left;
256 ramp_left = mtime_since_now(&td->epoch);
257 ramp_left = (ramp_left + 999) / 1000;
258 if (ramp_left <= t_eta)
264 rate_bytes = td->o.rate[DDIR_READ];
266 rate_bytes += td->o.rate[DDIR_WRITE];
268 rate_bytes += td->o.rate[DDIR_TRIM];
271 r_eta = bytes_total / rate_bytes;
272 r_eta += (td->o.start_delay / 1000000ULL);
276 eta_sec = min(r_eta, t_eta);
285 * thread is already done or waiting for fsync
293 static void calc_rate(int unified_rw_rep, unsigned long mtime,
294 unsigned long long *io_bytes,
295 unsigned long long *prev_io_bytes, uint64_t *rate)
299 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
300 unsigned long long diff, this_rate;
302 diff = io_bytes[i] - prev_io_bytes[i];
304 this_rate = ((1000 * diff) / mtime) / 1024;
308 if (unified_rw_rep) {
310 rate[0] += this_rate;
314 prev_io_bytes[i] = io_bytes[i];
318 static void calc_iops(int unified_rw_rep, unsigned long mtime,
319 unsigned long long *io_iops,
320 unsigned long long *prev_io_iops, unsigned int *iops)
324 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
325 unsigned long long diff, this_iops;
327 diff = io_iops[i] - prev_io_iops[i];
329 this_iops = (diff * 1000) / mtime;
333 if (unified_rw_rep) {
335 iops[0] += this_iops;
339 prev_io_iops[i] = io_iops[i];
344 * Print status of the jobs we know about. This includes rate estimates,
345 * ETA, thread state, etc.
347 bool calc_thread_status(struct jobs_eta *je, int force)
349 struct thread_data *td;
350 int i, unified_rw_rep;
351 uint64_t rate_time, disp_time, bw_avg_time, *eta_secs;
352 unsigned long long io_bytes[DDIR_RWDIR_CNT];
353 unsigned long long io_iops[DDIR_RWDIR_CNT];
356 static unsigned long long rate_io_bytes[DDIR_RWDIR_CNT];
357 static unsigned long long disp_io_bytes[DDIR_RWDIR_CNT];
358 static unsigned long long disp_io_iops[DDIR_RWDIR_CNT];
359 static struct timeval rate_prev_time, disp_prev_time;
362 if (!(output_format & FIO_OUTPUT_NORMAL) &&
365 if (temp_stall_ts || eta_print == FIO_ETA_NEVER)
368 if (!isatty(STDOUT_FILENO) && (eta_print != FIO_ETA_ALWAYS))
372 if (!ddir_rw_sum(rate_io_bytes))
373 fill_start_time(&rate_prev_time);
374 if (!ddir_rw_sum(disp_io_bytes))
375 fill_start_time(&disp_prev_time);
377 eta_secs = malloc(thread_number * sizeof(uint64_t));
378 memset(eta_secs, 0, thread_number * sizeof(uint64_t));
380 je->elapsed_sec = (mtime_since_genesis() + 999) / 1000;
382 io_bytes[DDIR_READ] = io_bytes[DDIR_WRITE] = io_bytes[DDIR_TRIM] = 0;
383 io_iops[DDIR_READ] = io_iops[DDIR_WRITE] = io_iops[DDIR_TRIM] = 0;
384 bw_avg_time = ULONG_MAX;
387 unified_rw_rep += td->o.unified_rw_rep;
388 if (is_power_of_2(td->o.kb_base))
390 je->unit_base = td->o.unit_base;
391 if (td->o.bw_avg_time < bw_avg_time)
392 bw_avg_time = td->o.bw_avg_time;
393 if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING
394 || td->runstate == TD_FSYNCING
395 || td->runstate == TD_PRE_READING
396 || td->runstate == TD_FINISHING) {
399 je->t_rate[0] += td->o.rate[DDIR_READ];
400 je->t_iops[0] += td->o.rate_iops[DDIR_READ];
401 je->m_rate[0] += td->o.ratemin[DDIR_READ];
402 je->m_iops[0] += td->o.rate_iops_min[DDIR_READ];
405 je->t_rate[1] += td->o.rate[DDIR_WRITE];
406 je->t_iops[1] += td->o.rate_iops[DDIR_WRITE];
407 je->m_rate[1] += td->o.ratemin[DDIR_WRITE];
408 je->m_iops[1] += td->o.rate_iops_min[DDIR_WRITE];
411 je->t_rate[2] += td->o.rate[DDIR_TRIM];
412 je->t_iops[2] += td->o.rate_iops[DDIR_TRIM];
413 je->m_rate[2] += td->o.ratemin[DDIR_TRIM];
414 je->m_iops[2] += td->o.rate_iops_min[DDIR_TRIM];
417 je->files_open += td->nr_open_files;
418 } else if (td->runstate == TD_RAMP) {
421 } else if (td->runstate == TD_SETTING_UP)
423 else if (td->runstate < TD_RUNNING)
426 if (je->elapsed_sec >= 3)
427 eta_secs[i] = thread_eta(td);
429 eta_secs[i] = INT_MAX;
431 check_str_update(td);
433 if (td->runstate > TD_SETTING_UP) {
436 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
437 if (unified_rw_rep) {
438 io_bytes[0] += td->io_bytes[ddir];
439 io_iops[0] += td->io_blocks[ddir];
441 io_bytes[ddir] += td->io_bytes[ddir];
442 io_iops[ddir] += td->io_blocks[ddir];
448 if (exitall_on_terminate) {
449 je->eta_sec = INT_MAX;
451 if (eta_secs[i] < je->eta_sec)
452 je->eta_sec = eta_secs[i];
455 unsigned long eta_stone = 0;
459 if ((td->runstate == TD_NOT_CREATED) && td->o.stonewall)
460 eta_stone += eta_secs[i];
462 if (eta_secs[i] > je->eta_sec)
463 je->eta_sec = eta_secs[i];
466 je->eta_sec += eta_stone;
471 fio_gettime(&now, NULL);
472 rate_time = mtime_since(&rate_prev_time, &now);
474 if (write_bw_log && rate_time > bw_avg_time && !in_ramp_time(td)) {
475 calc_rate(unified_rw_rep, rate_time, io_bytes, rate_io_bytes,
477 memcpy(&rate_prev_time, &now, sizeof(now));
478 add_agg_sample(je->rate[DDIR_READ], DDIR_READ, 0);
479 add_agg_sample(je->rate[DDIR_WRITE], DDIR_WRITE, 0);
480 add_agg_sample(je->rate[DDIR_TRIM], DDIR_TRIM, 0);
483 disp_time = mtime_since(&disp_prev_time, &now);
486 * Allow a little slack, the target is to print it every 1000 msecs
488 if (!force && disp_time < 900)
491 calc_rate(unified_rw_rep, disp_time, io_bytes, disp_io_bytes, je->rate);
492 calc_iops(unified_rw_rep, disp_time, io_iops, disp_io_iops, je->iops);
494 memcpy(&disp_prev_time, &now, sizeof(now));
496 if (!force && !je->nr_running && !je->nr_pending)
499 je->nr_threads = thread_number;
500 update_condensed_str(__run_str, run_str);
501 memcpy(je->run_str, run_str, strlen(run_str));
505 void display_thread_status(struct jobs_eta *je)
507 static struct timeval disp_eta_new_line;
508 static int eta_new_line_init, eta_new_line_pending;
509 static int linelen_last;
511 char output[REAL_MAX_JOBS + 512], *p = output;
515 if (je->eta_sec != INT_MAX && je->elapsed_sec) {
516 perc = (double) je->elapsed_sec / (double) (je->elapsed_sec + je->eta_sec);
517 eta_to_str(eta_str, je->eta_sec);
520 if (eta_new_line_pending) {
521 eta_new_line_pending = 0;
522 p += sprintf(p, "\n");
525 p += sprintf(p, "Jobs: %d (f=%d)", je->nr_running, je->files_open);
526 if (je->m_rate[0] || je->m_rate[1] || je->t_rate[0] || je->t_rate[1]) {
529 mr = num2str(je->m_rate[0] + je->m_rate[1], 4, 0, je->is_pow2, 8);
530 tr = num2str(je->t_rate[0] + je->t_rate[1], 4, 0, je->is_pow2, 8);
531 p += sprintf(p, ", CR=%s/%s KB/s", tr, mr);
534 } else if (je->m_iops[0] || je->m_iops[1] || je->t_iops[0] || je->t_iops[1]) {
535 p += sprintf(p, ", CR=%d/%d IOPS",
536 je->t_iops[0] + je->t_iops[1],
537 je->m_iops[0] + je->m_iops[1]);
539 if (je->eta_sec != INT_MAX && je->nr_running) {
541 char *iops_str[DDIR_RWDIR_CNT];
542 char *rate_str[DDIR_RWDIR_CNT];
547 if ((!je->eta_sec && !eta_good) || je->nr_ramp == je->nr_running ||
549 strcpy(perc_str, "-.-% done");
553 if (je->nr_setting_up && je->nr_running)
554 mult *= (1.0 - (double) je->nr_setting_up / (double) je->nr_running);
558 sprintf(perc_str, "%3.1f%% done", perc);
561 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
562 rate_str[ddir] = num2str(je->rate[ddir], 5,
563 1024, je->is_pow2, je->unit_base);
564 iops_str[ddir] = num2str(je->iops[ddir], 4, 1, 0, 0);
567 left = sizeof(output) - (p - output) - 1;
569 l = snprintf(p, left, ": [%s] [%s] [%s/%s/%s /s] [%s/%s/%s iops] [eta %s]",
570 je->run_str, perc_str, rate_str[DDIR_READ],
571 rate_str[DDIR_WRITE], rate_str[DDIR_TRIM],
572 iops_str[DDIR_READ], iops_str[DDIR_WRITE],
573 iops_str[DDIR_TRIM], eta_str);
575 if (l >= 0 && l < linelen_last)
576 p += sprintf(p, "%*s", linelen_last - l, "");
579 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
580 free(rate_str[ddir]);
581 free(iops_str[ddir]);
584 p += sprintf(p, "\r");
586 printf("%s", output);
588 if (!eta_new_line_init) {
589 fio_gettime(&disp_eta_new_line, NULL);
590 eta_new_line_init = 1;
591 } else if (eta_new_line && mtime_since_now(&disp_eta_new_line) > eta_new_line) {
592 fio_gettime(&disp_eta_new_line, NULL);
593 eta_new_line_pending = 1;
599 struct jobs_eta *get_jobs_eta(bool force, size_t *size)
606 *size = sizeof(*je) + THREAD_RUNSTR_SZ + 8;
610 memset(je, 0, *size);
612 if (!calc_thread_status(je, force)) {
617 *size = sizeof(*je) + strlen((char *) je->run_str) + 1;
621 void print_thread_status(void)
626 je = get_jobs_eta(false, &size);
628 display_thread_status(je);
633 void print_status_init(int thr_number)
635 __run_str[thr_number] = 'P';
636 update_condensed_str(__run_str, run_str);