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 int 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)
262 rate_bytes = ddir_rw_sum(td->o.rate);
264 r_eta = (bytes_total / 1024) / rate_bytes;
265 r_eta += (td->o.start_delay / 1000000ULL);
269 eta_sec = min(r_eta, t_eta);
278 * thread is already done or waiting for fsync
286 static void calc_rate(int unified_rw_rep, unsigned long mtime,
287 unsigned long long *io_bytes,
288 unsigned long long *prev_io_bytes, unsigned int *rate)
292 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
293 unsigned long long diff, this_rate;
295 diff = io_bytes[i] - prev_io_bytes[i];
297 this_rate = ((1000 * diff) / mtime) / 1024;
301 if (unified_rw_rep) {
303 rate[0] += this_rate;
307 prev_io_bytes[i] = io_bytes[i];
311 static void calc_iops(int unified_rw_rep, unsigned long mtime,
312 unsigned long long *io_iops,
313 unsigned long long *prev_io_iops, unsigned int *iops)
317 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
318 unsigned long long diff, this_iops;
320 diff = io_iops[i] - prev_io_iops[i];
322 this_iops = (diff * 1000) / mtime;
326 if (unified_rw_rep) {
328 iops[0] += this_iops;
332 prev_io_iops[i] = io_iops[i];
337 * Print status of the jobs we know about. This includes rate estimates,
338 * ETA, thread state, etc.
340 int calc_thread_status(struct jobs_eta *je, int force)
342 struct thread_data *td;
343 int i, unified_rw_rep;
344 unsigned long rate_time, disp_time, bw_avg_time, *eta_secs;
345 unsigned long long io_bytes[DDIR_RWDIR_CNT];
346 unsigned long long io_iops[DDIR_RWDIR_CNT];
349 static unsigned long long rate_io_bytes[DDIR_RWDIR_CNT];
350 static unsigned long long disp_io_bytes[DDIR_RWDIR_CNT];
351 static unsigned long long disp_io_iops[DDIR_RWDIR_CNT];
352 static struct timeval rate_prev_time, disp_prev_time;
355 if (!(output_format & FIO_OUTPUT_NORMAL) &&
358 if (temp_stall_ts || eta_print == FIO_ETA_NEVER)
361 if (!isatty(STDOUT_FILENO) && (eta_print != FIO_ETA_ALWAYS))
365 if (!ddir_rw_sum(rate_io_bytes))
366 fill_start_time(&rate_prev_time);
367 if (!ddir_rw_sum(disp_io_bytes))
368 fill_start_time(&disp_prev_time);
370 eta_secs = malloc(thread_number * sizeof(unsigned long));
371 memset(eta_secs, 0, thread_number * sizeof(unsigned long));
373 je->elapsed_sec = (mtime_since_genesis() + 999) / 1000;
375 io_bytes[DDIR_READ] = io_bytes[DDIR_WRITE] = io_bytes[DDIR_TRIM] = 0;
376 io_iops[DDIR_READ] = io_iops[DDIR_WRITE] = io_iops[DDIR_TRIM] = 0;
377 bw_avg_time = ULONG_MAX;
380 unified_rw_rep += td->o.unified_rw_rep;
381 if (is_power_of_2(td->o.kb_base))
383 je->unit_base = td->o.unit_base;
384 if (td->o.bw_avg_time < bw_avg_time)
385 bw_avg_time = td->o.bw_avg_time;
386 if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING
387 || td->runstate == TD_FSYNCING
388 || td->runstate == TD_PRE_READING
389 || td->runstate == TD_FINISHING) {
392 je->t_rate[0] += td->o.rate[DDIR_READ];
393 je->t_iops[0] += td->o.rate_iops[DDIR_READ];
394 je->m_rate[0] += td->o.ratemin[DDIR_READ];
395 je->m_iops[0] += td->o.rate_iops_min[DDIR_READ];
398 je->t_rate[1] += td->o.rate[DDIR_WRITE];
399 je->t_iops[1] += td->o.rate_iops[DDIR_WRITE];
400 je->m_rate[1] += td->o.ratemin[DDIR_WRITE];
401 je->m_iops[1] += td->o.rate_iops_min[DDIR_WRITE];
404 je->t_rate[2] += td->o.rate[DDIR_TRIM];
405 je->t_iops[2] += td->o.rate_iops[DDIR_TRIM];
406 je->m_rate[2] += td->o.ratemin[DDIR_TRIM];
407 je->m_iops[2] += td->o.rate_iops_min[DDIR_TRIM];
410 je->files_open += td->nr_open_files;
411 } else if (td->runstate == TD_RAMP) {
414 } else if (td->runstate == TD_SETTING_UP)
416 else if (td->runstate < TD_RUNNING)
419 if (je->elapsed_sec >= 3)
420 eta_secs[i] = thread_eta(td);
422 eta_secs[i] = INT_MAX;
424 check_str_update(td);
426 if (td->runstate > TD_SETTING_UP) {
429 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
430 if (unified_rw_rep) {
431 io_bytes[0] += td->io_bytes[ddir];
432 io_iops[0] += td->io_blocks[ddir];
434 io_bytes[ddir] += td->io_bytes[ddir];
435 io_iops[ddir] += td->io_blocks[ddir];
441 if (exitall_on_terminate) {
442 je->eta_sec = INT_MAX;
444 if (eta_secs[i] < je->eta_sec)
445 je->eta_sec = eta_secs[i];
448 unsigned long eta_stone = 0;
452 if ((td->runstate == TD_NOT_CREATED) && td->o.stonewall)
453 eta_stone += eta_secs[i];
455 if (eta_secs[i] > je->eta_sec)
456 je->eta_sec = eta_secs[i];
459 je->eta_sec += eta_stone;
464 fio_gettime(&now, NULL);
465 rate_time = mtime_since(&rate_prev_time, &now);
467 if (write_bw_log && rate_time > bw_avg_time && !in_ramp_time(td)) {
468 calc_rate(unified_rw_rep, rate_time, io_bytes, rate_io_bytes,
470 memcpy(&rate_prev_time, &now, sizeof(now));
471 add_agg_sample(je->rate[DDIR_READ], DDIR_READ, 0);
472 add_agg_sample(je->rate[DDIR_WRITE], DDIR_WRITE, 0);
473 add_agg_sample(je->rate[DDIR_TRIM], DDIR_TRIM, 0);
476 disp_time = mtime_since(&disp_prev_time, &now);
479 * Allow a little slack, the target is to print it every 1000 msecs
481 if (!force && disp_time < 900)
484 calc_rate(unified_rw_rep, disp_time, io_bytes, disp_io_bytes, je->rate);
485 calc_iops(unified_rw_rep, disp_time, io_iops, disp_io_iops, je->iops);
487 memcpy(&disp_prev_time, &now, sizeof(now));
489 if (!force && !je->nr_running && !je->nr_pending)
492 je->nr_threads = thread_number;
493 update_condensed_str(__run_str, run_str);
494 memcpy(je->run_str, run_str, strlen(run_str));
498 void display_thread_status(struct jobs_eta *je)
500 static struct timeval disp_eta_new_line;
501 static int eta_new_line_init, eta_new_line_pending;
502 static int linelen_last;
504 char output[REAL_MAX_JOBS + 512], *p = output;
508 if (je->eta_sec != INT_MAX && je->elapsed_sec) {
509 perc = (double) je->elapsed_sec / (double) (je->elapsed_sec + je->eta_sec);
510 eta_to_str(eta_str, je->eta_sec);
513 if (eta_new_line_pending) {
514 eta_new_line_pending = 0;
515 p += sprintf(p, "\n");
518 p += sprintf(p, "Jobs: %d (f=%d)", je->nr_running, je->files_open);
519 if (je->m_rate[0] || je->m_rate[1] || je->t_rate[0] || je->t_rate[1]) {
522 mr = num2str(je->m_rate[0] + je->m_rate[1], 4, 0, je->is_pow2, 8);
523 tr = num2str(je->t_rate[0] + je->t_rate[1], 4, 0, je->is_pow2, 8);
524 p += sprintf(p, ", CR=%s/%s KB/s", tr, mr);
527 } else if (je->m_iops[0] || je->m_iops[1] || je->t_iops[0] || je->t_iops[1]) {
528 p += sprintf(p, ", CR=%d/%d IOPS",
529 je->t_iops[0] + je->t_iops[1],
530 je->m_iops[0] + je->m_iops[1]);
532 if (je->eta_sec != INT_MAX && je->nr_running) {
534 char *iops_str[DDIR_RWDIR_CNT];
535 char *rate_str[DDIR_RWDIR_CNT];
540 if ((!je->eta_sec && !eta_good) || je->nr_ramp == je->nr_running ||
542 strcpy(perc_str, "-.-% done");
546 if (je->nr_setting_up && je->nr_running)
547 mult *= (1.0 - (double) je->nr_setting_up / (double) je->nr_running);
551 sprintf(perc_str, "%3.1f%% done", perc);
554 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
555 rate_str[ddir] = num2str(je->rate[ddir], 5,
556 1024, je->is_pow2, je->unit_base);
557 iops_str[ddir] = num2str(je->iops[ddir], 4, 1, 0, 0);
560 left = sizeof(output) - (p - output) - 1;
562 l = snprintf(p, left, ": [%s] [%s] [%s/%s/%s /s] [%s/%s/%s iops] [eta %s]",
563 je->run_str, perc_str, rate_str[DDIR_READ],
564 rate_str[DDIR_WRITE], rate_str[DDIR_TRIM],
565 iops_str[DDIR_READ], iops_str[DDIR_WRITE],
566 iops_str[DDIR_TRIM], eta_str);
568 if (l >= 0 && l < linelen_last)
569 p += sprintf(p, "%*s", linelen_last - l, "");
572 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
573 free(rate_str[ddir]);
574 free(iops_str[ddir]);
577 p += sprintf(p, "\r");
579 printf("%s", output);
581 if (!eta_new_line_init) {
582 fio_gettime(&disp_eta_new_line, NULL);
583 eta_new_line_init = 1;
584 } else if (eta_new_line && mtime_since_now(&disp_eta_new_line) > eta_new_line) {
585 fio_gettime(&disp_eta_new_line, NULL);
586 eta_new_line_pending = 1;
592 struct jobs_eta *get_jobs_eta(bool force, size_t *size)
599 *size = sizeof(*je) + THREAD_RUNSTR_SZ + 8;
603 memset(je, 0, *size);
605 if (!calc_thread_status(je, force)) {
610 *size = sizeof(*je) + strlen((char *) je->run_str) + 1;
614 void print_thread_status(void)
619 je = get_jobs_eta(false, &size);
621 display_thread_status(je);
626 void print_status_init(int thr_number)
628 __run_str[thr_number] = 'P';
629 update_condensed_str(__run_str, run_str);