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;
136 str += sprintf(str, "%02ud:", d);
140 str += sprintf(str, "%02uh:", h);
142 str += sprintf(str, "%02um:", m);
143 str += sprintf(str, "%02us", s);
147 * Best effort calculation of the estimated pending runtime of a job.
149 static int thread_eta(struct thread_data *td)
151 unsigned long long bytes_total, bytes_done;
152 unsigned long eta_sec = 0;
153 unsigned long elapsed;
156 elapsed = (mtime_since_now(&td->epoch) + 999) / 1000;
157 timeout = td->o.timeout / 1000000UL;
159 bytes_total = td->total_io_size;
161 if (td->o.fill_device && td->o.size == -1ULL) {
162 if (!td->fill_device_size || td->fill_device_size == -1ULL)
165 bytes_total = td->fill_device_size;
168 if (td->o.zone_size && td->o.zone_skip && bytes_total) {
169 unsigned int nr_zones;
172 zone_bytes = bytes_total + td->o.zone_size + td->o.zone_skip;
173 nr_zones = (zone_bytes - 1) / (td->o.zone_size + td->o.zone_skip);
174 bytes_total -= nr_zones * td->o.zone_skip;
178 * if writing and verifying afterwards, bytes_total will be twice the
179 * size. In a mixed workload, verify phase will be the size of the
180 * first stage writes.
182 if (td->o.do_verify && td->o.verify && td_write(td)) {
184 unsigned int perc = 50;
186 if (td->o.rwmix[DDIR_WRITE])
187 perc = td->o.rwmix[DDIR_WRITE];
189 bytes_total += (bytes_total * perc) / 100;
194 if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING) {
197 bytes_done = ddir_rw_sum(td->io_bytes);
200 perc = (double) bytes_done / (double) bytes_total;
206 if (td->o.time_based) {
208 perc_t = (double) elapsed / (double) timeout;
213 * Will never hit, we can't have time_based
214 * without a timeout set.
220 eta_sec = (unsigned long) (elapsed * (1.0 / perc)) - elapsed;
223 eta_sec > (timeout + done_secs - elapsed))
224 eta_sec = timeout + done_secs - elapsed;
225 } else if (td->runstate == TD_NOT_CREATED || td->runstate == TD_CREATED
226 || td->runstate == TD_INITIALIZED
227 || td->runstate == TD_SETTING_UP
228 || td->runstate == TD_RAMP
229 || td->runstate == TD_PRE_READING) {
230 int t_eta = 0, r_eta = 0;
231 unsigned long long rate_bytes;
234 * We can only guess - assume it'll run the full timeout
235 * if given, otherwise assume it'll run at the specified rate.
238 uint64_t __timeout = td->o.timeout;
239 uint64_t start_delay = td->o.start_delay;
240 uint64_t ramp_time = td->o.ramp_time;
242 t_eta = __timeout + start_delay + ramp_time;
245 if (in_ramp_time(td)) {
246 unsigned long ramp_left;
248 ramp_left = mtime_since_now(&td->epoch);
249 ramp_left = (ramp_left + 999) / 1000;
250 if (ramp_left <= t_eta)
254 rate_bytes = ddir_rw_sum(td->o.rate);
256 r_eta = (bytes_total / 1024) / rate_bytes;
257 r_eta += (td->o.start_delay / 1000000ULL);
261 eta_sec = min(r_eta, t_eta);
270 * thread is already done or waiting for fsync
278 static void calc_rate(int unified_rw_rep, unsigned long mtime,
279 unsigned long long *io_bytes,
280 unsigned long long *prev_io_bytes, unsigned int *rate)
284 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
285 unsigned long long diff;
287 diff = io_bytes[i] - prev_io_bytes[i];
288 if (unified_rw_rep) {
290 rate[0] += ((1000 * diff) / mtime) / 1024;
292 rate[i] = ((1000 * diff) / mtime) / 1024;
294 prev_io_bytes[i] = io_bytes[i];
298 static void calc_iops(int unified_rw_rep, unsigned long mtime,
299 unsigned long long *io_iops,
300 unsigned long long *prev_io_iops, unsigned int *iops)
304 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
305 unsigned long long diff;
307 diff = io_iops[i] - prev_io_iops[i];
308 if (unified_rw_rep) {
310 iops[0] += (diff * 1000) / mtime;
312 iops[i] = (diff * 1000) / mtime;
314 prev_io_iops[i] = io_iops[i];
319 * Print status of the jobs we know about. This includes rate estimates,
320 * ETA, thread state, etc.
322 int calc_thread_status(struct jobs_eta *je, int force)
324 struct thread_data *td;
325 int i, unified_rw_rep;
326 unsigned long rate_time, disp_time, bw_avg_time, *eta_secs;
327 unsigned long long io_bytes[DDIR_RWDIR_CNT];
328 unsigned long long io_iops[DDIR_RWDIR_CNT];
331 static unsigned long long rate_io_bytes[DDIR_RWDIR_CNT];
332 static unsigned long long disp_io_bytes[DDIR_RWDIR_CNT];
333 static unsigned long long disp_io_iops[DDIR_RWDIR_CNT];
334 static struct timeval rate_prev_time, disp_prev_time;
337 if (output_format != FIO_OUTPUT_NORMAL &&
340 if (temp_stall_ts || eta_print == FIO_ETA_NEVER)
343 if (!isatty(STDOUT_FILENO) && (eta_print != FIO_ETA_ALWAYS))
347 if (!ddir_rw_sum(rate_io_bytes))
348 fill_start_time(&rate_prev_time);
349 if (!ddir_rw_sum(disp_io_bytes))
350 fill_start_time(&disp_prev_time);
352 eta_secs = malloc(thread_number * sizeof(unsigned long));
353 memset(eta_secs, 0, thread_number * sizeof(unsigned long));
355 je->elapsed_sec = (mtime_since_genesis() + 999) / 1000;
357 io_bytes[DDIR_READ] = io_bytes[DDIR_WRITE] = io_bytes[DDIR_TRIM] = 0;
358 io_iops[DDIR_READ] = io_iops[DDIR_WRITE] = io_iops[DDIR_TRIM] = 0;
359 bw_avg_time = ULONG_MAX;
362 unified_rw_rep += td->o.unified_rw_rep;
363 if (is_power_of_2(td->o.kb_base))
365 je->unit_base = td->o.unit_base;
366 if (td->o.bw_avg_time < bw_avg_time)
367 bw_avg_time = td->o.bw_avg_time;
368 if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING
369 || td->runstate == TD_FSYNCING
370 || td->runstate == TD_PRE_READING
371 || td->runstate == TD_FINISHING) {
374 je->t_rate[0] += td->o.rate[DDIR_READ];
375 je->t_iops[0] += td->o.rate_iops[DDIR_READ];
376 je->m_rate[0] += td->o.ratemin[DDIR_READ];
377 je->m_iops[0] += td->o.rate_iops_min[DDIR_READ];
380 je->t_rate[1] += td->o.rate[DDIR_WRITE];
381 je->t_iops[1] += td->o.rate_iops[DDIR_WRITE];
382 je->m_rate[1] += td->o.ratemin[DDIR_WRITE];
383 je->m_iops[1] += td->o.rate_iops_min[DDIR_WRITE];
386 je->t_rate[2] += td->o.rate[DDIR_TRIM];
387 je->t_iops[2] += td->o.rate_iops[DDIR_TRIM];
388 je->m_rate[2] += td->o.ratemin[DDIR_TRIM];
389 je->m_iops[2] += td->o.rate_iops_min[DDIR_TRIM];
392 je->files_open += td->nr_open_files;
393 } else if (td->runstate == TD_RAMP) {
396 } else if (td->runstate == TD_SETTING_UP)
398 else if (td->runstate < TD_RUNNING)
401 if (je->elapsed_sec >= 3)
402 eta_secs[i] = thread_eta(td);
404 eta_secs[i] = INT_MAX;
406 check_str_update(td);
408 if (td->runstate > TD_SETTING_UP) {
411 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
412 if (unified_rw_rep) {
413 io_bytes[0] += td->io_bytes[ddir];
414 io_iops[0] += td->io_blocks[ddir];
416 io_bytes[ddir] += td->io_bytes[ddir];
417 io_iops[ddir] += td->io_blocks[ddir];
423 if (exitall_on_terminate)
424 je->eta_sec = INT_MAX;
429 if (exitall_on_terminate) {
430 if (eta_secs[i] < je->eta_sec)
431 je->eta_sec = eta_secs[i];
433 if (eta_secs[i] > je->eta_sec)
434 je->eta_sec = eta_secs[i];
440 fio_gettime(&now, NULL);
441 rate_time = mtime_since(&rate_prev_time, &now);
443 if (write_bw_log && rate_time > bw_avg_time && !in_ramp_time(td)) {
444 calc_rate(unified_rw_rep, rate_time, io_bytes, rate_io_bytes,
446 memcpy(&rate_prev_time, &now, sizeof(now));
447 add_agg_sample(je->rate[DDIR_READ], DDIR_READ, 0);
448 add_agg_sample(je->rate[DDIR_WRITE], DDIR_WRITE, 0);
449 add_agg_sample(je->rate[DDIR_TRIM], DDIR_TRIM, 0);
452 disp_time = mtime_since(&disp_prev_time, &now);
455 * Allow a little slack, the target is to print it every 1000 msecs
457 if (!force && disp_time < 900)
460 calc_rate(unified_rw_rep, disp_time, io_bytes, disp_io_bytes, je->rate);
461 calc_iops(unified_rw_rep, disp_time, io_iops, disp_io_iops, je->iops);
463 memcpy(&disp_prev_time, &now, sizeof(now));
465 if (!force && !je->nr_running && !je->nr_pending)
468 je->nr_threads = thread_number;
469 update_condensed_str(__run_str, run_str);
470 memcpy(je->run_str, run_str, strlen(run_str));
474 void display_thread_status(struct jobs_eta *je)
476 static struct timeval disp_eta_new_line;
477 static int eta_new_line_init, eta_new_line_pending;
478 static int linelen_last;
480 char output[REAL_MAX_JOBS + 512], *p = output;
484 if (je->eta_sec != INT_MAX && je->elapsed_sec) {
485 perc = (double) je->elapsed_sec / (double) (je->elapsed_sec + je->eta_sec);
486 eta_to_str(eta_str, je->eta_sec);
489 if (eta_new_line_pending) {
490 eta_new_line_pending = 0;
491 p += sprintf(p, "\n");
494 p += sprintf(p, "Jobs: %d (f=%d)", je->nr_running, je->files_open);
495 if (je->m_rate[0] || je->m_rate[1] || je->t_rate[0] || je->t_rate[1]) {
498 mr = num2str(je->m_rate[0] + je->m_rate[1], 4, 0, je->is_pow2, 8);
499 tr = num2str(je->t_rate[0] + je->t_rate[1], 4, 0, je->is_pow2, 8);
500 p += sprintf(p, ", CR=%s/%s KB/s", tr, mr);
503 } else if (je->m_iops[0] || je->m_iops[1] || je->t_iops[0] || je->t_iops[1]) {
504 p += sprintf(p, ", CR=%d/%d IOPS",
505 je->t_iops[0] + je->t_iops[1],
506 je->m_iops[0] + je->m_iops[1]);
508 if (je->eta_sec != INT_MAX && je->nr_running) {
510 char *iops_str[DDIR_RWDIR_CNT];
511 char *rate_str[DDIR_RWDIR_CNT];
516 if ((!je->eta_sec && !eta_good) || je->nr_ramp == je->nr_running)
517 strcpy(perc_str, "-.-% done");
521 if (je->nr_setting_up && je->nr_running)
522 mult *= (1.0 - (double) je->nr_setting_up / (double) je->nr_running);
526 sprintf(perc_str, "%3.1f%% done", perc);
529 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
530 rate_str[ddir] = num2str(je->rate[ddir], 5,
531 1024, je->is_pow2, je->unit_base);
532 iops_str[ddir] = num2str(je->iops[ddir], 4, 1, 0, 0);
535 left = sizeof(output) - (p - output) - 1;
537 l = snprintf(p, left, ": [%s] [%s] [%s/%s/%s /s] [%s/%s/%s iops] [eta %s]",
538 je->run_str, perc_str, rate_str[DDIR_READ],
539 rate_str[DDIR_WRITE], rate_str[DDIR_TRIM],
540 iops_str[DDIR_READ], iops_str[DDIR_WRITE],
541 iops_str[DDIR_TRIM], eta_str);
543 if (l >= 0 && l < linelen_last)
544 p += sprintf(p, "%*s", linelen_last - l, "");
547 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
548 free(rate_str[ddir]);
549 free(iops_str[ddir]);
552 p += sprintf(p, "\r");
554 printf("%s", output);
556 if (!eta_new_line_init) {
557 fio_gettime(&disp_eta_new_line, NULL);
558 eta_new_line_init = 1;
559 } else if (eta_new_line && mtime_since_now(&disp_eta_new_line) > eta_new_line) {
560 fio_gettime(&disp_eta_new_line, NULL);
561 eta_new_line_pending = 1;
567 struct jobs_eta *get_jobs_eta(int force, size_t *size)
574 *size = sizeof(*je) + THREAD_RUNSTR_SZ;
578 memset(je, 0, *size);
580 if (!calc_thread_status(je, force)) {
585 *size = sizeof(*je) + strlen((char *) je->run_str) + 1;
589 void print_thread_status(void)
594 je = get_jobs_eta(0, &size);
596 display_thread_status(je);
601 void print_status_init(int thr_number)
603 __run_str[thr_number] = 'P';
604 update_condensed_str(__run_str, run_str);