1 /*
2 * Status and ETA code
3 */
4 #include <unistd.h>
5 #include <fcntl.h>
6 #include <string.h>
7
8 #include "fio.h"
9 #include "lib/pow2.h"
10
11 static char __run_str[REAL_MAX_JOBS + 1];
12 static char run_str[__THREAD_RUNSTR_SZ(REAL_MAX_JOBS)];
13
update_condensed_str(char * rstr,char * run_str_condensed)14 static void update_condensed_str(char *rstr, char *run_str_condensed)
15 {
16 if (*rstr) {
17 while (*rstr) {
18 int nr = 1;
19
20 *run_str_condensed++ = *rstr++;
21 while (*(rstr - 1) == *rstr) {
22 rstr++;
23 nr++;
24 }
25 run_str_condensed += sprintf(run_str_condensed, "(%u),", nr);
26 }
27 run_str_condensed--;
28 }
29 *run_str_condensed = '\0';
30 }
31
32 /*
33 * Sets the status of the 'td' in the printed status map.
34 */
check_str_update(struct thread_data * td)35 static void check_str_update(struct thread_data *td)
36 {
37 char c = __run_str[td->thread_number - 1];
38
39 switch (td->runstate) {
40 case TD_REAPED:
41 if (td->error)
42 c = 'X';
43 else if (td->sig)
44 c = 'K';
45 else
46 c = '_';
47 break;
48 case TD_EXITED:
49 c = 'E';
50 break;
51 case TD_RAMP:
52 c = '/';
53 break;
54 case TD_RUNNING:
55 if (td_rw(td)) {
56 if (td_random(td)) {
57 if (td->o.rwmix[DDIR_READ] == 100)
58 c = 'r';
59 else if (td->o.rwmix[DDIR_WRITE] == 100)
60 c = 'w';
61 else
62 c = 'm';
63 } else {
64 if (td->o.rwmix[DDIR_READ] == 100)
65 c = 'R';
66 else if (td->o.rwmix[DDIR_WRITE] == 100)
67 c = 'W';
68 else
69 c = 'M';
70 }
71 } else if (td_read(td)) {
72 if (td_random(td))
73 c = 'r';
74 else
75 c = 'R';
76 } else if (td_write(td)) {
77 if (td_random(td))
78 c = 'w';
79 else
80 c = 'W';
81 } else {
82 if (td_random(td))
83 c = 'd';
84 else
85 c = 'D';
86 }
87 break;
88 case TD_PRE_READING:
89 c = 'p';
90 break;
91 case TD_VERIFYING:
92 c = 'V';
93 break;
94 case TD_FSYNCING:
95 c = 'F';
96 break;
97 case TD_FINISHING:
98 c = 'f';
99 break;
100 case TD_CREATED:
101 c = 'C';
102 break;
103 case TD_INITIALIZED:
104 case TD_SETTING_UP:
105 c = 'I';
106 break;
107 case TD_NOT_CREATED:
108 c = 'P';
109 break;
110 default:
111 log_err("state %d\n", td->runstate);
112 }
113
114 __run_str[td->thread_number - 1] = c;
115 update_condensed_str(__run_str, run_str);
116 }
117
118 /*
119 * Convert seconds to a printable string.
120 */
eta_to_str(char * str,unsigned long eta_sec)121 void eta_to_str(char *str, unsigned long eta_sec)
122 {
123 unsigned int d, h, m, s;
124 int disp_hour = 0;
125
126 if (eta_sec == -1) {
127 sprintf(str, "--");
128 return;
129 }
130
131 s = eta_sec % 60;
132 eta_sec /= 60;
133 m = eta_sec % 60;
134 eta_sec /= 60;
135 h = eta_sec % 24;
136 eta_sec /= 24;
137 d = eta_sec;
138
139 if (d) {
140 disp_hour = 1;
141 str += sprintf(str, "%02ud:", d);
142 }
143
144 if (h || disp_hour)
145 str += sprintf(str, "%02uh:", h);
146
147 str += sprintf(str, "%02um:", m);
148 str += sprintf(str, "%02us", s);
149 }
150
151 /*
152 * Best effort calculation of the estimated pending runtime of a job.
153 */
thread_eta(struct thread_data * td)154 static unsigned long thread_eta(struct thread_data *td)
155 {
156 unsigned long long bytes_total, bytes_done;
157 unsigned long eta_sec = 0;
158 unsigned long elapsed;
159 uint64_t timeout;
160
161 elapsed = (mtime_since_now(&td->epoch) + 999) / 1000;
162 timeout = td->o.timeout / 1000000UL;
163
164 bytes_total = td->total_io_size;
165
166 if (td->flags & TD_F_NO_PROGRESS)
167 return -1;
168
169 if (td->o.fill_device && td->o.size == -1ULL) {
170 if (!td->fill_device_size || td->fill_device_size == -1ULL)
171 return 0;
172
173 bytes_total = td->fill_device_size;
174 }
175
176 if (td->o.zone_size && td->o.zone_skip && bytes_total) {
177 unsigned int nr_zones;
178 uint64_t zone_bytes;
179
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;
183 }
184
185 /*
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.
189 */
190 if (td->o.do_verify && td->o.verify && td_write(td)) {
191 if (td_rw(td)) {
192 unsigned int perc = 50;
193
194 if (td->o.rwmix[DDIR_WRITE])
195 perc = td->o.rwmix[DDIR_WRITE];
196
197 bytes_total += (bytes_total * perc) / 100;
198 } else
199 bytes_total <<= 1;
200 }
201
202 if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING) {
203 double perc, perc_t;
204
205 bytes_done = ddir_rw_sum(td->io_bytes);
206
207 if (bytes_total) {
208 perc = (double) bytes_done / (double) bytes_total;
209 if (perc > 1.0)
210 perc = 1.0;
211 } else
212 perc = 0.0;
213
214 if (td->o.time_based) {
215 if (timeout) {
216 perc_t = (double) elapsed / (double) timeout;
217 if (perc_t < perc)
218 perc = perc_t;
219 } else {
220 /*
221 * Will never hit, we can't have time_based
222 * without a timeout set.
223 */
224 perc = 0.0;
225 }
226 }
227
228 if (perc == 0.0) {
229 eta_sec = timeout;
230 } else {
231 eta_sec = (unsigned long) (elapsed * (1.0 / perc)) - elapsed;
232 }
233
234 if (td->o.timeout &&
235 eta_sec > (timeout + done_secs - elapsed))
236 eta_sec = timeout + done_secs - elapsed;
237 } else if (td->runstate == TD_NOT_CREATED || td->runstate == TD_CREATED
238 || td->runstate == TD_INITIALIZED
239 || td->runstate == TD_SETTING_UP
240 || td->runstate == TD_RAMP
241 || td->runstate == TD_PRE_READING) {
242 int64_t t_eta = 0, r_eta = 0;
243 unsigned long long rate_bytes;
244
245 /*
246 * We can only guess - assume it'll run the full timeout
247 * if given, otherwise assume it'll run at the specified rate.
248 */
249 if (td->o.timeout) {
250 uint64_t __timeout = td->o.timeout;
251 uint64_t start_delay = td->o.start_delay;
252 uint64_t ramp_time = td->o.ramp_time;
253
254 t_eta = __timeout + start_delay;
255 if (!td->ramp_time_over) {
256 t_eta += ramp_time;
257 }
258 t_eta /= 1000000ULL;
259
260 if ((td->runstate == TD_RAMP) && in_ramp_time(td)) {
261 unsigned long ramp_left;
262
263 ramp_left = mtime_since_now(&td->epoch);
264 ramp_left = (ramp_left + 999) / 1000;
265 if (ramp_left <= t_eta)
266 t_eta -= ramp_left;
267 }
268 }
269 rate_bytes = 0;
270 if (td_read(td))
271 rate_bytes = td->o.rate[DDIR_READ];
272 if (td_write(td))
273 rate_bytes += td->o.rate[DDIR_WRITE];
274 if (td_trim(td))
275 rate_bytes += td->o.rate[DDIR_TRIM];
276
277 if (rate_bytes) {
278 r_eta = bytes_total / rate_bytes;
279 r_eta += (td->o.start_delay / 1000000ULL);
280 }
281
282 if (r_eta && t_eta)
283 eta_sec = min(r_eta, t_eta);
284 else if (r_eta)
285 eta_sec = r_eta;
286 else if (t_eta)
287 eta_sec = t_eta;
288 else
289 eta_sec = 0;
290 } else {
291 /*
292 * thread is already done or waiting for fsync
293 */
294 eta_sec = 0;
295 }
296
297 return eta_sec;
298 }
299
calc_rate(int unified_rw_rep,unsigned long mtime,unsigned long long * io_bytes,unsigned long long * prev_io_bytes,uint64_t * rate)300 static void calc_rate(int unified_rw_rep, unsigned long mtime,
301 unsigned long long *io_bytes,
302 unsigned long long *prev_io_bytes, uint64_t *rate)
303 {
304 int i;
305
306 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
307 unsigned long long diff, this_rate;
308
309 diff = io_bytes[i] - prev_io_bytes[i];
310 if (mtime)
311 this_rate = ((1000 * diff) / mtime) / 1024; /* KiB/s */
312 else
313 this_rate = 0;
314
315 if (unified_rw_rep) {
316 rate[i] = 0;
317 rate[0] += this_rate;
318 } else
319 rate[i] = this_rate;
320
321 prev_io_bytes[i] = io_bytes[i];
322 }
323 }
324
calc_iops(int unified_rw_rep,unsigned long mtime,unsigned long long * io_iops,unsigned long long * prev_io_iops,unsigned int * iops)325 static void calc_iops(int unified_rw_rep, unsigned long mtime,
326 unsigned long long *io_iops,
327 unsigned long long *prev_io_iops, unsigned int *iops)
328 {
329 int i;
330
331 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
332 unsigned long long diff, this_iops;
333
334 diff = io_iops[i] - prev_io_iops[i];
335 if (mtime)
336 this_iops = (diff * 1000) / mtime;
337 else
338 this_iops = 0;
339
340 if (unified_rw_rep) {
341 iops[i] = 0;
342 iops[0] += this_iops;
343 } else
344 iops[i] = this_iops;
345
346 prev_io_iops[i] = io_iops[i];
347 }
348 }
349
350 /*
351 * Print status of the jobs we know about. This includes rate estimates,
352 * ETA, thread state, etc.
353 */
calc_thread_status(struct jobs_eta * je,int force)354 bool calc_thread_status(struct jobs_eta *je, int force)
355 {
356 struct thread_data *td;
357 int i, unified_rw_rep;
358 uint64_t rate_time, disp_time, bw_avg_time, *eta_secs;
359 unsigned long long io_bytes[DDIR_RWDIR_CNT];
360 unsigned long long io_iops[DDIR_RWDIR_CNT];
361 struct timeval now;
362
363 static unsigned long long rate_io_bytes[DDIR_RWDIR_CNT];
364 static unsigned long long disp_io_bytes[DDIR_RWDIR_CNT];
365 static unsigned long long disp_io_iops[DDIR_RWDIR_CNT];
366 static struct timeval rate_prev_time, disp_prev_time;
367
368 if (!force) {
369 if (!(output_format & FIO_OUTPUT_NORMAL) &&
370 f_out == stdout)
371 return false;
372 if (temp_stall_ts || eta_print == FIO_ETA_NEVER)
373 return false;
374
375 if (!isatty(STDOUT_FILENO) && (eta_print != FIO_ETA_ALWAYS))
376 return false;
377 }
378
379 if (!ddir_rw_sum(rate_io_bytes))
380 fill_start_time(&rate_prev_time);
381 if (!ddir_rw_sum(disp_io_bytes))
382 fill_start_time(&disp_prev_time);
383
384 eta_secs = malloc(thread_number * sizeof(uint64_t));
385 memset(eta_secs, 0, thread_number * sizeof(uint64_t));
386
387 je->elapsed_sec = (mtime_since_genesis() + 999) / 1000;
388
389 io_bytes[DDIR_READ] = io_bytes[DDIR_WRITE] = io_bytes[DDIR_TRIM] = 0;
390 io_iops[DDIR_READ] = io_iops[DDIR_WRITE] = io_iops[DDIR_TRIM] = 0;
391 bw_avg_time = ULONG_MAX;
392 unified_rw_rep = 0;
393 for_each_td(td, i) {
394 unified_rw_rep += td->o.unified_rw_rep;
395 if (is_power_of_2(td->o.kb_base))
396 je->is_pow2 = 1;
397 je->unit_base = td->o.unit_base;
398 if (td->o.bw_avg_time < bw_avg_time)
399 bw_avg_time = td->o.bw_avg_time;
400 if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING
401 || td->runstate == TD_FSYNCING
402 || td->runstate == TD_PRE_READING
403 || td->runstate == TD_FINISHING) {
404 je->nr_running++;
405 if (td_read(td)) {
406 je->t_rate[0] += td->o.rate[DDIR_READ];
407 je->t_iops[0] += td->o.rate_iops[DDIR_READ];
408 je->m_rate[0] += td->o.ratemin[DDIR_READ];
409 je->m_iops[0] += td->o.rate_iops_min[DDIR_READ];
410 }
411 if (td_write(td)) {
412 je->t_rate[1] += td->o.rate[DDIR_WRITE];
413 je->t_iops[1] += td->o.rate_iops[DDIR_WRITE];
414 je->m_rate[1] += td->o.ratemin[DDIR_WRITE];
415 je->m_iops[1] += td->o.rate_iops_min[DDIR_WRITE];
416 }
417 if (td_trim(td)) {
418 je->t_rate[2] += td->o.rate[DDIR_TRIM];
419 je->t_iops[2] += td->o.rate_iops[DDIR_TRIM];
420 je->m_rate[2] += td->o.ratemin[DDIR_TRIM];
421 je->m_iops[2] += td->o.rate_iops_min[DDIR_TRIM];
422 }
423
424 je->files_open += td->nr_open_files;
425 } else if (td->runstate == TD_RAMP) {
426 je->nr_running++;
427 je->nr_ramp++;
428 } else if (td->runstate == TD_SETTING_UP)
429 je->nr_setting_up++;
430 else if (td->runstate < TD_RUNNING)
431 je->nr_pending++;
432
433 if (je->elapsed_sec >= 3)
434 eta_secs[i] = thread_eta(td);
435 else
436 eta_secs[i] = INT_MAX;
437
438 check_str_update(td);
439
440 if (td->runstate > TD_SETTING_UP) {
441 int ddir;
442
443 for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) {
444 if (unified_rw_rep) {
445 io_bytes[0] += td->io_bytes[ddir];
446 io_iops[0] += td->io_blocks[ddir];
447 } else {
448 io_bytes[ddir] += td->io_bytes[ddir];
449 io_iops[ddir] += td->io_blocks[ddir];
450 }
451 }
452 }
453 }
454
455 if (exitall_on_terminate) {
456 je->eta_sec = INT_MAX;
457 for_each_td(td, i) {
458 if (eta_secs[i] < je->eta_sec)
459 je->eta_sec = eta_secs[i];
460 }
461 } else {
462 unsigned long eta_stone = 0;
463
464 je->eta_sec = 0;
465 for_each_td(td, i) {
466 if ((td->runstate == TD_NOT_CREATED) && td->o.stonewall)
467 eta_stone += eta_secs[i];
468 else {
469 if (eta_secs[i] > je->eta_sec)
470 je->eta_sec = eta_secs[i];
471 }
472 }
473 je->eta_sec += eta_stone;
474 }
475
476 free(eta_secs);
477
478 fio_gettime(&now, NULL);
479 rate_time = mtime_since(&rate_prev_time, &now);
480
481 if (write_bw_log && rate_time > bw_avg_time && !in_ramp_time(td)) {
482 calc_rate(unified_rw_rep, rate_time, io_bytes, rate_io_bytes,
483 je->rate);
484 memcpy(&rate_prev_time, &now, sizeof(now));
485 add_agg_sample(sample_val(je->rate[DDIR_READ]), DDIR_READ, 0);
486 add_agg_sample(sample_val(je->rate[DDIR_WRITE]), DDIR_WRITE, 0);
487 add_agg_sample(sample_val(je->rate[DDIR_TRIM]), DDIR_TRIM, 0);
488 }
489
490 disp_time = mtime_since(&disp_prev_time, &now);
491
492 /*
493 * Allow a little slack, the target is to print it every 1000 msecs
494 */
495 if (!force && disp_time < 900)
496 return false;
497
498 calc_rate(unified_rw_rep, disp_time, io_bytes, disp_io_bytes, je->rate);
499 calc_iops(unified_rw_rep, disp_time, io_iops, disp_io_iops, je->iops);
500
501 memcpy(&disp_prev_time, &now, sizeof(now));
502
503 if (!force && !je->nr_running && !je->nr_pending)
504 return false;
505
506 je->nr_threads = thread_number;
507 update_condensed_str(__run_str, run_str);
508 memcpy(je->run_str, run_str, strlen(run_str));
509 return true;
510 }
511
display_thread_status(struct jobs_eta * je)512 void display_thread_status(struct jobs_eta *je)
513 {
514 static struct timeval disp_eta_new_line;
515 static int eta_new_line_init, eta_new_line_pending;
516 static int linelen_last;
517 static int eta_good;
518 char output[REAL_MAX_JOBS + 512], *p = output;
519 char eta_str[128];
520 double perc = 0.0;
521
522 if (je->eta_sec != INT_MAX && je->elapsed_sec) {
523 perc = (double) je->elapsed_sec / (double) (je->elapsed_sec + je->eta_sec);
524 eta_to_str(eta_str, je->eta_sec);
525 }
526
527 if (eta_new_line_pending) {
528 eta_new_line_pending = 0;
529 p += sprintf(p, "\n");
530 }
531
532 p += sprintf(p, "Jobs: %d (f=%d)", je->nr_running, je->files_open);
533
534 /* rate limits, if any */
535 if (je->m_rate[0] || je->m_rate[1] || je->m_rate[2] ||
536 je->t_rate[0] || je->t_rate[1] || je->t_rate[2]) {
537 char *tr, *mr;
538
539 mr = num2str(je->m_rate[0] + je->m_rate[1] + je->m_rate[2],
540 4, 0, je->is_pow2, N2S_BYTEPERSEC);
541 tr = num2str(je->t_rate[0] + je->t_rate[1] + je->t_rate[2],
542 4, 0, je->is_pow2, N2S_BYTEPERSEC);
543
544 p += sprintf(p, ", %s-%s", mr, tr);
545 free(tr);
546 free(mr);
547 } else if (je->m_iops[0] || je->m_iops[1] || je->m_iops[2] ||
548 je->t_iops[0] || je->t_iops[1] || je->t_iops[2]) {
549 p += sprintf(p, ", %d-%d IOPS",
550 je->m_iops[0] + je->m_iops[1] + je->m_iops[2],
551 je->t_iops[0] + je->t_iops[1] + je->t_iops[2]);
552 }
553
554 /* current run string, % done, bandwidth, iops, eta */
555 if (je->eta_sec != INT_MAX && je->nr_running) {
556 char perc_str[32];
557 char *iops_str[DDIR_RWDIR_CNT];
558 char *rate_str[DDIR_RWDIR_CNT];
559 size_t left;
560 int l;
561 int ddir;
562
563 if ((!je->eta_sec && !eta_good) || je->nr_ramp == je->nr_running ||
564 je->eta_sec == -1)
565 strcpy(perc_str, "-.-%");
566 else {
567 double mult = 100.0;
568
569 if (je->nr_setting_up && je->nr_running)
570 mult *= (1.0 - (double) je->nr_setting_up / (double) je->nr_running);
571
572 eta_good = 1;
573 perc *= mult;
574 sprintf(perc_str, "%3.1f%%", perc);
575 }
576
577 for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) {
578 rate_str[ddir] = num2str(je->rate[ddir], 4,
579 1024, je->is_pow2, je->unit_base);
580 iops_str[ddir] = num2str(je->iops[ddir], 4, 1, 0, N2S_NONE);
581 }
582
583 left = sizeof(output) - (p - output) - 1;
584
585 if (je->rate[DDIR_TRIM] || je->iops[DDIR_TRIM])
586 l = snprintf(p, left,
587 ": [%s][%s][r=%s,w=%s,t=%s][r=%s,w=%s,t=%s IOPS][eta %s]",
588 je->run_str, perc_str, rate_str[DDIR_READ],
589 rate_str[DDIR_WRITE], rate_str[DDIR_TRIM],
590 iops_str[DDIR_READ], iops_str[DDIR_WRITE],
591 iops_str[DDIR_TRIM], eta_str);
592 else
593 l = snprintf(p, left,
594 ": [%s][%s][r=%s,w=%s][r=%s,w=%s IOPS][eta %s]",
595 je->run_str, perc_str,
596 rate_str[DDIR_READ], rate_str[DDIR_WRITE],
597 iops_str[DDIR_READ], iops_str[DDIR_WRITE],
598 eta_str);
599 p += l;
600 if (l >= 0 && l < linelen_last)
601 p += sprintf(p, "%*s", linelen_last - l, "");
602 linelen_last = l;
603
604 for (ddir = 0; ddir < DDIR_RWDIR_CNT; ddir++) {
605 free(rate_str[ddir]);
606 free(iops_str[ddir]);
607 }
608 }
609 p += sprintf(p, "\r");
610
611 printf("%s", output);
612
613 if (!eta_new_line_init) {
614 fio_gettime(&disp_eta_new_line, NULL);
615 eta_new_line_init = 1;
616 } else if (eta_new_line && mtime_since_now(&disp_eta_new_line) > eta_new_line) {
617 fio_gettime(&disp_eta_new_line, NULL);
618 eta_new_line_pending = 1;
619 }
620
621 fflush(stdout);
622 }
623
get_jobs_eta(bool force,size_t * size)624 struct jobs_eta *get_jobs_eta(bool force, size_t *size)
625 {
626 struct jobs_eta *je;
627
628 if (!thread_number)
629 return NULL;
630
631 *size = sizeof(*je) + THREAD_RUNSTR_SZ + 8;
632 je = malloc(*size);
633 if (!je)
634 return NULL;
635 memset(je, 0, *size);
636
637 if (!calc_thread_status(je, force)) {
638 free(je);
639 return NULL;
640 }
641
642 *size = sizeof(*je) + strlen((char *) je->run_str) + 1;
643 return je;
644 }
645
print_thread_status(void)646 void print_thread_status(void)
647 {
648 struct jobs_eta *je;
649 size_t size;
650
651 je = get_jobs_eta(false, &size);
652 if (je)
653 display_thread_status(je);
654
655 free(je);
656 }
657
print_status_init(int thr_number)658 void print_status_init(int thr_number)
659 {
660 __run_str[thr_number] = 'P';
661 update_condensed_str(__run_str, run_str);
662 }
663