eta: fix potential divide by zero
[fio.git] / eta.c
CommitLineData
263e529f
JA
1/*
2 * Status and ETA code
3 */
4#include <unistd.h>
5#include <fcntl.h>
6#include <string.h>
7
8#include "fio.h"
263e529f 9
fca70358 10static char run_str[REAL_MAX_JOBS + 1];
263e529f
JA
11
12/*
13 * Sets the status of the 'td' in the printed status map.
14 */
15static void check_str_update(struct thread_data *td)
16{
17 char c = run_str[td->thread_number - 1];
18
19 switch (td->runstate) {
5ec10eaa 20 case TD_REAPED:
4f7e57a4
JA
21 if (td->error)
22 c = 'X';
a5e371a6
JA
23 else if (td->sig)
24 c = 'K';
4f7e57a4
JA
25 else
26 c = '_';
5ec10eaa
JA
27 break;
28 case TD_EXITED:
29 c = 'E';
30 break;
b29ee5b3
JA
31 case TD_RAMP:
32 c = '/';
33 break;
5ec10eaa
JA
34 case TD_RUNNING:
35 if (td_rw(td)) {
e3b3f81e
JA
36 if (td_random(td)) {
37 if (td->o.rwmix[DDIR_READ] == 100)
38 c = 'r';
39 else if (td->o.rwmix[DDIR_WRITE] == 100)
40 c = 'w';
41 else
42 c = 'm';
43 } else {
44 if (td->o.rwmix[DDIR_READ] == 100)
45 c = 'R';
46 else if (td->o.rwmix[DDIR_WRITE] == 100)
47 c = 'W';
48 else
49 c = 'M';
50 }
5ec10eaa
JA
51 } else if (td_read(td)) {
52 if (td_random(td))
53 c = 'r';
54 else
55 c = 'R';
6eaf09d6 56 } else if (td_write(td)) {
5ec10eaa
JA
57 if (td_random(td))
58 c = 'w';
59 else
60 c = 'W';
6eaf09d6
SL
61 } else {
62 if (td_random(td))
63 c = 'd';
64 else
65 c = 'D';
5ec10eaa
JA
66 }
67 break;
b0f65863
JA
68 case TD_PRE_READING:
69 c = 'p';
70 break;
5ec10eaa
JA
71 case TD_VERIFYING:
72 c = 'V';
73 break;
74 case TD_FSYNCING:
75 c = 'F';
76 break;
3d434057
JA
77 case TD_FINISHING:
78 c = 'f';
79 break;
5ec10eaa
JA
80 case TD_CREATED:
81 c = 'C';
82 break;
83 case TD_INITIALIZED:
9c6f6316 84 case TD_SETTING_UP:
5ec10eaa
JA
85 c = 'I';
86 break;
87 case TD_NOT_CREATED:
88 c = 'P';
89 break;
90 default:
91 log_err("state %d\n", td->runstate);
263e529f
JA
92 }
93
94 run_str[td->thread_number - 1] = c;
95}
96
97/*
98 * Convert seconds to a printable string.
99 */
3e47bd25 100void eta_to_str(char *str, unsigned long eta_sec)
263e529f
JA
101{
102 unsigned int d, h, m, s;
165faf16 103 int disp_hour = 0;
263e529f 104
263e529f
JA
105 s = eta_sec % 60;
106 eta_sec /= 60;
107 m = eta_sec % 60;
108 eta_sec /= 60;
109 h = eta_sec % 24;
110 eta_sec /= 24;
111 d = eta_sec;
112
165faf16
JA
113 if (d) {
114 disp_hour = 1;
1e97cce9 115 str += sprintf(str, "%02ud:", d);
263e529f 116 }
165faf16
JA
117
118 if (h || disp_hour)
1e97cce9 119 str += sprintf(str, "%02uh:", h);
263e529f 120
1e97cce9
JA
121 str += sprintf(str, "%02um:", m);
122 str += sprintf(str, "%02us", s);
263e529f
JA
123}
124
125/*
126 * Best effort calculation of the estimated pending runtime of a job.
127 */
b29ee5b3 128static int thread_eta(struct thread_data *td)
263e529f
JA
129{
130 unsigned long long bytes_total, bytes_done;
1e97cce9 131 unsigned long eta_sec = 0;
b29ee5b3 132 unsigned long elapsed;
0de5b26f 133 uint64_t timeout;
b29ee5b3
JA
134
135 elapsed = (mtime_since_now(&td->epoch) + 999) / 1000;
0de5b26f 136 timeout = td->o.timeout / 1000000UL;
263e529f
JA
137
138 bytes_total = td->total_io_size;
139
2e3bd4c2
JA
140 if (td->o.fill_device && td->o.size == -1ULL) {
141 if (!td->fill_device_size || td->fill_device_size == -1ULL)
142 return 0;
143
144 bytes_total = td->fill_device_size;
145 }
146
0d73a2f9
JA
147 if (td->o.zone_size && td->o.zone_skip && bytes_total) {
148 unsigned int nr_zones;
149 uint64_t zone_bytes;
150
151 zone_bytes = bytes_total + td->o.zone_size + td->o.zone_skip;
152 nr_zones = (zone_bytes - 1) / (td->o.zone_size + td->o.zone_skip);
153 bytes_total -= nr_zones * td->o.zone_skip;
154 }
155
74939e38 156 /*
165eb050
JA
157 * if writing and verifying afterwards, bytes_total will be twice the
158 * size. In a mixed workload, verify phase will be the size of the
159 * first stage writes.
74939e38 160 */
0dd8377f 161 if (td->o.do_verify && td->o.verify && td_write(td)) {
165eb050
JA
162 if (td_rw(td)) {
163 unsigned int perc = 50;
164
165 if (td->o.rwmix[DDIR_WRITE])
166 perc = td->o.rwmix[DDIR_WRITE];
167
168 bytes_total += (bytes_total * perc) / 100;
169 } else
74939e38
JA
170 bytes_total <<= 1;
171 }
172
263e529f 173 if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING) {
cf4464ca 174 double perc, perc_t;
263e529f 175
342f4be4 176 bytes_done = ddir_rw_sum(td->io_bytes);
476882d7
JA
177
178 if (bytes_total) {
179 perc = (double) bytes_done / (double) bytes_total;
180 if (perc > 1.0)
181 perc = 1.0;
182 } else
183 perc = 0.0;
263e529f 184
cf4464ca 185 if (td->o.time_based) {
476882d7
JA
186 if (timeout) {
187 perc_t = (double) elapsed / (double) timeout;
188 if (perc_t < perc)
189 perc = perc_t;
190 } else {
191 /*
192 * Will never hit, we can't have time_based
193 * without a timeout set.
194 */
195 perc = 0.0;
196 }
cf4464ca
JA
197 }
198
1e97cce9 199 eta_sec = (unsigned long) (elapsed * (1.0 / perc)) - elapsed;
263e529f 200
d3eeeabc 201 if (td->o.timeout &&
0de5b26f
JA
202 eta_sec > (timeout + done_secs - elapsed))
203 eta_sec = timeout + done_secs - elapsed;
263e529f 204 } else if (td->runstate == TD_NOT_CREATED || td->runstate == TD_CREATED
b29ee5b3 205 || td->runstate == TD_INITIALIZED
714e85f3 206 || td->runstate == TD_SETTING_UP
b0f65863
JA
207 || td->runstate == TD_RAMP
208 || td->runstate == TD_PRE_READING) {
263e529f 209 int t_eta = 0, r_eta = 0;
342f4be4 210 unsigned long long rate_bytes;
263e529f
JA
211
212 /*
213 * We can only guess - assume it'll run the full timeout
214 * if given, otherwise assume it'll run at the specified rate.
215 */
b29ee5b3 216 if (td->o.timeout) {
0de5b26f
JA
217 uint64_t timeout = td->o.timeout;
218 uint64_t start_delay = td->o.start_delay;
219 uint64_t ramp_time = td->o.ramp_time;
220
221 t_eta = timeout + start_delay + ramp_time;
222 t_eta /= 1000000ULL;
b29ee5b3
JA
223
224 if (in_ramp_time(td)) {
225 unsigned long ramp_left;
226
cda99fa0 227 ramp_left = mtime_since_now(&td->epoch);
b29ee5b3
JA
228 ramp_left = (ramp_left + 999) / 1000;
229 if (ramp_left <= t_eta)
230 t_eta -= ramp_left;
231 }
232 }
342f4be4
JA
233 rate_bytes = ddir_rw_sum(td->o.rate);
234 if (rate_bytes) {
235 r_eta = (bytes_total / 1024) / rate_bytes;
0de5b26f 236 r_eta += (td->o.start_delay / 1000000ULL);
263e529f
JA
237 }
238
239 if (r_eta && t_eta)
240 eta_sec = min(r_eta, t_eta);
241 else if (r_eta)
242 eta_sec = r_eta;
243 else if (t_eta)
244 eta_sec = t_eta;
245 else
246 eta_sec = 0;
247 } else {
248 /*
249 * thread is already done or waiting for fsync
250 */
251 eta_sec = 0;
252 }
253
254 return eta_sec;
255}
256
771e58be
JA
257static void calc_rate(int unified_rw_rep, unsigned long mtime,
258 unsigned long long *io_bytes,
46fda8d0
JA
259 unsigned long long *prev_io_bytes, unsigned int *rate)
260{
b7f05eb0
JA
261 int i;
262
6eaf09d6 263 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
b7f05eb0
JA
264 unsigned long long diff;
265
266 diff = io_bytes[i] - prev_io_bytes[i];
771e58be
JA
267 if (unified_rw_rep) {
268 rate[i] = 0;
269 rate[0] += ((1000 * diff) / mtime) / 1024;
270 } else
271 rate[i] = ((1000 * diff) / mtime) / 1024;
6eaf09d6
SL
272
273 prev_io_bytes[i] = io_bytes[i];
b7f05eb0 274 }
46fda8d0 275}
5ec10eaa 276
771e58be
JA
277static void calc_iops(int unified_rw_rep, unsigned long mtime,
278 unsigned long long *io_iops,
adb02ba8
JA
279 unsigned long long *prev_io_iops, unsigned int *iops)
280{
6eaf09d6
SL
281 int i;
282
283 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
771e58be
JA
284 unsigned long long diff;
285
286 diff = io_iops[i] - prev_io_iops[i];
287 if (unified_rw_rep) {
288 iops[i] = 0;
289 iops[0] += (diff * 1000) / mtime;
290 } else
291 iops[i] = (diff * 1000) / mtime;
292
6eaf09d6
SL
293 prev_io_iops[i] = io_iops[i];
294 }
adb02ba8
JA
295}
296
263e529f
JA
297/*
298 * Print status of the jobs we know about. This includes rate estimates,
299 * ETA, thread state, etc.
300 */
af9c9fb3 301int calc_thread_status(struct jobs_eta *je, int force)
263e529f 302{
34572e28 303 struct thread_data *td;
771e58be 304 int i, unified_rw_rep;
b75a394f 305 unsigned long rate_time, disp_time, bw_avg_time, *eta_secs;
6eaf09d6
SL
306 unsigned long long io_bytes[DDIR_RWDIR_CNT];
307 unsigned long long io_iops[DDIR_RWDIR_CNT];
46fda8d0
JA
308 struct timeval now;
309
6eaf09d6
SL
310 static unsigned long long rate_io_bytes[DDIR_RWDIR_CNT];
311 static unsigned long long disp_io_bytes[DDIR_RWDIR_CNT];
312 static unsigned long long disp_io_iops[DDIR_RWDIR_CNT];
46fda8d0 313 static struct timeval rate_prev_time, disp_prev_time;
6043c579 314
af9c9fb3 315 if (!force) {
fc47559b
JA
316 if (output_format != FIO_OUTPUT_NORMAL &&
317 f_out == stdout)
f3afa57e
JA
318 return 0;
319 if (temp_stall_ts || eta_print == FIO_ETA_NEVER)
af9c9fb3 320 return 0;
e592a06b 321
af9c9fb3
JA
322 if (!isatty(STDOUT_FILENO) && (eta_print != FIO_ETA_ALWAYS))
323 return 0;
324 }
263e529f 325
342f4be4 326 if (!ddir_rw_sum(rate_io_bytes))
46fda8d0 327 fill_start_time(&rate_prev_time);
342f4be4 328 if (!ddir_rw_sum(disp_io_bytes))
46fda8d0 329 fill_start_time(&disp_prev_time);
6043c579 330
d3eeeabc
JA
331 eta_secs = malloc(thread_number * sizeof(unsigned long));
332 memset(eta_secs, 0, thread_number * sizeof(unsigned long));
263e529f 333
b75a394f
JA
334 je->elapsed_sec = (mtime_since_genesis() + 999) / 1000;
335
6eaf09d6
SL
336 io_bytes[DDIR_READ] = io_bytes[DDIR_WRITE] = io_bytes[DDIR_TRIM] = 0;
337 io_iops[DDIR_READ] = io_iops[DDIR_WRITE] = io_iops[DDIR_TRIM] = 0;
bb3884d8 338 bw_avg_time = ULONG_MAX;
771e58be 339 unified_rw_rep = 0;
34572e28 340 for_each_td(td, i) {
771e58be 341 unified_rw_rep += td->o.unified_rw_rep;
b7f05eb0
JA
342 if (is_power_of_2(td->o.kb_base))
343 je->is_pow2 = 1;
ad705bcb 344 je->unit_base = td->o.unit_base;
2dc1bbeb
JA
345 if (td->o.bw_avg_time < bw_avg_time)
346 bw_avg_time = td->o.bw_avg_time;
46fda8d0 347 if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING
b0f65863 348 || td->runstate == TD_FSYNCING
3d434057
JA
349 || td->runstate == TD_PRE_READING
350 || td->runstate == TD_FINISHING) {
b75a394f 351 je->nr_running++;
242c38db 352 if (td_read(td)) {
c2e9cc4d
JA
353 je->t_rate[0] += td->o.rate[DDIR_READ];
354 je->t_iops[0] += td->o.rate_iops[DDIR_READ];
355 je->m_rate[0] += td->o.ratemin[DDIR_READ];
356 je->m_iops[0] += td->o.rate_iops_min[DDIR_READ];
242c38db
JA
357 }
358 if (td_write(td)) {
c2e9cc4d
JA
359 je->t_rate[1] += td->o.rate[DDIR_WRITE];
360 je->t_iops[1] += td->o.rate_iops[DDIR_WRITE];
361 je->m_rate[1] += td->o.ratemin[DDIR_WRITE];
362 je->m_iops[1] += td->o.rate_iops_min[DDIR_WRITE];
242c38db 363 }
6eaf09d6 364 if (td_trim(td)) {
d79db122
JA
365 je->t_rate[2] += td->o.rate[DDIR_TRIM];
366 je->t_iops[2] += td->o.rate_iops[DDIR_TRIM];
367 je->m_rate[2] += td->o.ratemin[DDIR_TRIM];
368 je->m_iops[2] += td->o.rate_iops_min[DDIR_TRIM];
6eaf09d6
SL
369 }
370
b75a394f 371 je->files_open += td->nr_open_files;
b29ee5b3 372 } else if (td->runstate == TD_RAMP) {
b75a394f
JA
373 je->nr_running++;
374 je->nr_ramp++;
714e85f3 375 } else if (td->runstate == TD_SETTING_UP) {
9c6f6316 376 je->nr_running++;
714e85f3
JA
377 je->nr_setting_up++;
378 } else if (td->runstate < TD_RUNNING)
b75a394f 379 je->nr_pending++;
263e529f 380
b75a394f 381 if (je->elapsed_sec >= 3)
b29ee5b3 382 eta_secs[i] = thread_eta(td);
263e529f
JA
383 else
384 eta_secs[i] = INT_MAX;
385
386 check_str_update(td);
b29ee5b3 387
714e85f3 388 if (td->runstate > TD_SETTING_UP) {
6eaf09d6 389 int ddir;
771e58be 390
6eaf09d6 391 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
771e58be
JA
392 if (unified_rw_rep) {
393 io_bytes[0] += td->io_bytes[ddir];
394 io_iops[0] += td->io_blocks[ddir];
395 } else {
396 io_bytes[ddir] += td->io_bytes[ddir];
397 io_iops[ddir] += td->io_blocks[ddir];
398 }
6eaf09d6 399 }
b29ee5b3 400 }
263e529f
JA
401 }
402
403 if (exitall_on_terminate)
b75a394f 404 je->eta_sec = INT_MAX;
263e529f 405 else
b75a394f 406 je->eta_sec = 0;
263e529f 407
34572e28 408 for_each_td(td, i) {
9c5a3854 409 if (exitall_on_terminate) {
b75a394f
JA
410 if (eta_secs[i] < je->eta_sec)
411 je->eta_sec = eta_secs[i];
9c5a3854 412 } else {
b75a394f
JA
413 if (eta_secs[i] > je->eta_sec)
414 je->eta_sec = eta_secs[i];
9c5a3854 415 }
263e529f
JA
416 }
417
eecf272f
JA
418 free(eta_secs);
419
46fda8d0
JA
420 fio_gettime(&now, NULL);
421 rate_time = mtime_since(&rate_prev_time, &now);
422
b29ee5b3 423 if (write_bw_log && rate_time > bw_avg_time && !in_ramp_time(td)) {
771e58be
JA
424 calc_rate(unified_rw_rep, rate_time, io_bytes, rate_io_bytes,
425 je->rate);
46fda8d0 426 memcpy(&rate_prev_time, &now, sizeof(now));
b75a394f
JA
427 add_agg_sample(je->rate[DDIR_READ], DDIR_READ, 0);
428 add_agg_sample(je->rate[DDIR_WRITE], DDIR_WRITE, 0);
6eaf09d6 429 add_agg_sample(je->rate[DDIR_TRIM], DDIR_TRIM, 0);
6043c579
JA
430 }
431
46fda8d0 432 disp_time = mtime_since(&disp_prev_time, &now);
a5b01f1b
JA
433
434 /*
435 * Allow a little slack, the target is to print it every 1000 msecs
436 */
af9c9fb3 437 if (!force && disp_time < 900)
b75a394f 438 return 0;
46fda8d0 439
771e58be
JA
440 calc_rate(unified_rw_rep, disp_time, io_bytes, disp_io_bytes, je->rate);
441 calc_iops(unified_rw_rep, disp_time, io_iops, disp_io_iops, je->iops);
adb02ba8 442
46fda8d0
JA
443 memcpy(&disp_prev_time, &now, sizeof(now));
444
af9c9fb3 445 if (!force && !je->nr_running && !je->nr_pending)
b75a394f
JA
446 return 0;
447
1d1f45ae
JA
448 je->nr_threads = thread_number;
449 memcpy(je->run_str, run_str, thread_number * sizeof(char));
b75a394f
JA
450 return 1;
451}
452
cf451d1e 453void display_thread_status(struct jobs_eta *je)
b75a394f 454{
e382e661
JA
455 static struct timeval disp_eta_new_line;
456 static int eta_new_line_init, eta_new_line_pending;
b75a394f
JA
457 static int linelen_last;
458 static int eta_good;
79074e7c 459 char output[REAL_MAX_JOBS + 512], *p = output;
b75a394f
JA
460 char eta_str[128];
461 double perc = 0.0;
b75a394f 462
cf451d1e
JA
463 if (je->eta_sec != INT_MAX && je->elapsed_sec) {
464 perc = (double) je->elapsed_sec / (double) (je->elapsed_sec + je->eta_sec);
465 eta_to_str(eta_str, je->eta_sec);
b75a394f
JA
466 }
467
e382e661
JA
468 if (eta_new_line_pending) {
469 eta_new_line_pending = 0;
470 p += sprintf(p, "\n");
471 }
472
cf451d1e 473 p += sprintf(p, "Jobs: %d (f=%d)", je->nr_running, je->files_open);
3e47bd25 474 if (je->m_rate[0] || je->m_rate[1] || je->t_rate[0] || je->t_rate[1]) {
581e7141
JA
475 char *tr, *mr;
476
22f80458
JA
477 mr = num2str(je->m_rate[0] + je->m_rate[1], 4, 0, je->is_pow2, 8);
478 tr = num2str(je->t_rate[0] + je->t_rate[1], 4, 0, je->is_pow2, 8);
37db14fe 479 p += sprintf(p, ", CR=%s/%s KB/s", tr, mr);
581e7141
JA
480 free(tr);
481 free(mr);
3e47bd25
JA
482 } else if (je->m_iops[0] || je->m_iops[1] || je->t_iops[0] || je->t_iops[1]) {
483 p += sprintf(p, ", CR=%d/%d IOPS",
484 je->t_iops[0] + je->t_iops[1],
88e4beaa 485 je->m_iops[0] + je->m_iops[1]);
3e47bd25 486 }
cf451d1e 487 if (je->eta_sec != INT_MAX && je->nr_running) {
0721d11e 488 char perc_str[32];
6eaf09d6
SL
489 char *iops_str[DDIR_RWDIR_CNT];
490 char *rate_str[DDIR_RWDIR_CNT];
79074e7c 491 size_t left;
adb02ba8 492 int l;
6eaf09d6 493 int ddir;
5ec10eaa 494
cf451d1e 495 if ((!je->eta_sec && !eta_good) || je->nr_ramp == je->nr_running)
cac58fd5
JA
496 strcpy(perc_str, "-.-% done");
497 else {
714e85f3
JA
498 double mult = 100.0;
499
500 if (je->nr_setting_up && je->nr_running)
501 mult *= (1.0 - (double) je->nr_setting_up / (double) je->nr_running);
502
cac58fd5 503 eta_good = 1;
714e85f3 504 perc *= mult;
0721d11e 505 sprintf(perc_str, "%3.1f%% done", perc);
cac58fd5 506 }
0721d11e 507
6eaf09d6
SL
508 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
509 rate_str[ddir] = num2str(je->rate[ddir], 5,
ad705bcb 510 1024, je->is_pow2, je->unit_base);
73798eb2 511 iops_str[ddir] = num2str(je->iops[ddir], 4, 1, 0, 0);
6eaf09d6 512 }
adb02ba8 513
79074e7c
JA
514 left = sizeof(output) - (p - output) - 1;
515
6eaf09d6
SL
516 l = snprintf(p, left, ": [%s] [%s] [%s/%s/%s /s] [%s/%s/%s iops] [eta %s]",
517 je->run_str, perc_str, rate_str[DDIR_READ],
518 rate_str[DDIR_WRITE], rate_str[DDIR_TRIM],
519 iops_str[DDIR_READ], iops_str[DDIR_WRITE],
520 iops_str[DDIR_TRIM], eta_str);
37db14fe 521 p += l;
adb02ba8 522 if (l >= 0 && l < linelen_last)
37db14fe 523 p += sprintf(p, "%*s", linelen_last - l, "");
adb02ba8 524 linelen_last = l;
d1bd7213 525
6eaf09d6
SL
526 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
527 free(rate_str[ddir]);
528 free(iops_str[ddir]);
529 }
263e529f 530 }
37db14fe
JA
531 p += sprintf(p, "\r");
532
3e47bd25 533 printf("%s", output);
e382e661
JA
534
535 if (!eta_new_line_init) {
536 fio_gettime(&disp_eta_new_line, NULL);
537 eta_new_line_init = 1;
538 } else if (eta_new_line &&
539 mtime_since_now(&disp_eta_new_line) > eta_new_line * 1000) {
540 fio_gettime(&disp_eta_new_line, NULL);
541 eta_new_line_pending = 1;
542 }
543
3e47bd25 544 fflush(stdout);
263e529f
JA
545}
546
cf451d1e
JA
547void print_thread_status(void)
548{
1d1f45ae 549 struct jobs_eta *je;
b814fb26 550 size_t size;
1d1f45ae 551
b814fb26
JA
552 if (!thread_number)
553 return;
cf451d1e 554
b814fb26
JA
555 size = sizeof(*je) + thread_number * sizeof(char) + 1;
556 je = malloc(size);
557 memset(je, 0, size);
cf451d1e 558
af9c9fb3 559 if (calc_thread_status(je, 0))
1d1f45ae 560 display_thread_status(je);
cf451d1e 561
1d1f45ae 562 free(je);
cf451d1e 563}
b75a394f 564
2b13e716 565void print_status_init(int thr_number)
263e529f 566{
2b13e716 567 run_str[thr_number] = 'P';
263e529f 568}