eta: fix potential divide by zero
[fio.git] / eta.c
... / ...
CommitLineData
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
10static char run_str[REAL_MAX_JOBS + 1];
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) {
20 case TD_REAPED:
21 if (td->error)
22 c = 'X';
23 else if (td->sig)
24 c = 'K';
25 else
26 c = '_';
27 break;
28 case TD_EXITED:
29 c = 'E';
30 break;
31 case TD_RAMP:
32 c = '/';
33 break;
34 case TD_RUNNING:
35 if (td_rw(td)) {
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 }
51 } else if (td_read(td)) {
52 if (td_random(td))
53 c = 'r';
54 else
55 c = 'R';
56 } else if (td_write(td)) {
57 if (td_random(td))
58 c = 'w';
59 else
60 c = 'W';
61 } else {
62 if (td_random(td))
63 c = 'd';
64 else
65 c = 'D';
66 }
67 break;
68 case TD_PRE_READING:
69 c = 'p';
70 break;
71 case TD_VERIFYING:
72 c = 'V';
73 break;
74 case TD_FSYNCING:
75 c = 'F';
76 break;
77 case TD_FINISHING:
78 c = 'f';
79 break;
80 case TD_CREATED:
81 c = 'C';
82 break;
83 case TD_INITIALIZED:
84 case TD_SETTING_UP:
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);
92 }
93
94 run_str[td->thread_number - 1] = c;
95}
96
97/*
98 * Convert seconds to a printable string.
99 */
100void eta_to_str(char *str, unsigned long eta_sec)
101{
102 unsigned int d, h, m, s;
103 int disp_hour = 0;
104
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
113 if (d) {
114 disp_hour = 1;
115 str += sprintf(str, "%02ud:", d);
116 }
117
118 if (h || disp_hour)
119 str += sprintf(str, "%02uh:", h);
120
121 str += sprintf(str, "%02um:", m);
122 str += sprintf(str, "%02us", s);
123}
124
125/*
126 * Best effort calculation of the estimated pending runtime of a job.
127 */
128static int thread_eta(struct thread_data *td)
129{
130 unsigned long long bytes_total, bytes_done;
131 unsigned long eta_sec = 0;
132 unsigned long elapsed;
133 uint64_t timeout;
134
135 elapsed = (mtime_since_now(&td->epoch) + 999) / 1000;
136 timeout = td->o.timeout / 1000000UL;
137
138 bytes_total = td->total_io_size;
139
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
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
156 /*
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.
160 */
161 if (td->o.do_verify && td->o.verify && td_write(td)) {
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
170 bytes_total <<= 1;
171 }
172
173 if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING) {
174 double perc, perc_t;
175
176 bytes_done = ddir_rw_sum(td->io_bytes);
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;
184
185 if (td->o.time_based) {
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 }
197 }
198
199 eta_sec = (unsigned long) (elapsed * (1.0 / perc)) - elapsed;
200
201 if (td->o.timeout &&
202 eta_sec > (timeout + done_secs - elapsed))
203 eta_sec = timeout + done_secs - elapsed;
204 } else if (td->runstate == TD_NOT_CREATED || td->runstate == TD_CREATED
205 || td->runstate == TD_INITIALIZED
206 || td->runstate == TD_SETTING_UP
207 || td->runstate == TD_RAMP
208 || td->runstate == TD_PRE_READING) {
209 int t_eta = 0, r_eta = 0;
210 unsigned long long rate_bytes;
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 */
216 if (td->o.timeout) {
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;
223
224 if (in_ramp_time(td)) {
225 unsigned long ramp_left;
226
227 ramp_left = mtime_since_now(&td->epoch);
228 ramp_left = (ramp_left + 999) / 1000;
229 if (ramp_left <= t_eta)
230 t_eta -= ramp_left;
231 }
232 }
233 rate_bytes = ddir_rw_sum(td->o.rate);
234 if (rate_bytes) {
235 r_eta = (bytes_total / 1024) / rate_bytes;
236 r_eta += (td->o.start_delay / 1000000ULL);
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
257static void calc_rate(int unified_rw_rep, unsigned long mtime,
258 unsigned long long *io_bytes,
259 unsigned long long *prev_io_bytes, unsigned int *rate)
260{
261 int i;
262
263 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
264 unsigned long long diff;
265
266 diff = io_bytes[i] - prev_io_bytes[i];
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;
272
273 prev_io_bytes[i] = io_bytes[i];
274 }
275}
276
277static void calc_iops(int unified_rw_rep, unsigned long mtime,
278 unsigned long long *io_iops,
279 unsigned long long *prev_io_iops, unsigned int *iops)
280{
281 int i;
282
283 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
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
293 prev_io_iops[i] = io_iops[i];
294 }
295}
296
297/*
298 * Print status of the jobs we know about. This includes rate estimates,
299 * ETA, thread state, etc.
300 */
301int calc_thread_status(struct jobs_eta *je, int force)
302{
303 struct thread_data *td;
304 int i, unified_rw_rep;
305 unsigned long rate_time, disp_time, bw_avg_time, *eta_secs;
306 unsigned long long io_bytes[DDIR_RWDIR_CNT];
307 unsigned long long io_iops[DDIR_RWDIR_CNT];
308 struct timeval now;
309
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];
313 static struct timeval rate_prev_time, disp_prev_time;
314
315 if (!force) {
316 if (output_format != FIO_OUTPUT_NORMAL &&
317 f_out == stdout)
318 return 0;
319 if (temp_stall_ts || eta_print == FIO_ETA_NEVER)
320 return 0;
321
322 if (!isatty(STDOUT_FILENO) && (eta_print != FIO_ETA_ALWAYS))
323 return 0;
324 }
325
326 if (!ddir_rw_sum(rate_io_bytes))
327 fill_start_time(&rate_prev_time);
328 if (!ddir_rw_sum(disp_io_bytes))
329 fill_start_time(&disp_prev_time);
330
331 eta_secs = malloc(thread_number * sizeof(unsigned long));
332 memset(eta_secs, 0, thread_number * sizeof(unsigned long));
333
334 je->elapsed_sec = (mtime_since_genesis() + 999) / 1000;
335
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;
338 bw_avg_time = ULONG_MAX;
339 unified_rw_rep = 0;
340 for_each_td(td, i) {
341 unified_rw_rep += td->o.unified_rw_rep;
342 if (is_power_of_2(td->o.kb_base))
343 je->is_pow2 = 1;
344 je->unit_base = td->o.unit_base;
345 if (td->o.bw_avg_time < bw_avg_time)
346 bw_avg_time = td->o.bw_avg_time;
347 if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING
348 || td->runstate == TD_FSYNCING
349 || td->runstate == TD_PRE_READING
350 || td->runstate == TD_FINISHING) {
351 je->nr_running++;
352 if (td_read(td)) {
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];
357 }
358 if (td_write(td)) {
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];
363 }
364 if (td_trim(td)) {
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];
369 }
370
371 je->files_open += td->nr_open_files;
372 } else if (td->runstate == TD_RAMP) {
373 je->nr_running++;
374 je->nr_ramp++;
375 } else if (td->runstate == TD_SETTING_UP) {
376 je->nr_running++;
377 je->nr_setting_up++;
378 } else if (td->runstate < TD_RUNNING)
379 je->nr_pending++;
380
381 if (je->elapsed_sec >= 3)
382 eta_secs[i] = thread_eta(td);
383 else
384 eta_secs[i] = INT_MAX;
385
386 check_str_update(td);
387
388 if (td->runstate > TD_SETTING_UP) {
389 int ddir;
390
391 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
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 }
399 }
400 }
401 }
402
403 if (exitall_on_terminate)
404 je->eta_sec = INT_MAX;
405 else
406 je->eta_sec = 0;
407
408 for_each_td(td, i) {
409 if (exitall_on_terminate) {
410 if (eta_secs[i] < je->eta_sec)
411 je->eta_sec = eta_secs[i];
412 } else {
413 if (eta_secs[i] > je->eta_sec)
414 je->eta_sec = eta_secs[i];
415 }
416 }
417
418 free(eta_secs);
419
420 fio_gettime(&now, NULL);
421 rate_time = mtime_since(&rate_prev_time, &now);
422
423 if (write_bw_log && rate_time > bw_avg_time && !in_ramp_time(td)) {
424 calc_rate(unified_rw_rep, rate_time, io_bytes, rate_io_bytes,
425 je->rate);
426 memcpy(&rate_prev_time, &now, sizeof(now));
427 add_agg_sample(je->rate[DDIR_READ], DDIR_READ, 0);
428 add_agg_sample(je->rate[DDIR_WRITE], DDIR_WRITE, 0);
429 add_agg_sample(je->rate[DDIR_TRIM], DDIR_TRIM, 0);
430 }
431
432 disp_time = mtime_since(&disp_prev_time, &now);
433
434 /*
435 * Allow a little slack, the target is to print it every 1000 msecs
436 */
437 if (!force && disp_time < 900)
438 return 0;
439
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);
442
443 memcpy(&disp_prev_time, &now, sizeof(now));
444
445 if (!force && !je->nr_running && !je->nr_pending)
446 return 0;
447
448 je->nr_threads = thread_number;
449 memcpy(je->run_str, run_str, thread_number * sizeof(char));
450 return 1;
451}
452
453void display_thread_status(struct jobs_eta *je)
454{
455 static struct timeval disp_eta_new_line;
456 static int eta_new_line_init, eta_new_line_pending;
457 static int linelen_last;
458 static int eta_good;
459 char output[REAL_MAX_JOBS + 512], *p = output;
460 char eta_str[128];
461 double perc = 0.0;
462
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);
466 }
467
468 if (eta_new_line_pending) {
469 eta_new_line_pending = 0;
470 p += sprintf(p, "\n");
471 }
472
473 p += sprintf(p, "Jobs: %d (f=%d)", je->nr_running, je->files_open);
474 if (je->m_rate[0] || je->m_rate[1] || je->t_rate[0] || je->t_rate[1]) {
475 char *tr, *mr;
476
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);
479 p += sprintf(p, ", CR=%s/%s KB/s", tr, mr);
480 free(tr);
481 free(mr);
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],
485 je->m_iops[0] + je->m_iops[1]);
486 }
487 if (je->eta_sec != INT_MAX && je->nr_running) {
488 char perc_str[32];
489 char *iops_str[DDIR_RWDIR_CNT];
490 char *rate_str[DDIR_RWDIR_CNT];
491 size_t left;
492 int l;
493 int ddir;
494
495 if ((!je->eta_sec && !eta_good) || je->nr_ramp == je->nr_running)
496 strcpy(perc_str, "-.-% done");
497 else {
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
503 eta_good = 1;
504 perc *= mult;
505 sprintf(perc_str, "%3.1f%% done", perc);
506 }
507
508 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
509 rate_str[ddir] = num2str(je->rate[ddir], 5,
510 1024, je->is_pow2, je->unit_base);
511 iops_str[ddir] = num2str(je->iops[ddir], 4, 1, 0, 0);
512 }
513
514 left = sizeof(output) - (p - output) - 1;
515
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);
521 p += l;
522 if (l >= 0 && l < linelen_last)
523 p += sprintf(p, "%*s", linelen_last - l, "");
524 linelen_last = l;
525
526 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
527 free(rate_str[ddir]);
528 free(iops_str[ddir]);
529 }
530 }
531 p += sprintf(p, "\r");
532
533 printf("%s", output);
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
544 fflush(stdout);
545}
546
547void print_thread_status(void)
548{
549 struct jobs_eta *je;
550 size_t size;
551
552 if (!thread_number)
553 return;
554
555 size = sizeof(*je) + thread_number * sizeof(char) + 1;
556 je = malloc(size);
557 memset(je, 0, size);
558
559 if (calc_thread_status(je, 0))
560 display_thread_status(je);
561
562 free(je);
563}
564
565void print_status_init(int thr_number)
566{
567 run_str[thr_number] = 'P';
568}