eta: improve accuracy for rwmix workloads and verify
[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;
77 case TD_CREATED:
78 c = 'C';
79 break;
80 case TD_INITIALIZED:
9c6f6316 81 case TD_SETTING_UP:
5ec10eaa
JA
82 c = 'I';
83 break;
84 case TD_NOT_CREATED:
85 c = 'P';
86 break;
87 default:
88 log_err("state %d\n", td->runstate);
263e529f
JA
89 }
90
91 run_str[td->thread_number - 1] = c;
92}
93
94/*
95 * Convert seconds to a printable string.
96 */
9c02f6ef 97static void eta_to_str(char *str, unsigned long eta_sec)
263e529f
JA
98{
99 unsigned int d, h, m, s;
165faf16 100 int disp_hour = 0;
263e529f 101
263e529f
JA
102 s = eta_sec % 60;
103 eta_sec /= 60;
104 m = eta_sec % 60;
105 eta_sec /= 60;
106 h = eta_sec % 24;
107 eta_sec /= 24;
108 d = eta_sec;
109
165faf16
JA
110 if (d) {
111 disp_hour = 1;
1e97cce9 112 str += sprintf(str, "%02ud:", d);
263e529f 113 }
165faf16
JA
114
115 if (h || disp_hour)
1e97cce9 116 str += sprintf(str, "%02uh:", h);
263e529f 117
1e97cce9
JA
118 str += sprintf(str, "%02um:", m);
119 str += sprintf(str, "%02us", s);
263e529f
JA
120}
121
122/*
123 * Best effort calculation of the estimated pending runtime of a job.
124 */
b29ee5b3 125static int thread_eta(struct thread_data *td)
263e529f
JA
126{
127 unsigned long long bytes_total, bytes_done;
1e97cce9 128 unsigned long eta_sec = 0;
b29ee5b3
JA
129 unsigned long elapsed;
130
131 elapsed = (mtime_since_now(&td->epoch) + 999) / 1000;
263e529f
JA
132
133 bytes_total = td->total_io_size;
134
2e3bd4c2
JA
135 if (td->o.fill_device && td->o.size == -1ULL) {
136 if (!td->fill_device_size || td->fill_device_size == -1ULL)
137 return 0;
138
139 bytes_total = td->fill_device_size;
140 }
141
74939e38 142 /*
165eb050
JA
143 * if writing and verifying afterwards, bytes_total will be twice the
144 * size. In a mixed workload, verify phase will be the size of the
145 * first stage writes.
74939e38 146 */
0dd8377f 147 if (td->o.do_verify && td->o.verify && td_write(td)) {
165eb050
JA
148 if (td_rw(td)) {
149 unsigned int perc = 50;
150
151 if (td->o.rwmix[DDIR_WRITE])
152 perc = td->o.rwmix[DDIR_WRITE];
153
154 bytes_total += (bytes_total * perc) / 100;
155 } else
74939e38
JA
156 bytes_total <<= 1;
157 }
158
2dc1bbeb
JA
159 if (td->o.zone_size && td->o.zone_skip)
160 bytes_total /= (td->o.zone_skip / td->o.zone_size);
263e529f
JA
161
162 if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING) {
cf4464ca 163 double perc, perc_t;
263e529f 164
342f4be4 165 bytes_done = ddir_rw_sum(td->io_bytes);
263e529f
JA
166 perc = (double) bytes_done / (double) bytes_total;
167 if (perc > 1.0)
168 perc = 1.0;
169
cf4464ca
JA
170 if (td->o.time_based) {
171 perc_t = (double) elapsed / (double) td->o.timeout;
172 if (perc_t < perc)
173 perc = perc_t;
174 }
175
1e97cce9 176 eta_sec = (unsigned long) (elapsed * (1.0 / perc)) - elapsed;
263e529f 177
d3eeeabc
JA
178 if (td->o.timeout &&
179 eta_sec > (td->o.timeout + done_secs - elapsed))
180 eta_sec = td->o.timeout + done_secs - elapsed;
263e529f 181 } else if (td->runstate == TD_NOT_CREATED || td->runstate == TD_CREATED
b29ee5b3 182 || td->runstate == TD_INITIALIZED
b0f65863
JA
183 || td->runstate == TD_RAMP
184 || td->runstate == TD_PRE_READING) {
263e529f 185 int t_eta = 0, r_eta = 0;
342f4be4 186 unsigned long long rate_bytes;
263e529f
JA
187
188 /*
189 * We can only guess - assume it'll run the full timeout
190 * if given, otherwise assume it'll run at the specified rate.
191 */
b29ee5b3 192 if (td->o.timeout) {
cda99fa0
SSP
193 t_eta = td->o.timeout + td->o.start_delay +
194 td->o.ramp_time;
b29ee5b3
JA
195
196 if (in_ramp_time(td)) {
197 unsigned long ramp_left;
198
cda99fa0 199 ramp_left = mtime_since_now(&td->epoch);
b29ee5b3
JA
200 ramp_left = (ramp_left + 999) / 1000;
201 if (ramp_left <= t_eta)
202 t_eta -= ramp_left;
203 }
204 }
342f4be4
JA
205 rate_bytes = ddir_rw_sum(td->o.rate);
206 if (rate_bytes) {
207 r_eta = (bytes_total / 1024) / rate_bytes;
d3eeeabc 208 r_eta += td->o.start_delay;
263e529f
JA
209 }
210
211 if (r_eta && t_eta)
212 eta_sec = min(r_eta, t_eta);
213 else if (r_eta)
214 eta_sec = r_eta;
215 else if (t_eta)
216 eta_sec = t_eta;
217 else
218 eta_sec = 0;
219 } else {
220 /*
221 * thread is already done or waiting for fsync
222 */
223 eta_sec = 0;
224 }
225
226 return eta_sec;
227}
228
46fda8d0
JA
229static void calc_rate(unsigned long mtime, unsigned long long *io_bytes,
230 unsigned long long *prev_io_bytes, unsigned int *rate)
231{
b7f05eb0
JA
232 int i;
233
6eaf09d6 234 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
b7f05eb0
JA
235 unsigned long long diff;
236
237 diff = io_bytes[i] - prev_io_bytes[i];
238 rate[i] = ((1000 * diff) / mtime) / 1024;
6eaf09d6
SL
239
240 prev_io_bytes[i] = io_bytes[i];
b7f05eb0 241 }
46fda8d0 242}
5ec10eaa 243
adb02ba8
JA
244static void calc_iops(unsigned long mtime, unsigned long long *io_iops,
245 unsigned long long *prev_io_iops, unsigned int *iops)
246{
6eaf09d6
SL
247 int i;
248
249 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
250 iops[i] = ((io_iops[i] - prev_io_iops[i]) * 1000) / mtime;
251 prev_io_iops[i] = io_iops[i];
252 }
adb02ba8
JA
253}
254
263e529f
JA
255/*
256 * Print status of the jobs we know about. This includes rate estimates,
257 * ETA, thread state, etc.
258 */
af9c9fb3 259int calc_thread_status(struct jobs_eta *je, int force)
263e529f 260{
34572e28 261 struct thread_data *td;
b75a394f
JA
262 int i;
263 unsigned long rate_time, disp_time, bw_avg_time, *eta_secs;
6eaf09d6
SL
264 unsigned long long io_bytes[DDIR_RWDIR_CNT];
265 unsigned long long io_iops[DDIR_RWDIR_CNT];
46fda8d0
JA
266 struct timeval now;
267
6eaf09d6
SL
268 static unsigned long long rate_io_bytes[DDIR_RWDIR_CNT];
269 static unsigned long long disp_io_bytes[DDIR_RWDIR_CNT];
270 static unsigned long long disp_io_iops[DDIR_RWDIR_CNT];
46fda8d0 271 static struct timeval rate_prev_time, disp_prev_time;
6043c579 272
af9c9fb3 273 if (!force) {
f3afa57e
JA
274 if (output_format != FIO_OUTPUT_NORMAL)
275 return 0;
276 if (temp_stall_ts || eta_print == FIO_ETA_NEVER)
af9c9fb3 277 return 0;
e592a06b 278
af9c9fb3
JA
279 if (!isatty(STDOUT_FILENO) && (eta_print != FIO_ETA_ALWAYS))
280 return 0;
281 }
263e529f 282
342f4be4 283 if (!ddir_rw_sum(rate_io_bytes))
46fda8d0 284 fill_start_time(&rate_prev_time);
342f4be4 285 if (!ddir_rw_sum(disp_io_bytes))
46fda8d0 286 fill_start_time(&disp_prev_time);
6043c579 287
d3eeeabc
JA
288 eta_secs = malloc(thread_number * sizeof(unsigned long));
289 memset(eta_secs, 0, thread_number * sizeof(unsigned long));
263e529f 290
b75a394f
JA
291 je->elapsed_sec = (mtime_since_genesis() + 999) / 1000;
292
6eaf09d6
SL
293 io_bytes[DDIR_READ] = io_bytes[DDIR_WRITE] = io_bytes[DDIR_TRIM] = 0;
294 io_iops[DDIR_READ] = io_iops[DDIR_WRITE] = io_iops[DDIR_TRIM] = 0;
bb3884d8 295 bw_avg_time = ULONG_MAX;
34572e28 296 for_each_td(td, i) {
b7f05eb0
JA
297 if (is_power_of_2(td->o.kb_base))
298 je->is_pow2 = 1;
2dc1bbeb
JA
299 if (td->o.bw_avg_time < bw_avg_time)
300 bw_avg_time = td->o.bw_avg_time;
46fda8d0 301 if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING
b0f65863
JA
302 || td->runstate == TD_FSYNCING
303 || td->runstate == TD_PRE_READING) {
b75a394f 304 je->nr_running++;
242c38db
JA
305 if (td_read(td)) {
306 je->t_rate += td->o.rate[DDIR_READ];
307 je->t_iops += td->o.rate_iops[DDIR_READ];
308 je->m_rate += td->o.ratemin[DDIR_READ];
309 je->m_iops += td->o.rate_iops_min[DDIR_READ];
310 }
311 if (td_write(td)) {
312 je->t_rate += td->o.rate[DDIR_WRITE];
313 je->t_iops += td->o.rate_iops[DDIR_WRITE];
314 je->m_rate += td->o.ratemin[DDIR_WRITE];
315 je->m_iops += td->o.rate_iops_min[DDIR_WRITE];
316 }
6eaf09d6
SL
317 if (td_trim(td)) {
318 je->t_rate += td->o.rate[DDIR_TRIM];
319 je->t_iops += td->o.rate_iops[DDIR_TRIM];
320 je->m_rate += td->o.ratemin[DDIR_TRIM];
321 je->m_iops += td->o.rate_iops_min[DDIR_TRIM];
322 }
323
b75a394f 324 je->files_open += td->nr_open_files;
b29ee5b3 325 } else if (td->runstate == TD_RAMP) {
b75a394f
JA
326 je->nr_running++;
327 je->nr_ramp++;
9c6f6316
JA
328 } else if (td->runstate == TD_SETTING_UP)
329 je->nr_running++;
330 else if (td->runstate < TD_RUNNING)
b75a394f 331 je->nr_pending++;
263e529f 332
b75a394f 333 if (je->elapsed_sec >= 3)
b29ee5b3 334 eta_secs[i] = thread_eta(td);
263e529f
JA
335 else
336 eta_secs[i] = INT_MAX;
337
338 check_str_update(td);
b29ee5b3
JA
339
340 if (td->runstate > TD_RAMP) {
6eaf09d6
SL
341 int ddir;
342 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
343 io_bytes[ddir] += td->io_bytes[ddir];
344 io_iops[ddir] += td->io_blocks[ddir];
345 }
b29ee5b3 346 }
263e529f
JA
347 }
348
349 if (exitall_on_terminate)
b75a394f 350 je->eta_sec = INT_MAX;
263e529f 351 else
b75a394f 352 je->eta_sec = 0;
263e529f 353
34572e28 354 for_each_td(td, i) {
9c5a3854 355 if (exitall_on_terminate) {
b75a394f
JA
356 if (eta_secs[i] < je->eta_sec)
357 je->eta_sec = eta_secs[i];
9c5a3854 358 } else {
b75a394f
JA
359 if (eta_secs[i] > je->eta_sec)
360 je->eta_sec = eta_secs[i];
9c5a3854 361 }
263e529f
JA
362 }
363
eecf272f
JA
364 free(eta_secs);
365
46fda8d0
JA
366 fio_gettime(&now, NULL);
367 rate_time = mtime_since(&rate_prev_time, &now);
368
b29ee5b3 369 if (write_bw_log && rate_time > bw_avg_time && !in_ramp_time(td)) {
b75a394f 370 calc_rate(rate_time, io_bytes, rate_io_bytes, je->rate);
46fda8d0 371 memcpy(&rate_prev_time, &now, sizeof(now));
b75a394f
JA
372 add_agg_sample(je->rate[DDIR_READ], DDIR_READ, 0);
373 add_agg_sample(je->rate[DDIR_WRITE], DDIR_WRITE, 0);
6eaf09d6 374 add_agg_sample(je->rate[DDIR_TRIM], DDIR_TRIM, 0);
6043c579
JA
375 }
376
46fda8d0 377 disp_time = mtime_since(&disp_prev_time, &now);
a5b01f1b
JA
378
379 /*
380 * Allow a little slack, the target is to print it every 1000 msecs
381 */
af9c9fb3 382 if (!force && disp_time < 900)
b75a394f 383 return 0;
46fda8d0 384
b75a394f
JA
385 calc_rate(disp_time, io_bytes, disp_io_bytes, je->rate);
386 calc_iops(disp_time, io_iops, disp_io_iops, je->iops);
adb02ba8 387
46fda8d0
JA
388 memcpy(&disp_prev_time, &now, sizeof(now));
389
af9c9fb3 390 if (!force && !je->nr_running && !je->nr_pending)
b75a394f
JA
391 return 0;
392
1d1f45ae
JA
393 je->nr_threads = thread_number;
394 memcpy(je->run_str, run_str, thread_number * sizeof(char));
395
b75a394f
JA
396 return 1;
397}
398
cf451d1e 399void display_thread_status(struct jobs_eta *je)
b75a394f 400{
b75a394f
JA
401 static int linelen_last;
402 static int eta_good;
79074e7c 403 char output[REAL_MAX_JOBS + 512], *p = output;
b75a394f
JA
404 char eta_str[128];
405 double perc = 0.0;
b75a394f 406
cf451d1e
JA
407 if (je->eta_sec != INT_MAX && je->elapsed_sec) {
408 perc = (double) je->elapsed_sec / (double) (je->elapsed_sec + je->eta_sec);
409 eta_to_str(eta_str, je->eta_sec);
b75a394f
JA
410 }
411
cf451d1e
JA
412 p += sprintf(p, "Jobs: %d (f=%d)", je->nr_running, je->files_open);
413 if (je->m_rate || je->t_rate) {
581e7141
JA
414 char *tr, *mr;
415
b7f05eb0
JA
416 mr = num2str(je->m_rate, 4, 0, je->is_pow2);
417 tr = num2str(je->t_rate, 4, 0, je->is_pow2);
37db14fe 418 p += sprintf(p, ", CR=%s/%s KB/s", tr, mr);
581e7141
JA
419 free(tr);
420 free(mr);
cf451d1e
JA
421 } else if (je->m_iops || je->t_iops)
422 p += sprintf(p, ", CR=%d/%d IOPS", je->t_iops, je->m_iops);
423 if (je->eta_sec != INT_MAX && je->nr_running) {
0721d11e 424 char perc_str[32];
6eaf09d6
SL
425 char *iops_str[DDIR_RWDIR_CNT];
426 char *rate_str[DDIR_RWDIR_CNT];
79074e7c 427 size_t left;
adb02ba8 428 int l;
6eaf09d6 429 int ddir;
5ec10eaa 430
cf451d1e 431 if ((!je->eta_sec && !eta_good) || je->nr_ramp == je->nr_running)
cac58fd5
JA
432 strcpy(perc_str, "-.-% done");
433 else {
434 eta_good = 1;
435 perc *= 100.0;
0721d11e 436 sprintf(perc_str, "%3.1f%% done", perc);
cac58fd5 437 }
0721d11e 438
6eaf09d6
SL
439 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
440 rate_str[ddir] = num2str(je->rate[ddir], 5,
441 1024, je->is_pow2);
442 iops_str[ddir] = num2str(je->iops[ddir], 4, 1, 0);
443 }
adb02ba8 444
79074e7c
JA
445 left = sizeof(output) - (p - output) - 1;
446
6eaf09d6
SL
447 l = snprintf(p, left, ": [%s] [%s] [%s/%s/%s /s] [%s/%s/%s iops] [eta %s]",
448 je->run_str, perc_str, rate_str[DDIR_READ],
449 rate_str[DDIR_WRITE], rate_str[DDIR_TRIM],
450 iops_str[DDIR_READ], iops_str[DDIR_WRITE],
451 iops_str[DDIR_TRIM], eta_str);
37db14fe 452 p += l;
adb02ba8 453 if (l >= 0 && l < linelen_last)
37db14fe 454 p += sprintf(p, "%*s", linelen_last - l, "");
adb02ba8 455 linelen_last = l;
d1bd7213 456
6eaf09d6
SL
457 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
458 free(rate_str[ddir]);
459 free(iops_str[ddir]);
460 }
263e529f 461 }
37db14fe
JA
462 p += sprintf(p, "\r");
463
cf451d1e
JA
464 printf("%s", output);
465 fflush(stdout);
263e529f
JA
466}
467
cf451d1e
JA
468void print_thread_status(void)
469{
1d1f45ae 470 struct jobs_eta *je;
b814fb26 471 size_t size;
1d1f45ae 472
b814fb26
JA
473 if (!thread_number)
474 return;
cf451d1e 475
b814fb26
JA
476 size = sizeof(*je) + thread_number * sizeof(char) + 1;
477 je = malloc(size);
478 memset(je, 0, size);
cf451d1e 479
af9c9fb3 480 if (calc_thread_status(je, 0))
1d1f45ae 481 display_thread_status(je);
cf451d1e 482
1d1f45ae 483 free(je);
cf451d1e 484}
b75a394f 485
2b13e716 486void print_status_init(int thr_number)
263e529f 487{
2b13e716 488 run_str[thr_number] = 'P';
263e529f 489}