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
10 static char __run_str[REAL_MAX_JOBS + 1];
11 static char run_str[__THREAD_RUNSTR_SZ(REAL_MAX_JOBS)];
12
update_condensed_str(char * rstr,char * run_str_condensed)13 static void update_condensed_str(char *rstr, char *run_str_condensed)
14 {
15 if (*rstr) {
16 while (*rstr) {
17 int nr = 1;
18
19 *run_str_condensed++ = *rstr++;
20 while (*(rstr - 1) == *rstr) {
21 rstr++;
22 nr++;
23 }
24 run_str_condensed += sprintf(run_str_condensed, "(%u),", nr);
25 }
26 run_str_condensed--;
27 }
28 *run_str_condensed = '\0';
29 }
30
31 /*
32 * Sets the status of the 'td' in the printed status map.
33 */
check_str_update(struct thread_data * td)34 static void check_str_update(struct thread_data *td)
35 {
36 char c = __run_str[td->thread_number - 1];
37
38 switch (td->runstate) {
39 case TD_REAPED:
40 if (td->error)
41 c = 'X';
42 else if (td->sig)
43 c = 'K';
44 else
45 c = '_';
46 break;
47 case TD_EXITED:
48 c = 'E';
49 break;
50 case TD_RAMP:
51 c = '/';
52 break;
53 case TD_RUNNING:
54 if (td_rw(td)) {
55 if (td_random(td)) {
56 if (td->o.rwmix[DDIR_READ] == 100)
57 c = 'r';
58 else if (td->o.rwmix[DDIR_WRITE] == 100)
59 c = 'w';
60 else
61 c = 'm';
62 } else {
63 if (td->o.rwmix[DDIR_READ] == 100)
64 c = 'R';
65 else if (td->o.rwmix[DDIR_WRITE] == 100)
66 c = 'W';
67 else
68 c = 'M';
69 }
70 } else if (td_read(td)) {
71 if (td_random(td))
72 c = 'r';
73 else
74 c = 'R';
75 } else if (td_write(td)) {
76 if (td_random(td))
77 c = 'w';
78 else
79 c = 'W';
80 } else {
81 if (td_random(td))
82 c = 'd';
83 else
84 c = 'D';
85 }
86 break;
87 case TD_PRE_READING:
88 c = 'p';
89 break;
90 case TD_VERIFYING:
91 c = 'V';
92 break;
93 case TD_FSYNCING:
94 c = 'F';
95 break;
96 case TD_FINISHING:
97 c = 'f';
98 break;
99 case TD_CREATED:
100 c = 'C';
101 break;
102 case TD_INITIALIZED:
103 case TD_SETTING_UP:
104 c = 'I';
105 break;
106 case TD_NOT_CREATED:
107 c = 'P';
108 break;
109 default:
110 log_err("state %d\n", td->runstate);
111 }
112
113 __run_str[td->thread_number - 1] = c;
114 update_condensed_str(__run_str, run_str);
115 }
116
117 /*
118 * Convert seconds to a printable string.
119 */
eta_to_str(char * str,unsigned long eta_sec)120 void eta_to_str(char *str, unsigned long eta_sec)
121 {
122 unsigned int d, h, m, s;
123 int disp_hour = 0;
124
125 s = eta_sec % 60;
126 eta_sec /= 60;
127 m = eta_sec % 60;
128 eta_sec /= 60;
129 h = eta_sec % 24;
130 eta_sec /= 24;
131 d = eta_sec;
132
133 if (d) {
134 disp_hour = 1;
135 str += sprintf(str, "%02ud:", d);
136 }
137
138 if (h || disp_hour)
139 str += sprintf(str, "%02uh:", h);
140
141 str += sprintf(str, "%02um:", m);
142 str += sprintf(str, "%02us", s);
143 }
144
145 /*
146 * Best effort calculation of the estimated pending runtime of a job.
147 */
thread_eta(struct thread_data * td)148 static int thread_eta(struct thread_data *td)
149 {
150 unsigned long long bytes_total, bytes_done;
151 unsigned long eta_sec = 0;
152 unsigned long elapsed;
153 uint64_t timeout;
154
155 elapsed = (mtime_since_now(&td->epoch) + 999) / 1000;
156 timeout = td->o.timeout / 1000000UL;
157
158 bytes_total = td->total_io_size;
159
160 if (td->o.fill_device && td->o.size == -1ULL) {
161 if (!td->fill_device_size || td->fill_device_size == -1ULL)
162 return 0;
163
164 bytes_total = td->fill_device_size;
165 }
166
167 if (td->o.zone_size && td->o.zone_skip && bytes_total) {
168 unsigned int nr_zones;
169 uint64_t zone_bytes;
170
171 zone_bytes = bytes_total + td->o.zone_size + td->o.zone_skip;
172 nr_zones = (zone_bytes - 1) / (td->o.zone_size + td->o.zone_skip);
173 bytes_total -= nr_zones * td->o.zone_skip;
174 }
175
176 /*
177 * if writing and verifying afterwards, bytes_total will be twice the
178 * size. In a mixed workload, verify phase will be the size of the
179 * first stage writes.
180 */
181 if (td->o.do_verify && td->o.verify && td_write(td)) {
182 if (td_rw(td)) {
183 unsigned int perc = 50;
184
185 if (td->o.rwmix[DDIR_WRITE])
186 perc = td->o.rwmix[DDIR_WRITE];
187
188 bytes_total += (bytes_total * perc) / 100;
189 } else
190 bytes_total <<= 1;
191 }
192
193 if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING) {
194 double perc, perc_t;
195
196 bytes_done = ddir_rw_sum(td->io_bytes);
197
198 if (bytes_total) {
199 perc = (double) bytes_done / (double) bytes_total;
200 if (perc > 1.0)
201 perc = 1.0;
202 } else
203 perc = 0.0;
204
205 if (td->o.time_based) {
206 if (timeout) {
207 perc_t = (double) elapsed / (double) timeout;
208 if (perc_t < perc)
209 perc = perc_t;
210 } else {
211 /*
212 * Will never hit, we can't have time_based
213 * without a timeout set.
214 */
215 perc = 0.0;
216 }
217 }
218
219 eta_sec = (unsigned long) (elapsed * (1.0 / perc)) - elapsed;
220
221 if (td->o.timeout &&
222 eta_sec > (timeout + done_secs - elapsed))
223 eta_sec = timeout + done_secs - elapsed;
224 } else if (td->runstate == TD_NOT_CREATED || td->runstate == TD_CREATED
225 || td->runstate == TD_INITIALIZED
226 || td->runstate == TD_SETTING_UP
227 || td->runstate == TD_RAMP
228 || td->runstate == TD_PRE_READING) {
229 int t_eta = 0, r_eta = 0;
230 unsigned long long rate_bytes;
231
232 /*
233 * We can only guess - assume it'll run the full timeout
234 * if given, otherwise assume it'll run at the specified rate.
235 */
236 if (td->o.timeout) {
237 uint64_t __timeout = td->o.timeout;
238 uint64_t start_delay = td->o.start_delay;
239 uint64_t ramp_time = td->o.ramp_time;
240
241 t_eta = __timeout + start_delay + ramp_time;
242 t_eta /= 1000000ULL;
243
244 if (in_ramp_time(td)) {
245 unsigned long ramp_left;
246
247 ramp_left = mtime_since_now(&td->epoch);
248 ramp_left = (ramp_left + 999) / 1000;
249 if (ramp_left <= t_eta)
250 t_eta -= ramp_left;
251 }
252 }
253 rate_bytes = ddir_rw_sum(td->o.rate);
254 if (rate_bytes) {
255 r_eta = (bytes_total / 1024) / rate_bytes;
256 r_eta += (td->o.start_delay / 1000000ULL);
257 }
258
259 if (r_eta && t_eta)
260 eta_sec = min(r_eta, t_eta);
261 else if (r_eta)
262 eta_sec = r_eta;
263 else if (t_eta)
264 eta_sec = t_eta;
265 else
266 eta_sec = 0;
267 } else {
268 /*
269 * thread is already done or waiting for fsync
270 */
271 eta_sec = 0;
272 }
273
274 return eta_sec;
275 }
276
calc_rate(int unified_rw_rep,unsigned long mtime,unsigned long long * io_bytes,unsigned long long * prev_io_bytes,unsigned int * rate)277 static void calc_rate(int unified_rw_rep, unsigned long mtime,
278 unsigned long long *io_bytes,
279 unsigned long long *prev_io_bytes, unsigned int *rate)
280 {
281 int i;
282
283 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
284 unsigned long long diff;
285
286 diff = io_bytes[i] - prev_io_bytes[i];
287 if (unified_rw_rep) {
288 rate[i] = 0;
289 rate[0] += ((1000 * diff) / mtime) / 1024;
290 } else
291 rate[i] = ((1000 * diff) / mtime) / 1024;
292
293 prev_io_bytes[i] = io_bytes[i];
294 }
295 }
296
calc_iops(int unified_rw_rep,unsigned long mtime,unsigned long long * io_iops,unsigned long long * prev_io_iops,unsigned int * iops)297 static void calc_iops(int unified_rw_rep, unsigned long mtime,
298 unsigned long long *io_iops,
299 unsigned long long *prev_io_iops, unsigned int *iops)
300 {
301 int i;
302
303 for (i = 0; i < DDIR_RWDIR_CNT; i++) {
304 unsigned long long diff;
305
306 diff = io_iops[i] - prev_io_iops[i];
307 if (unified_rw_rep) {
308 iops[i] = 0;
309 iops[0] += (diff * 1000) / mtime;
310 } else
311 iops[i] = (diff * 1000) / mtime;
312
313 prev_io_iops[i] = io_iops[i];
314 }
315 }
316
317 /*
318 * Print status of the jobs we know about. This includes rate estimates,
319 * ETA, thread state, etc.
320 */
calc_thread_status(struct jobs_eta * je,int force)321 int calc_thread_status(struct jobs_eta *je, int force)
322 {
323 struct thread_data *td;
324 int i, unified_rw_rep;
325 unsigned long rate_time, disp_time, bw_avg_time, *eta_secs;
326 unsigned long long io_bytes[DDIR_RWDIR_CNT];
327 unsigned long long io_iops[DDIR_RWDIR_CNT];
328 struct timeval now;
329
330 static unsigned long long rate_io_bytes[DDIR_RWDIR_CNT];
331 static unsigned long long disp_io_bytes[DDIR_RWDIR_CNT];
332 static unsigned long long disp_io_iops[DDIR_RWDIR_CNT];
333 static struct timeval rate_prev_time, disp_prev_time;
334
335 if (!force) {
336 if (output_format != FIO_OUTPUT_NORMAL &&
337 f_out == stdout)
338 return 0;
339 if (temp_stall_ts || eta_print == FIO_ETA_NEVER)
340 return 0;
341
342 if (!isatty(STDOUT_FILENO) && (eta_print != FIO_ETA_ALWAYS))
343 return 0;
344 }
345
346 if (!ddir_rw_sum(rate_io_bytes))
347 fill_start_time(&rate_prev_time);
348 if (!ddir_rw_sum(disp_io_bytes))
349 fill_start_time(&disp_prev_time);
350
351 eta_secs = malloc(thread_number * sizeof(unsigned long));
352 memset(eta_secs, 0, thread_number * sizeof(unsigned long));
353
354 je->elapsed_sec = (mtime_since_genesis() + 999) / 1000;
355
356 io_bytes[DDIR_READ] = io_bytes[DDIR_WRITE] = io_bytes[DDIR_TRIM] = 0;
357 io_iops[DDIR_READ] = io_iops[DDIR_WRITE] = io_iops[DDIR_TRIM] = 0;
358 bw_avg_time = ULONG_MAX;
359 unified_rw_rep = 0;
360 for_each_td(td, i) {
361 unified_rw_rep += td->o.unified_rw_rep;
362 if (is_power_of_2(td->o.kb_base))
363 je->is_pow2 = 1;
364 je->unit_base = td->o.unit_base;
365 if (td->o.bw_avg_time < bw_avg_time)
366 bw_avg_time = td->o.bw_avg_time;
367 if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING
368 || td->runstate == TD_FSYNCING
369 || td->runstate == TD_PRE_READING
370 || td->runstate == TD_FINISHING) {
371 je->nr_running++;
372 if (td_read(td)) {
373 je->t_rate[0] += td->o.rate[DDIR_READ];
374 je->t_iops[0] += td->o.rate_iops[DDIR_READ];
375 je->m_rate[0] += td->o.ratemin[DDIR_READ];
376 je->m_iops[0] += td->o.rate_iops_min[DDIR_READ];
377 }
378 if (td_write(td)) {
379 je->t_rate[1] += td->o.rate[DDIR_WRITE];
380 je->t_iops[1] += td->o.rate_iops[DDIR_WRITE];
381 je->m_rate[1] += td->o.ratemin[DDIR_WRITE];
382 je->m_iops[1] += td->o.rate_iops_min[DDIR_WRITE];
383 }
384 if (td_trim(td)) {
385 je->t_rate[2] += td->o.rate[DDIR_TRIM];
386 je->t_iops[2] += td->o.rate_iops[DDIR_TRIM];
387 je->m_rate[2] += td->o.ratemin[DDIR_TRIM];
388 je->m_iops[2] += td->o.rate_iops_min[DDIR_TRIM];
389 }
390
391 je->files_open += td->nr_open_files;
392 } else if (td->runstate == TD_RAMP) {
393 je->nr_running++;
394 je->nr_ramp++;
395 } else if (td->runstate == TD_SETTING_UP)
396 je->nr_setting_up++;
397 else if (td->runstate < TD_RUNNING)
398 je->nr_pending++;
399
400 if (je->elapsed_sec >= 3)
401 eta_secs[i] = thread_eta(td);
402 else
403 eta_secs[i] = INT_MAX;
404
405 check_str_update(td);
406
407 if (td->runstate > TD_SETTING_UP) {
408 int ddir;
409
410 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
411 if (unified_rw_rep) {
412 io_bytes[0] += td->io_bytes[ddir];
413 io_iops[0] += td->io_blocks[ddir];
414 } else {
415 io_bytes[ddir] += td->io_bytes[ddir];
416 io_iops[ddir] += td->io_blocks[ddir];
417 }
418 }
419 }
420 }
421
422 if (exitall_on_terminate)
423 je->eta_sec = INT_MAX;
424 else
425 je->eta_sec = 0;
426
427 for_each_td(td, i) {
428 if (exitall_on_terminate) {
429 if (eta_secs[i] < je->eta_sec)
430 je->eta_sec = eta_secs[i];
431 } else {
432 if (eta_secs[i] > je->eta_sec)
433 je->eta_sec = eta_secs[i];
434 }
435 }
436
437 free(eta_secs);
438
439 fio_gettime(&now, NULL);
440 rate_time = mtime_since(&rate_prev_time, &now);
441
442 if (write_bw_log && rate_time > bw_avg_time && !in_ramp_time(td)) {
443 calc_rate(unified_rw_rep, rate_time, io_bytes, rate_io_bytes,
444 je->rate);
445 memcpy(&rate_prev_time, &now, sizeof(now));
446 add_agg_sample(je->rate[DDIR_READ], DDIR_READ, 0);
447 add_agg_sample(je->rate[DDIR_WRITE], DDIR_WRITE, 0);
448 add_agg_sample(je->rate[DDIR_TRIM], DDIR_TRIM, 0);
449 }
450
451 disp_time = mtime_since(&disp_prev_time, &now);
452
453 /*
454 * Allow a little slack, the target is to print it every 1000 msecs
455 */
456 if (!force && disp_time < 900)
457 return 0;
458
459 calc_rate(unified_rw_rep, disp_time, io_bytes, disp_io_bytes, je->rate);
460 calc_iops(unified_rw_rep, disp_time, io_iops, disp_io_iops, je->iops);
461
462 memcpy(&disp_prev_time, &now, sizeof(now));
463
464 if (!force && !je->nr_running && !je->nr_pending)
465 return 0;
466
467 je->nr_threads = thread_number;
468 update_condensed_str(__run_str, run_str);
469 memcpy(je->run_str, run_str, strlen(run_str));
470 return 1;
471 }
472
display_thread_status(struct jobs_eta * je)473 void display_thread_status(struct jobs_eta *je)
474 {
475 static struct timeval disp_eta_new_line;
476 static int eta_new_line_init, eta_new_line_pending;
477 static int linelen_last;
478 static int eta_good;
479 char output[REAL_MAX_JOBS + 512], *p = output;
480 char eta_str[128];
481 double perc = 0.0;
482
483 if (je->eta_sec != INT_MAX && je->elapsed_sec) {
484 perc = (double) je->elapsed_sec / (double) (je->elapsed_sec + je->eta_sec);
485 eta_to_str(eta_str, je->eta_sec);
486 }
487
488 if (eta_new_line_pending) {
489 eta_new_line_pending = 0;
490 p += sprintf(p, "\n");
491 }
492
493 p += sprintf(p, "Jobs: %d (f=%d)", je->nr_running, je->files_open);
494 if (je->m_rate[0] || je->m_rate[1] || je->t_rate[0] || je->t_rate[1]) {
495 char *tr, *mr;
496
497 mr = num2str(je->m_rate[0] + je->m_rate[1], 4, 0, je->is_pow2, 8);
498 tr = num2str(je->t_rate[0] + je->t_rate[1], 4, 0, je->is_pow2, 8);
499 p += sprintf(p, ", CR=%s/%s KB/s", tr, mr);
500 free(tr);
501 free(mr);
502 } else if (je->m_iops[0] || je->m_iops[1] || je->t_iops[0] || je->t_iops[1]) {
503 p += sprintf(p, ", CR=%d/%d IOPS",
504 je->t_iops[0] + je->t_iops[1],
505 je->m_iops[0] + je->m_iops[1]);
506 }
507 if (je->eta_sec != INT_MAX && je->nr_running) {
508 char perc_str[32];
509 char *iops_str[DDIR_RWDIR_CNT];
510 char *rate_str[DDIR_RWDIR_CNT];
511 size_t left;
512 int l;
513 int ddir;
514
515 if ((!je->eta_sec && !eta_good) || je->nr_ramp == je->nr_running)
516 strcpy(perc_str, "-.-% done");
517 else {
518 double mult = 100.0;
519
520 if (je->nr_setting_up && je->nr_running)
521 mult *= (1.0 - (double) je->nr_setting_up / (double) je->nr_running);
522
523 eta_good = 1;
524 perc *= mult;
525 sprintf(perc_str, "%3.1f%% done", perc);
526 }
527
528 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
529 rate_str[ddir] = num2str(je->rate[ddir], 5,
530 1024, je->is_pow2, je->unit_base);
531 iops_str[ddir] = num2str(je->iops[ddir], 4, 1, 0, 0);
532 }
533
534 left = sizeof(output) - (p - output) - 1;
535
536 l = snprintf(p, left, ": [%s] [%s] [%s/%s/%s /s] [%s/%s/%s iops] [eta %s]",
537 je->run_str, perc_str, rate_str[DDIR_READ],
538 rate_str[DDIR_WRITE], rate_str[DDIR_TRIM],
539 iops_str[DDIR_READ], iops_str[DDIR_WRITE],
540 iops_str[DDIR_TRIM], eta_str);
541 p += l;
542 if (l >= 0 && l < linelen_last)
543 p += sprintf(p, "%*s", linelen_last - l, "");
544 linelen_last = l;
545
546 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
547 free(rate_str[ddir]);
548 free(iops_str[ddir]);
549 }
550 }
551 p += sprintf(p, "\r");
552
553 printf("%s", output);
554
555 if (!eta_new_line_init) {
556 fio_gettime(&disp_eta_new_line, NULL);
557 eta_new_line_init = 1;
558 } else if (eta_new_line && mtime_since_now(&disp_eta_new_line) > eta_new_line) {
559 fio_gettime(&disp_eta_new_line, NULL);
560 eta_new_line_pending = 1;
561 }
562
563 fflush(stdout);
564 }
565
get_jobs_eta(int force,size_t * size)566 struct jobs_eta *get_jobs_eta(int force, size_t *size)
567 {
568 struct jobs_eta *je;
569
570 if (!thread_number)
571 return NULL;
572
573 *size = sizeof(*je) + THREAD_RUNSTR_SZ;
574 je = malloc(*size);
575 if (!je)
576 return NULL;
577 memset(je, 0, *size);
578
579 if (!calc_thread_status(je, force)) {
580 free(je);
581 return NULL;
582 }
583
584 *size = sizeof(*je) + strlen((char *) je->run_str) + 1;
585 return je;
586 }
587
print_thread_status(void)588 void print_thread_status(void)
589 {
590 struct jobs_eta *je;
591 size_t size;
592
593 je = get_jobs_eta(0, &size);
594 if (je)
595 display_thread_status(je);
596
597 free(je);
598 }
599
print_status_init(int thr_number)600 void print_status_init(int thr_number)
601 {
602 __run_str[thr_number] = 'P';
603 update_condensed_str(__run_str, run_str);
604 }
605