Merge branch 'client-hostfile' of git://github.com/bengland2/fio
[fio.git] / eta.c
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
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  */
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  */
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         s = eta_sec % 60;
127         eta_sec /= 60;
128         m = eta_sec % 60;
129         eta_sec /= 60;
130         h = eta_sec % 24;
131         eta_sec /= 24;
132         d = eta_sec;
133
134         if (d) {
135                 disp_hour = 1;
136                 str += sprintf(str, "%02ud:", d);
137         }
138
139         if (h || disp_hour)
140                 str += sprintf(str, "%02uh:", h);
141
142         str += sprintf(str, "%02um:", m);
143         str += sprintf(str, "%02us", s);
144 }
145
146 /*
147  * Best effort calculation of the estimated pending runtime of a job.
148  */
149 static int thread_eta(struct thread_data *td)
150 {
151         unsigned long long bytes_total, bytes_done;
152         unsigned long eta_sec = 0;
153         unsigned long elapsed;
154         uint64_t timeout;
155
156         elapsed = (mtime_since_now(&td->epoch) + 999) / 1000;
157         timeout = td->o.timeout / 1000000UL;
158
159         bytes_total = td->total_io_size;
160
161         if (td->o.fill_device && td->o.size  == -1ULL) {
162                 if (!td->fill_device_size || td->fill_device_size == -1ULL)
163                         return 0;
164
165                 bytes_total = td->fill_device_size;
166         }
167
168         if (td->o.zone_size && td->o.zone_skip && bytes_total) {
169                 unsigned int nr_zones;
170                 uint64_t zone_bytes;
171
172                 zone_bytes = bytes_total + td->o.zone_size + td->o.zone_skip;
173                 nr_zones = (zone_bytes - 1) / (td->o.zone_size + td->o.zone_skip);
174                 bytes_total -= nr_zones * td->o.zone_skip;
175         }
176
177         /*
178          * if writing and verifying afterwards, bytes_total will be twice the
179          * size. In a mixed workload, verify phase will be the size of the
180          * first stage writes.
181          */
182         if (td->o.do_verify && td->o.verify && td_write(td)) {
183                 if (td_rw(td)) {
184                         unsigned int perc = 50;
185
186                         if (td->o.rwmix[DDIR_WRITE])
187                                 perc = td->o.rwmix[DDIR_WRITE];
188
189                         bytes_total += (bytes_total * perc) / 100;
190                 } else
191                         bytes_total <<= 1;
192         }
193
194         if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING) {
195                 double perc, perc_t;
196
197                 bytes_done = ddir_rw_sum(td->io_bytes);
198
199                 if (bytes_total) {
200                         perc = (double) bytes_done / (double) bytes_total;
201                         if (perc > 1.0)
202                                 perc = 1.0;
203                 } else
204                         perc = 0.0;
205
206                 if (td->o.time_based) {
207                         if (timeout) {
208                                 perc_t = (double) elapsed / (double) timeout;
209                                 if (perc_t < perc)
210                                         perc = perc_t;
211                         } else {
212                                 /*
213                                  * Will never hit, we can't have time_based
214                                  * without a timeout set.
215                                  */
216                                 perc = 0.0;
217                         }
218                 }
219
220                 eta_sec = (unsigned long) (elapsed * (1.0 / perc)) - elapsed;
221
222                 if (td->o.timeout &&
223                     eta_sec > (timeout + done_secs - elapsed))
224                         eta_sec = timeout + done_secs - elapsed;
225         } else if (td->runstate == TD_NOT_CREATED || td->runstate == TD_CREATED
226                         || td->runstate == TD_INITIALIZED
227                         || td->runstate == TD_SETTING_UP
228                         || td->runstate == TD_RAMP
229                         || td->runstate == TD_PRE_READING) {
230                 int t_eta = 0, r_eta = 0;
231                 unsigned long long rate_bytes;
232
233                 /*
234                  * We can only guess - assume it'll run the full timeout
235                  * if given, otherwise assume it'll run at the specified rate.
236                  */
237                 if (td->o.timeout) {
238                         uint64_t __timeout = td->o.timeout;
239                         uint64_t start_delay = td->o.start_delay;
240                         uint64_t ramp_time = td->o.ramp_time;
241
242                         t_eta = __timeout + start_delay + ramp_time;
243                         t_eta /= 1000000ULL;
244
245                         if (in_ramp_time(td)) {
246                                 unsigned long ramp_left;
247
248                                 ramp_left = mtime_since_now(&td->epoch);
249                                 ramp_left = (ramp_left + 999) / 1000;
250                                 if (ramp_left <= t_eta)
251                                         t_eta -= ramp_left;
252                         }
253                 }
254                 rate_bytes = ddir_rw_sum(td->o.rate);
255                 if (rate_bytes) {
256                         r_eta = (bytes_total / 1024) / rate_bytes;
257                         r_eta += (td->o.start_delay / 1000000ULL);
258                 }
259
260                 if (r_eta && t_eta)
261                         eta_sec = min(r_eta, t_eta);
262                 else if (r_eta)
263                         eta_sec = r_eta;
264                 else if (t_eta)
265                         eta_sec = t_eta;
266                 else
267                         eta_sec = 0;
268         } else {
269                 /*
270                  * thread is already done or waiting for fsync
271                  */
272                 eta_sec = 0;
273         }
274
275         return eta_sec;
276 }
277
278 static void calc_rate(int unified_rw_rep, unsigned long mtime,
279                       unsigned long long *io_bytes,
280                       unsigned long long *prev_io_bytes, unsigned int *rate)
281 {
282         int i;
283
284         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
285                 unsigned long long diff;
286
287                 diff = io_bytes[i] - prev_io_bytes[i];
288                 if (unified_rw_rep) {
289                         rate[i] = 0;
290                         rate[0] += ((1000 * diff) / mtime) / 1024;
291                 } else
292                         rate[i] = ((1000 * diff) / mtime) / 1024;
293
294                 prev_io_bytes[i] = io_bytes[i];
295         }
296 }
297
298 static void calc_iops(int unified_rw_rep, unsigned long mtime,
299                       unsigned long long *io_iops,
300                       unsigned long long *prev_io_iops, unsigned int *iops)
301 {
302         int i;
303
304         for (i = 0; i < DDIR_RWDIR_CNT; i++) {
305                 unsigned long long diff;
306
307                 diff = io_iops[i] - prev_io_iops[i];
308                 if (unified_rw_rep) {
309                         iops[i] = 0;
310                         iops[0] += (diff * 1000) / mtime;
311                 } else
312                         iops[i] = (diff * 1000) / mtime;
313
314                 prev_io_iops[i] = io_iops[i];
315         }
316 }
317
318 /*
319  * Print status of the jobs we know about. This includes rate estimates,
320  * ETA, thread state, etc.
321  */
322 int calc_thread_status(struct jobs_eta *je, int force)
323 {
324         struct thread_data *td;
325         int i, unified_rw_rep;
326         unsigned long rate_time, disp_time, bw_avg_time, *eta_secs;
327         unsigned long long io_bytes[DDIR_RWDIR_CNT];
328         unsigned long long io_iops[DDIR_RWDIR_CNT];
329         struct timeval now;
330
331         static unsigned long long rate_io_bytes[DDIR_RWDIR_CNT];
332         static unsigned long long disp_io_bytes[DDIR_RWDIR_CNT];
333         static unsigned long long disp_io_iops[DDIR_RWDIR_CNT];
334         static struct timeval rate_prev_time, disp_prev_time;
335
336         if (!force) {
337                 if (output_format != FIO_OUTPUT_NORMAL &&
338                     f_out == stdout)
339                         return 0;
340                 if (temp_stall_ts || eta_print == FIO_ETA_NEVER)
341                         return 0;
342
343                 if (!isatty(STDOUT_FILENO) && (eta_print != FIO_ETA_ALWAYS))
344                         return 0;
345         }
346
347         if (!ddir_rw_sum(rate_io_bytes))
348                 fill_start_time(&rate_prev_time);
349         if (!ddir_rw_sum(disp_io_bytes))
350                 fill_start_time(&disp_prev_time);
351
352         eta_secs = malloc(thread_number * sizeof(unsigned long));
353         memset(eta_secs, 0, thread_number * sizeof(unsigned long));
354
355         je->elapsed_sec = (mtime_since_genesis() + 999) / 1000;
356
357         io_bytes[DDIR_READ] = io_bytes[DDIR_WRITE] = io_bytes[DDIR_TRIM] = 0;
358         io_iops[DDIR_READ] = io_iops[DDIR_WRITE] = io_iops[DDIR_TRIM] = 0;
359         bw_avg_time = ULONG_MAX;
360         unified_rw_rep = 0;
361         for_each_td(td, i) {
362                 unified_rw_rep += td->o.unified_rw_rep;
363                 if (is_power_of_2(td->o.kb_base))
364                         je->is_pow2 = 1;
365                 je->unit_base = td->o.unit_base;
366                 if (td->o.bw_avg_time < bw_avg_time)
367                         bw_avg_time = td->o.bw_avg_time;
368                 if (td->runstate == TD_RUNNING || td->runstate == TD_VERIFYING
369                     || td->runstate == TD_FSYNCING
370                     || td->runstate == TD_PRE_READING
371                     || td->runstate == TD_FINISHING) {
372                         je->nr_running++;
373                         if (td_read(td)) {
374                                 je->t_rate[0] += td->o.rate[DDIR_READ];
375                                 je->t_iops[0] += td->o.rate_iops[DDIR_READ];
376                                 je->m_rate[0] += td->o.ratemin[DDIR_READ];
377                                 je->m_iops[0] += td->o.rate_iops_min[DDIR_READ];
378                         }
379                         if (td_write(td)) {
380                                 je->t_rate[1] += td->o.rate[DDIR_WRITE];
381                                 je->t_iops[1] += td->o.rate_iops[DDIR_WRITE];
382                                 je->m_rate[1] += td->o.ratemin[DDIR_WRITE];
383                                 je->m_iops[1] += td->o.rate_iops_min[DDIR_WRITE];
384                         }
385                         if (td_trim(td)) {
386                                 je->t_rate[2] += td->o.rate[DDIR_TRIM];
387                                 je->t_iops[2] += td->o.rate_iops[DDIR_TRIM];
388                                 je->m_rate[2] += td->o.ratemin[DDIR_TRIM];
389                                 je->m_iops[2] += td->o.rate_iops_min[DDIR_TRIM];
390                         }
391
392                         je->files_open += td->nr_open_files;
393                 } else if (td->runstate == TD_RAMP) {
394                         je->nr_running++;
395                         je->nr_ramp++;
396                 } else if (td->runstate == TD_SETTING_UP)
397                         je->nr_setting_up++;
398                 else if (td->runstate < TD_RUNNING)
399                         je->nr_pending++;
400
401                 if (je->elapsed_sec >= 3)
402                         eta_secs[i] = thread_eta(td);
403                 else
404                         eta_secs[i] = INT_MAX;
405
406                 check_str_update(td);
407
408                 if (td->runstate > TD_SETTING_UP) {
409                         int ddir;
410
411                         for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
412                                 if (unified_rw_rep) {
413                                         io_bytes[0] += td->io_bytes[ddir];
414                                         io_iops[0] += td->io_blocks[ddir];
415                                 } else {
416                                         io_bytes[ddir] += td->io_bytes[ddir];
417                                         io_iops[ddir] += td->io_blocks[ddir];
418                                 }
419                         }
420                 }
421         }
422
423         if (exitall_on_terminate)
424                 je->eta_sec = INT_MAX;
425         else
426                 je->eta_sec = 0;
427
428         for_each_td(td, i) {
429                 if (exitall_on_terminate) {
430                         if (eta_secs[i] < je->eta_sec)
431                                 je->eta_sec = eta_secs[i];
432                 } else {
433                         if (eta_secs[i] > je->eta_sec)
434                                 je->eta_sec = eta_secs[i];
435                 }
436         }
437
438         free(eta_secs);
439
440         fio_gettime(&now, NULL);
441         rate_time = mtime_since(&rate_prev_time, &now);
442
443         if (write_bw_log && rate_time > bw_avg_time && !in_ramp_time(td)) {
444                 calc_rate(unified_rw_rep, rate_time, io_bytes, rate_io_bytes,
445                                 je->rate);
446                 memcpy(&rate_prev_time, &now, sizeof(now));
447                 add_agg_sample(je->rate[DDIR_READ], DDIR_READ, 0);
448                 add_agg_sample(je->rate[DDIR_WRITE], DDIR_WRITE, 0);
449                 add_agg_sample(je->rate[DDIR_TRIM], DDIR_TRIM, 0);
450         }
451
452         disp_time = mtime_since(&disp_prev_time, &now);
453
454         /*
455          * Allow a little slack, the target is to print it every 1000 msecs
456          */
457         if (!force && disp_time < 900)
458                 return 0;
459
460         calc_rate(unified_rw_rep, disp_time, io_bytes, disp_io_bytes, je->rate);
461         calc_iops(unified_rw_rep, disp_time, io_iops, disp_io_iops, je->iops);
462
463         memcpy(&disp_prev_time, &now, sizeof(now));
464
465         if (!force && !je->nr_running && !je->nr_pending)
466                 return 0;
467
468         je->nr_threads = thread_number;
469         update_condensed_str(__run_str, run_str);
470         memcpy(je->run_str, run_str, strlen(run_str));
471         return 1;
472 }
473
474 void display_thread_status(struct jobs_eta *je)
475 {
476         static struct timeval disp_eta_new_line;
477         static int eta_new_line_init, eta_new_line_pending;
478         static int linelen_last;
479         static int eta_good;
480         char output[REAL_MAX_JOBS + 512], *p = output;
481         char eta_str[128];
482         double perc = 0.0;
483
484         if (je->eta_sec != INT_MAX && je->elapsed_sec) {
485                 perc = (double) je->elapsed_sec / (double) (je->elapsed_sec + je->eta_sec);
486                 eta_to_str(eta_str, je->eta_sec);
487         }
488
489         if (eta_new_line_pending) {
490                 eta_new_line_pending = 0;
491                 p += sprintf(p, "\n");
492         }
493
494         p += sprintf(p, "Jobs: %d (f=%d)", je->nr_running, je->files_open);
495         if (je->m_rate[0] || je->m_rate[1] || je->t_rate[0] || je->t_rate[1]) {
496                 char *tr, *mr;
497
498                 mr = num2str(je->m_rate[0] + je->m_rate[1], 4, 0, je->is_pow2, 8);
499                 tr = num2str(je->t_rate[0] + je->t_rate[1], 4, 0, je->is_pow2, 8);
500                 p += sprintf(p, ", CR=%s/%s KB/s", tr, mr);
501                 free(tr);
502                 free(mr);
503         } else if (je->m_iops[0] || je->m_iops[1] || je->t_iops[0] || je->t_iops[1]) {
504                 p += sprintf(p, ", CR=%d/%d IOPS",
505                                         je->t_iops[0] + je->t_iops[1],
506                                         je->m_iops[0] + je->m_iops[1]);
507         }
508         if (je->eta_sec != INT_MAX && je->nr_running) {
509                 char perc_str[32];
510                 char *iops_str[DDIR_RWDIR_CNT];
511                 char *rate_str[DDIR_RWDIR_CNT];
512                 size_t left;
513                 int l;
514                 int ddir;
515
516                 if ((!je->eta_sec && !eta_good) || je->nr_ramp == je->nr_running)
517                         strcpy(perc_str, "-.-% done");
518                 else {
519                         double mult = 100.0;
520
521                         if (je->nr_setting_up && je->nr_running)
522                                 mult *= (1.0 - (double) je->nr_setting_up / (double) je->nr_running);
523
524                         eta_good = 1;
525                         perc *= mult;
526                         sprintf(perc_str, "%3.1f%% done", perc);
527                 }
528
529                 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
530                         rate_str[ddir] = num2str(je->rate[ddir], 5,
531                                                 1024, je->is_pow2, je->unit_base);
532                         iops_str[ddir] = num2str(je->iops[ddir], 4, 1, 0, 0);
533                 }
534
535                 left = sizeof(output) - (p - output) - 1;
536
537                 l = snprintf(p, left, ": [%s] [%s] [%s/%s/%s /s] [%s/%s/%s iops] [eta %s]",
538                                 je->run_str, perc_str, rate_str[DDIR_READ],
539                                 rate_str[DDIR_WRITE], rate_str[DDIR_TRIM],
540                                 iops_str[DDIR_READ], iops_str[DDIR_WRITE],
541                                 iops_str[DDIR_TRIM], eta_str);
542                 p += l;
543                 if (l >= 0 && l < linelen_last)
544                         p += sprintf(p, "%*s", linelen_last - l, "");
545                 linelen_last = l;
546
547                 for (ddir = DDIR_READ; ddir < DDIR_RWDIR_CNT; ddir++) {
548                         free(rate_str[ddir]);
549                         free(iops_str[ddir]);
550                 }
551         }
552         p += sprintf(p, "\r");
553
554         printf("%s", output);
555
556         if (!eta_new_line_init) {
557                 fio_gettime(&disp_eta_new_line, NULL);
558                 eta_new_line_init = 1;
559         } else if (eta_new_line && mtime_since_now(&disp_eta_new_line) > eta_new_line) {
560                 fio_gettime(&disp_eta_new_line, NULL);
561                 eta_new_line_pending = 1;
562         }
563
564         fflush(stdout);
565 }
566
567 struct jobs_eta *get_jobs_eta(int force, size_t *size)
568 {
569         struct jobs_eta *je;
570
571         if (!thread_number)
572                 return NULL;
573
574         *size = sizeof(*je) + THREAD_RUNSTR_SZ;
575         je = malloc(*size);
576         if (!je)
577                 return NULL;
578         memset(je, 0, *size);
579
580         if (!calc_thread_status(je, force)) {
581                 free(je);
582                 return NULL;
583         }
584
585         *size = sizeof(*je) + strlen((char *) je->run_str) + 1;
586         return je;
587 }
588
589 void print_thread_status(void)
590 {
591         struct jobs_eta *je;
592         size_t size;
593
594         je = get_jobs_eta(0, &size);
595         if (je)
596                 display_thread_status(je);
597
598         free(je);
599 }
600
601 void print_status_init(int thr_number)
602 {
603         __run_str[thr_number] = 'P';
604         update_condensed_str(__run_str, run_str);
605 }