[PATCH] Fix random seed for numjobs > 1
[fio.git] / stat.c
1 #include <stdio.h>
2 #include <string.h>
3 #include <sys/time.h>
4 #include <sys/types.h>
5 #include <sys/stat.h>
6 #include <dirent.h>
7 #include <libgen.h>
8 #include <math.h>
9
10 #include "fio.h"
11
12 static struct itimerval itimer;
13 static struct list_head disk_list = LIST_HEAD_INIT(disk_list);
14
15 /*
16  * Cheasy number->string conversion, complete with carry rounding error.
17  */
18 static char *num2str(unsigned long num, int maxlen, int base)
19 {
20         /*
21          * could be passed in for 10^3 base, but every caller expects
22          * 2^10 base right now.
23          */
24         const unsigned int thousand = 1024;
25         char postfix[] = { 'K', 'M', 'G', 'P', 'E' };
26         char *buf;
27         int i;
28
29         buf = malloc(128);
30
31         for (i = 0; base > 1; i++)
32                 base /= thousand;
33
34         do {
35                 int len, carry = 0;
36
37                 len = sprintf(buf, "%'lu", num);
38                 if (len <= maxlen) {
39                         buf[len] = postfix[i];
40                         buf[len + 1] = '\0';
41                         return buf;
42                 }
43
44                 if ((num % thousand) >= (thousand / 2))
45                         carry = 1;
46
47                 num /= thousand;
48                 num += carry;
49                 i++;
50         } while (i <= 5);
51
52         return buf;
53 }
54
55 static int get_io_ticks(struct disk_util *du, struct disk_util_stat *dus)
56 {
57         unsigned in_flight;
58         char line[256];
59         FILE *f;
60         char *p;
61
62         f = fopen(du->path, "r");
63         if (!f)
64                 return 1;
65
66         p = fgets(line, sizeof(line), f);
67         if (!p) {
68                 fclose(f);
69                 return 1;
70         }
71
72         if (sscanf(p, "%u %u %llu %u %u %u %llu %u %u %u %u\n", &dus->ios[0], &dus->merges[0], &dus->sectors[0], &dus->ticks[0], &dus->ios[1], &dus->merges[1], &dus->sectors[1], &dus->ticks[1], &in_flight, &dus->io_ticks, &dus->time_in_queue) != 11) {
73                 fclose(f);
74                 return 1;
75         }
76
77         fclose(f);
78         return 0;
79 }
80
81 static void update_io_tick_disk(struct disk_util *du)
82 {
83         struct disk_util_stat __dus, *dus, *ldus;
84         struct timeval t;
85
86         if (get_io_ticks(du, &__dus))
87                 return;
88
89         dus = &du->dus;
90         ldus = &du->last_dus;
91
92         dus->sectors[0] += (__dus.sectors[0] - ldus->sectors[0]);
93         dus->sectors[1] += (__dus.sectors[1] - ldus->sectors[1]);
94         dus->ios[0] += (__dus.ios[0] - ldus->ios[0]);
95         dus->ios[1] += (__dus.ios[1] - ldus->ios[1]);
96         dus->merges[0] += (__dus.merges[0] - ldus->merges[0]);
97         dus->merges[1] += (__dus.merges[1] - ldus->merges[1]);
98         dus->ticks[0] += (__dus.ticks[0] - ldus->ticks[0]);
99         dus->ticks[1] += (__dus.ticks[1] - ldus->ticks[1]);
100         dus->io_ticks += (__dus.io_ticks - ldus->io_ticks);
101         dus->time_in_queue += (__dus.time_in_queue - ldus->time_in_queue);
102
103         fio_gettime(&t, NULL);
104         du->msec += mtime_since(&du->time, &t);
105         memcpy(&du->time, &t, sizeof(t));
106         memcpy(ldus, &__dus, sizeof(__dus));
107 }
108
109 void update_io_ticks(void)
110 {
111         struct list_head *entry;
112         struct disk_util *du;
113
114         list_for_each(entry, &disk_list) {
115                 du = list_entry(entry, struct disk_util, list);
116                 update_io_tick_disk(du);
117         }
118 }
119
120 static int disk_util_exists(dev_t dev)
121 {
122         struct list_head *entry;
123         struct disk_util *du;
124
125         list_for_each(entry, &disk_list) {
126                 du = list_entry(entry, struct disk_util, list);
127
128                 if (du->dev == dev)
129                         return 1;
130         }
131
132         return 0;
133 }
134
135 static void disk_util_add(dev_t dev, char *path)
136 {
137         struct disk_util *du, *__du;
138         struct list_head *entry;
139
140         du = malloc(sizeof(*du));
141         memset(du, 0, sizeof(*du));
142         INIT_LIST_HEAD(&du->list);
143         sprintf(du->path, "%s/stat", path);
144         du->name = strdup(basename(path));
145         du->dev = dev;
146
147         list_for_each(entry, &disk_list) {
148                 __du = list_entry(entry, struct disk_util, list);
149
150                 if (!strcmp(du->name, __du->name)) {
151                         free(du->name);
152                         free(du);
153                         return;
154                 }
155         }
156
157         fio_gettime(&du->time, NULL);
158         get_io_ticks(du, &du->last_dus);
159
160         list_add_tail(&du->list, &disk_list);
161 }
162
163 static int check_dev_match(dev_t dev, char *path)
164 {
165         unsigned int major, minor;
166         char line[256], *p;
167         FILE *f;
168
169         f = fopen(path, "r");
170         if (!f) {
171                 perror("open path");
172                 return 1;
173         }
174
175         p = fgets(line, sizeof(line), f);
176         if (!p) {
177                 fclose(f);
178                 return 1;
179         }
180
181         if (sscanf(p, "%u:%u", &major, &minor) != 2) {
182                 fclose(f);
183                 return 1;
184         }
185
186         if (((major << 8) | minor) == dev) {
187                 fclose(f);
188                 return 0;
189         }
190
191         fclose(f);
192         return 1;
193 }
194
195 static int find_block_dir(dev_t dev, char *path)
196 {
197         struct dirent *dir;
198         struct stat st;
199         int found = 0;
200         DIR *D;
201
202         D = opendir(path);
203         if (!D)
204                 return 0;
205
206         while ((dir = readdir(D)) != NULL) {
207                 char full_path[256];
208
209                 if (!strcmp(dir->d_name, ".") || !strcmp(dir->d_name, ".."))
210                         continue;
211
212                 sprintf(full_path, "%s/%s", path, dir->d_name);
213
214                 if (!strcmp(dir->d_name, "dev")) {
215                         if (!check_dev_match(dev, full_path)) {
216                                 found = 1;
217                                 break;
218                         }
219                 }
220
221                 if (lstat(full_path, &st) == -1) {
222                         perror("stat");
223                         break;
224                 }
225
226                 if (!S_ISDIR(st.st_mode) || S_ISLNK(st.st_mode))
227                         continue;
228
229                 found = find_block_dir(dev, full_path);
230                 if (found) {
231                         strcpy(path, full_path);
232                         break;
233                 }
234         }
235
236         closedir(D);
237         return found;
238 }
239
240 void init_disk_util(struct thread_data *td)
241 {
242         struct fio_file *f;
243         struct stat st;
244         char foo[PATH_MAX], tmp[PATH_MAX];
245         dev_t dev;
246         char *p;
247
248         if (!td->do_disk_util || (td->io_ops->flags & (FIO_NETIO | FIO_NULLIO)))
249                 return;
250
251         /*
252          * Just use the same file, they are on the same device.
253          */
254         f = &td->files[0];
255         if (!stat(f->file_name, &st)) {
256                 if (S_ISBLK(st.st_mode))
257                         dev = st.st_rdev;
258                 else
259                         dev = st.st_dev;
260         } else {
261                 /*
262                  * must be a file, open "." in that path
263                  */
264                 strncpy(foo, f->file_name, PATH_MAX - 1);
265                 p = dirname(foo);
266                 if (stat(p, &st)) {
267                         perror("disk util stat");
268                         return;
269                 }
270
271                 dev = st.st_dev;
272         }
273
274         if (disk_util_exists(dev))
275                 return;
276                 
277         sprintf(foo, "/sys/block");
278         if (!find_block_dir(dev, foo))
279                 return;
280
281         /*
282          * If there's a ../queue/ directory there, we are inside a partition.
283          * Check if that is the case and jump back. For loop/md/dm etc we
284          * are already in the right spot.
285          */
286         sprintf(tmp, "%s/../queue", foo);
287         if (!stat(tmp, &st)) {
288                 p = dirname(foo);
289                 sprintf(tmp, "%s/queue", p);
290                 if (stat(tmp, &st)) {
291                         log_err("unknown sysfs layout\n");
292                         return;
293                 }
294                 strncpy(tmp, p, PATH_MAX - 1);
295                 sprintf(foo, "%s", tmp);
296         }
297
298         if (td->ioscheduler)
299                 td->sysfs_root = strdup(foo);
300
301         disk_util_add(dev, foo);
302 }
303
304 void disk_util_timer_arm(void)
305 {
306         itimer.it_value.tv_sec = 0;
307         itimer.it_value.tv_usec = DISK_UTIL_MSEC * 1000;
308         setitimer(ITIMER_REAL, &itimer, NULL);
309 }
310
311 void update_rusage_stat(struct thread_data *td)
312 {
313         getrusage(RUSAGE_SELF, &td->ru_end);
314
315         td->usr_time += mtime_since(&td->ru_start.ru_utime, &td->ru_end.ru_utime);
316         td->sys_time += mtime_since(&td->ru_start.ru_stime, &td->ru_end.ru_stime);
317         td->ctx += td->ru_end.ru_nvcsw + td->ru_end.ru_nivcsw - (td->ru_start.ru_nvcsw + td->ru_start.ru_nivcsw);
318         
319         memcpy(&td->ru_start, &td->ru_end, sizeof(td->ru_end));
320 }
321
322 static int calc_lat(struct io_stat *is, unsigned long *min, unsigned long *max,
323                     double *mean, double *dev)
324 {
325         double n = is->samples;
326
327         if (is->samples == 0)
328                 return 0;
329
330         *min = is->min_val;
331         *max = is->max_val;
332
333         n = (double) is->samples;
334         *mean = is->mean;
335
336         if (n > 1.0)
337                 *dev = sqrt(is->S / (n - 1.0));
338         else
339                 *dev = -1.0;
340
341         return 1;
342 }
343
344 static void show_group_stats(struct group_run_stats *rs, int id)
345 {
346         char *p1, *p2, *p3, *p4;
347         const char *ddir_str[] = { "   READ", "  WRITE" };
348         int i;
349
350         fprintf(f_out, "\nRun status group %d (all jobs):\n", id);
351
352         for (i = 0; i <= DDIR_WRITE; i++) {
353                 if (!rs->max_run[i])
354                         continue;
355
356                 p1 = num2str(rs->io_kb[i], 6, 1);
357                 p2 = num2str(rs->agg[i], 6, 1);
358                 p3 = num2str(rs->min_bw[i], 6, 1);
359                 p4 = num2str(rs->max_bw[i], 6, 1);
360
361                 fprintf(f_out, "%s: io=%siB, aggrb=%siB/s, minb=%siB/s, maxb=%siB/s, mint=%llumsec, maxt=%llumsec\n", ddir_str[i], p1, p2, p3, p4, rs->min_run[0], rs->max_run[0]);
362
363                 free(p1);
364                 free(p2);
365                 free(p3);
366                 free(p4);
367         }
368 }
369
370 static void show_disk_util(void)
371 {
372         struct disk_util_stat *dus;
373         struct list_head *entry, *next;
374         struct disk_util *du;
375         double util;
376
377         fprintf(f_out, "\nDisk stats (read/write):\n");
378
379         list_for_each(entry, &disk_list) {
380                 du = list_entry(entry, struct disk_util, list);
381                 dus = &du->dus;
382
383                 util = (double) 100 * du->dus.io_ticks / (double) du->msec;
384                 if (util > 100.0)
385                         util = 100.0;
386
387                 fprintf(f_out, "  %s: ios=%u/%u, merge=%u/%u, ticks=%u/%u, in_queue=%u, util=%3.2f%%\n", du->name, dus->ios[0], dus->ios[1], dus->merges[0], dus->merges[1], dus->ticks[0], dus->ticks[1], dus->time_in_queue, util);
388         }
389
390         /*
391          * now free the list
392          */
393         list_for_each_safe(entry, next, &disk_list) {
394                 list_del(entry);
395                 du = list_entry(entry, struct disk_util, list);
396                 free(du->name);
397                 free(du);
398         }
399 }
400
401 static void show_ddir_status(struct thread_data *td, struct group_run_stats *rs,
402                              int ddir)
403 {
404         const char *ddir_str[] = { "read ", "write" };
405         unsigned long min, max;
406         unsigned long long bw;
407         double mean, dev;
408         char *io_p, *bw_p;
409
410         if (!td->runtime[ddir])
411                 return;
412
413         bw = td->io_bytes[ddir] / td->runtime[ddir];
414         io_p = num2str(td->io_bytes[ddir] >> 10, 6, 1);
415         bw_p = num2str(bw, 6, 1);
416
417         fprintf(f_out, "  %s: io=%siB, bw=%siB/s, runt=%6lumsec\n", ddir_str[ddir], io_p, bw_p, td->runtime[ddir]);
418
419         free(io_p);
420         free(bw_p);
421
422         if (calc_lat(&td->slat_stat[ddir], &min, &max, &mean, &dev))
423                 fprintf(f_out, "    slat (msec): min=%5lu, max=%5lu, avg=%5.02f, stdev=%5.02f\n", min, max, mean, dev);
424
425         if (calc_lat(&td->clat_stat[ddir], &min, &max, &mean, &dev))
426                 fprintf(f_out, "    clat (msec): min=%5lu, max=%5lu, avg=%5.02f, stdev=%5.02f\n", min, max, mean, dev);
427
428         if (calc_lat(&td->bw_stat[ddir], &min, &max, &mean, &dev)) {
429                 double p_of_agg;
430
431                 p_of_agg = mean * 100 / (double) rs->agg[ddir];
432                 fprintf(f_out, "    bw (KiB/s) : min=%5lu, max=%5lu, per=%3.2f%%, avg=%5.02f, stdev=%5.02f\n", min, max, p_of_agg, mean, dev);
433         }
434 }
435
436 static void show_thread_status(struct thread_data *td,
437                                struct group_run_stats *rs)
438 {
439         double usr_cpu, sys_cpu;
440         unsigned long runtime;
441         double io_u_dist[FIO_IO_U_MAP_NR];
442         int i;
443
444         if (!(td->io_bytes[0] + td->io_bytes[1]))
445                 return;
446
447         if (!td->error)
448                 fprintf(f_out, "%s: (groupid=%d): err=%2d: pid=%d\n",td->name, td->groupid, td->error, td->pid);
449         else
450                 fprintf(f_out, "%s: (groupid=%d): err=%2d (%s): pid=%d\n",td->name, td->groupid, td->error, td->verror, td->pid);
451
452         show_ddir_status(td, rs, td->ddir);
453         if (td->io_bytes[td->ddir ^ 1])
454                 show_ddir_status(td, rs, td->ddir ^ 1);
455
456         runtime = mtime_since(&td->epoch, &td->end_time);
457         if (runtime) {
458                 double runt = (double) runtime;
459
460                 usr_cpu = (double) td->usr_time * 100 / runt;
461                 sys_cpu = (double) td->sys_time * 100 / runt;
462         } else {
463                 usr_cpu = 0;
464                 sys_cpu = 0;
465         }
466
467         fprintf(f_out, "  cpu          : usr=%3.2f%%, sys=%3.2f%%, ctx=%lu\n", usr_cpu, sys_cpu, td->ctx);
468
469         /*
470          * Do depth distribution calculations
471          */
472         for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
473                 io_u_dist[i] = (double) td->io_u_map[i] / (double) td->total_io_u;
474                 io_u_dist[i] *= 100.0;
475         }
476
477         fprintf(f_out, "  IO depths    : 1=%3.1f%%, 2=%3.1f%%, 4=%3.1f%%, 8=%3.1f%%, 16=%3.1f%%, 32=%3.1f%%, >=64=%3.1f%%\n", io_u_dist[0], io_u_dist[1], io_u_dist[2], io_u_dist[3], io_u_dist[4], io_u_dist[5], io_u_dist[6]);
478
479         if (td->description)
480                 fprintf(f_out, "%s\n", td->description);
481 }
482
483 static void show_ddir_status_terse(struct thread_data *td,
484                                    struct group_run_stats *rs, int ddir)
485 {
486         unsigned long min, max;
487         unsigned long long bw;
488         double mean, dev;
489
490         bw = 0;
491         if (td->runtime[ddir])
492                 bw = td->io_bytes[ddir] / td->runtime[ddir];
493
494         fprintf(f_out, ",%llu,%llu,%lu", td->io_bytes[ddir] >> 10, bw, td->runtime[ddir]);
495
496         if (calc_lat(&td->slat_stat[ddir], &min, &max, &mean, &dev))
497                 fprintf(f_out, ",%lu,%lu,%f,%f", min, max, mean, dev);
498         else
499                 fprintf(f_out, ",%lu,%lu,%f,%f", 0UL, 0UL, 0.0, 0.0);
500
501         if (calc_lat(&td->clat_stat[ddir], &min, &max, &mean, &dev))
502                 fprintf(f_out, ",%lu,%lu,%f,%f", min, max, mean, dev);
503         else
504                 fprintf(f_out, ",%lu,%lu,%f,%f", 0UL, 0UL, 0.0, 0.0);
505
506         if (calc_lat(&td->bw_stat[ddir], &min, &max, &mean, &dev)) {
507                 double p_of_agg;
508
509                 p_of_agg = mean * 100 / (double) rs->agg[ddir];
510                 fprintf(f_out, ",%lu,%lu,%f%%,%f,%f", min, max, p_of_agg, mean, dev);
511         } else
512                 fprintf(f_out, ",%lu,%lu,%f%%,%f,%f", 0UL, 0UL, 0.0, 0.0, 0.0);
513                 
514 }
515
516
517 static void show_thread_status_terse(struct thread_data *td,
518                                      struct group_run_stats *rs)
519 {
520         double usr_cpu, sys_cpu;
521
522         fprintf(f_out, "%s,%d,%d",td->name, td->groupid, td->error);
523
524         show_ddir_status_terse(td, rs, 0);
525         show_ddir_status_terse(td, rs, 1);
526
527         if (td->runtime[0] + td->runtime[1]) {
528                 double runt = (double) (td->runtime[0] + td->runtime[1]);
529
530                 usr_cpu = (double) td->usr_time * 100 / runt;
531                 sys_cpu = (double) td->sys_time * 100 / runt;
532         } else {
533                 usr_cpu = 0;
534                 sys_cpu = 0;
535         }
536
537         fprintf(f_out, ",%f%%,%f%%,%lu\n", usr_cpu, sys_cpu, td->ctx);
538 }
539
540 void show_run_stats(void)
541 {
542         struct group_run_stats *runstats, *rs;
543         struct thread_data *td;
544         int i;
545
546         runstats = malloc(sizeof(struct group_run_stats) * (groupid + 1));
547
548         for (i = 0; i < groupid + 1; i++) {
549                 rs = &runstats[i];
550
551                 memset(rs, 0, sizeof(*rs));
552                 rs->min_bw[0] = rs->min_run[0] = ~0UL;
553                 rs->min_bw[1] = rs->min_run[1] = ~0UL;
554         }
555
556         for_each_td(td, i) {
557                 unsigned long long rbw, wbw;
558
559                 rs = &runstats[td->groupid];
560
561                 if (td->runtime[0] < rs->min_run[0] || !rs->min_run[0])
562                         rs->min_run[0] = td->runtime[0];
563                 if (td->runtime[0] > rs->max_run[0])
564                         rs->max_run[0] = td->runtime[0];
565                 if (td->runtime[1] < rs->min_run[1] || !rs->min_run[1])
566                         rs->min_run[1] = td->runtime[1];
567                 if (td->runtime[1] > rs->max_run[1])
568                         rs->max_run[1] = td->runtime[1];
569
570                 rbw = wbw = 0;
571                 if (td->runtime[0])
572                         rbw = td->io_bytes[0] / (unsigned long long) td->runtime[0];
573                 if (td->runtime[1])
574                         wbw = td->io_bytes[1] / (unsigned long long) td->runtime[1];
575
576                 if (rbw < rs->min_bw[0])
577                         rs->min_bw[0] = rbw;
578                 if (wbw < rs->min_bw[1])
579                         rs->min_bw[1] = wbw;
580                 if (rbw > rs->max_bw[0])
581                         rs->max_bw[0] = rbw;
582                 if (wbw > rs->max_bw[1])
583                         rs->max_bw[1] = wbw;
584
585                 rs->io_kb[0] += td->io_bytes[0] >> 10;
586                 rs->io_kb[1] += td->io_bytes[1] >> 10;
587         }
588
589         for (i = 0; i < groupid + 1; i++) {
590                 rs = &runstats[i];
591
592                 if (rs->max_run[0])
593                         rs->agg[0] = (rs->io_kb[0]*1024) / rs->max_run[0];
594                 if (rs->max_run[1])
595                         rs->agg[1] = (rs->io_kb[1]*1024) / rs->max_run[1];
596         }
597
598         /*
599          * don't overwrite last signal output
600          */
601         if (!terse_output)
602                 printf("\n");
603
604         for_each_td(td, i) {
605                 rs = &runstats[td->groupid];
606
607                 if (terse_output)
608                         show_thread_status_terse(td, rs);
609                 else
610                         show_thread_status(td, rs);
611         }
612
613         if (!terse_output) {
614                 for (i = 0; i < groupid + 1; i++)
615                         show_group_stats(&runstats[i], i);
616
617                 show_disk_util();
618         }
619
620         free(runstats);
621 }
622
623 static inline void add_stat_sample(struct io_stat *is, unsigned long data)
624 {
625         double val = data;
626         double delta, n;
627
628         if (data > is->max_val)
629                 is->max_val = data;
630         if (data < is->min_val)
631                 is->min_val = data;
632
633         delta = val - is->mean;
634         n = is->samples + 1.0;
635         is->mean += delta / n;
636         is->S += delta * (val - is->mean);
637
638         is->samples++;
639 }
640
641 static void __add_log_sample(struct io_log *iolog, unsigned long val,
642                              enum fio_ddir ddir, unsigned long time)
643 {
644         if (iolog->nr_samples == iolog->max_samples) {
645                 int new_size = sizeof(struct io_sample) * iolog->max_samples*2;
646
647                 iolog->log = realloc(iolog->log, new_size);
648                 iolog->max_samples <<= 1;
649         }
650
651         iolog->log[iolog->nr_samples].val = val;
652         iolog->log[iolog->nr_samples].time = time;
653         iolog->log[iolog->nr_samples].ddir = ddir;
654         iolog->nr_samples++;
655 }
656
657 static void add_log_sample(struct thread_data *td, struct io_log *iolog,
658                            unsigned long val, enum fio_ddir ddir)
659 {
660         __add_log_sample(iolog, val, ddir, mtime_since_now(&td->epoch));
661 }
662
663 void add_agg_sample(unsigned long val, enum fio_ddir ddir)
664 {
665         struct io_log *iolog = agg_io_log[ddir];
666
667         __add_log_sample(iolog, val, ddir, mtime_since_genesis());
668 }
669
670 void add_clat_sample(struct thread_data *td, enum fio_ddir ddir,
671                      unsigned long msec)
672 {
673         add_stat_sample(&td->clat_stat[ddir], msec);
674
675         if (td->clat_log)
676                 add_log_sample(td, td->clat_log, msec, ddir);
677 }
678
679 void add_slat_sample(struct thread_data *td, enum fio_ddir ddir,
680                      unsigned long msec)
681 {
682         add_stat_sample(&td->slat_stat[ddir], msec);
683
684         if (td->slat_log)
685                 add_log_sample(td, td->slat_log, msec, ddir);
686 }
687
688 void add_bw_sample(struct thread_data *td, enum fio_ddir ddir,
689                    struct timeval *t)
690 {
691         unsigned long spent = mtime_since(&td->stat_sample_time[ddir], t);
692         unsigned long rate;
693
694         if (spent < td->bw_avg_time)
695                 return;
696
697         rate = (td->this_io_bytes[ddir] - td->stat_io_bytes[ddir]) / spent;
698         add_stat_sample(&td->bw_stat[ddir], rate);
699
700         if (td->bw_log)
701                 add_log_sample(td, td->bw_log, rate, ddir);
702
703         fio_gettime(&td->stat_sample_time[ddir], NULL);
704         td->stat_io_bytes[ddir] = td->this_io_bytes[ddir];
705 }
706
707