projects / fio.git / blame_incremental
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (non-incremental) | history | HEAD
Properly log errors in server
[fio.git] / stat.c
... / ...
CommitLineData
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#include "diskutil.h"
12
13void update_rusage_stat(struct thread_data *td)
14{
15 struct thread_stat *ts = &td->ts;
16
17 getrusage(RUSAGE_SELF, &ts->ru_end);
18
19 ts->usr_time += mtime_since(&ts->ru_start.ru_utime,
20 &ts->ru_end.ru_utime);
21 ts->sys_time += mtime_since(&ts->ru_start.ru_stime,
22 &ts->ru_end.ru_stime);
23 ts->ctx += ts->ru_end.ru_nvcsw + ts->ru_end.ru_nivcsw
24 - (ts->ru_start.ru_nvcsw + ts->ru_start.ru_nivcsw);
25 ts->minf += ts->ru_end.ru_minflt - ts->ru_start.ru_minflt;
26 ts->majf += ts->ru_end.ru_majflt - ts->ru_start.ru_majflt;
27
28 memcpy(&ts->ru_start, &ts->ru_end, sizeof(ts->ru_end));
29}
30
31/*
32 * Given a latency, return the index of the corresponding bucket in
33 * the structure tracking percentiles.
34 *
35 * (1) find the group (and error bits) that the value (latency)
36 * belongs to by looking at its MSB. (2) find the bucket number in the
37 * group by looking at the index bits.
38 *
39 */
40static unsigned int plat_val_to_idx(unsigned int val)
41{
42 unsigned int msb, error_bits, base, offset, idx;
43
44 /* Find MSB starting from bit 0 */
45 if (val == 0)
46 msb = 0;
47 else
48 msb = (sizeof(val)*8) - __builtin_clz(val) - 1;
49
50 /*
51 * MSB <= (FIO_IO_U_PLAT_BITS-1), cannot be rounded off. Use
52 * all bits of the sample as index
53 */
54 if (msb <= FIO_IO_U_PLAT_BITS)
55 return val;
56
57 /* Compute the number of error bits to discard*/
58 error_bits = msb - FIO_IO_U_PLAT_BITS;
59
60 /* Compute the number of buckets before the group */
61 base = (error_bits + 1) << FIO_IO_U_PLAT_BITS;
62
63 /*
64 * Discard the error bits and apply the mask to find the
65 * index for the buckets in the group
66 */
67 offset = (FIO_IO_U_PLAT_VAL - 1) & (val >> error_bits);
68
69 /* Make sure the index does not exceed (array size - 1) */
70 idx = (base + offset) < (FIO_IO_U_PLAT_NR - 1)?
71 (base + offset) : (FIO_IO_U_PLAT_NR - 1);
72
73 return idx;
74}
75
76/*
77 * Convert the given index of the bucket array to the value
78 * represented by the bucket
79 */
80static unsigned int plat_idx_to_val(unsigned int idx)
81{
82 unsigned int error_bits, k, base;
83
84 assert(idx < FIO_IO_U_PLAT_NR);
85
86 /* MSB <= (FIO_IO_U_PLAT_BITS-1), cannot be rounded off. Use
87 * all bits of the sample as index */
88 if (idx < (FIO_IO_U_PLAT_VAL << 1) )
89 return idx;
90
91 /* Find the group and compute the minimum value of that group */
92 error_bits = (idx >> FIO_IO_U_PLAT_BITS) -1;
93 base = 1 << (error_bits + FIO_IO_U_PLAT_BITS);
94
95 /* Find its bucket number of the group */
96 k = idx % FIO_IO_U_PLAT_VAL;
97
98 /* Return the mean of the range of the bucket */
99 return base + ((k + 0.5) * (1 << error_bits));
100}
101
102static int double_cmp(const void *a, const void *b)
103{
104 const double fa = *(const double *)a;
105 const double fb = *(const double *)b;
106 int cmp = 0;
107
108 if (fa > fb)
109 cmp = 1;
110 else if (fa < fb)
111 cmp = -1;
112
113 return cmp;
114}
115
116/*
117 * Find and display the p-th percentile of clat
118 */
119static void show_clat_percentiles(unsigned int* io_u_plat, unsigned long nr,
120 double* user_list)
121{
122 unsigned long sum = 0;
123 unsigned int len, i, j = 0;
124 const double *plist;
125 int is_last = 0;
126 static const double def_list[FIO_IO_U_LIST_MAX_LEN] = {
127 1.0, 5.0, 10.0, 20.0, 30.0,
128 40.0, 50.0, 60.0, 70.0, 80.0,
129 90.0, 95.0, 99.0, 99.5, 99.9};
130
131 plist = user_list;
132 if (!plist)
133 plist = def_list;
134
135 for (len = 0; len <FIO_IO_U_LIST_MAX_LEN && plist[len] != 0; len++)
136 ;
137
138 /*
139 * Sort the user-specified list. Note that this does not work
140 * for NaN values
141 */
142 if (user_list && len > 1)
143 qsort((void*)user_list, len, sizeof(user_list[0]), double_cmp);
144
145 log_info(" clat percentiles (usec) :");
146
147 for (i = 0; i < FIO_IO_U_PLAT_NR && !is_last; i++) {
148 sum += io_u_plat[i];
149 while (sum >= (plist[j] / 100 * nr)) {
150 assert(plist[j] <= 100.0);
151
152 /* for formatting */
153 if (j != 0 && (j % 4) == 0)
154 log_info(" ");
155
156 /* end of the list */
157 is_last = (j == len - 1);
158
159 log_info(" %2.2fth=%u%c", plist[j], plat_idx_to_val(i),
160 (is_last? '\n' : ','));
161
162 if (is_last)
163 break;
164
165 if (j % 4 == 3) /* for formatting */
166 log_info("\n");
167 j++;
168 }
169 }
170}
171
172static int calc_lat(struct io_stat *is, unsigned long *min, unsigned long *max,
173 double *mean, double *dev)
174{
175 double n = is->samples;
176
177 if (is->samples == 0)
178 return 0;
179
180 *min = is->min_val;
181 *max = is->max_val;
182
183 n = (double) is->samples;
184 *mean = is->mean;
185
186 if (n > 1.0)
187 *dev = sqrt(is->S / (n - 1.0));
188 else
189 *dev = 0;
190
191 return 1;
192}
193
194static void show_group_stats(struct group_run_stats *rs, int id)
195{
196 char *p1, *p2, *p3, *p4;
197 const char *ddir_str[] = { " READ", " WRITE" };
198 int i;
199
200 log_info("\nRun status group %d (all jobs):\n", id);
201
202 for (i = 0; i <= DDIR_WRITE; i++) {
203 const int i2p = is_power_of_2(rs->kb_base);
204
205 if (!rs->max_run[i])
206 continue;
207
208 p1 = num2str(rs->io_kb[i], 6, rs->kb_base, i2p);
209 p2 = num2str(rs->agg[i], 6, rs->kb_base, i2p);
210 p3 = num2str(rs->min_bw[i], 6, rs->kb_base, i2p);
211 p4 = num2str(rs->max_bw[i], 6, rs->kb_base, i2p);
212
213 log_info("%s: io=%sB, aggrb=%sB/s, minb=%sB/s, maxb=%sB/s,"
214 " mint=%llumsec, maxt=%llumsec\n", ddir_str[i], p1, p2,
215 p3, p4, rs->min_run[i],
216 rs->max_run[i]);
217
218 free(p1);
219 free(p2);
220 free(p3);
221 free(p4);
222 }
223}
224
225#define ts_total_io_u(ts) \
226 ((ts)->total_io_u[0] + (ts)->total_io_u[1])
227
228static void stat_calc_dist(unsigned int *map, unsigned long total,
229 double *io_u_dist)
230{
231 int i;
232
233 /*
234 * Do depth distribution calculations
235 */
236 for (i = 0; i < FIO_IO_U_MAP_NR; i++) {
237 if (total) {
238 io_u_dist[i] = (double) map[i] / (double) total;
239 io_u_dist[i] *= 100.0;
240 if (io_u_dist[i] < 0.1 && map[i])
241 io_u_dist[i] = 0.1;
242 } else
243 io_u_dist[i] = 0.0;
244 }
245}
246
247static void stat_calc_lat(struct thread_stat *ts, double *dst,
248 unsigned int *src, int nr)
249{
250 unsigned long total = ts_total_io_u(ts);
251 int i;
252
253 /*
254 * Do latency distribution calculations
255 */
256 for (i = 0; i < nr; i++) {
257 if (total) {
258 dst[i] = (double) src[i] / (double) total;
259 dst[i] *= 100.0;
260 if (dst[i] < 0.01 && src[i])
261 dst[i] = 0.01;
262 } else
263 dst[i] = 0.0;
264 }
265}
266
267static void stat_calc_lat_u(struct thread_stat *ts, double *io_u_lat)
268{
269 stat_calc_lat(ts, io_u_lat, ts->io_u_lat_u, FIO_IO_U_LAT_U_NR);
270}
271
272static void stat_calc_lat_m(struct thread_stat *ts, double *io_u_lat)
273{
274 stat_calc_lat(ts, io_u_lat, ts->io_u_lat_m, FIO_IO_U_LAT_M_NR);
275}
276
277static int usec_to_msec(unsigned long *min, unsigned long *max, double *mean,
278 double *dev)
279{
280 if (*min > 1000 && *max > 1000 && *mean > 1000.0 && *dev > 1000.0) {
281 *min /= 1000;
282 *max /= 1000;
283 *mean /= 1000.0;
284 *dev /= 1000.0;
285 return 0;
286 }
287
288 return 1;
289}
290
291static void show_ddir_status(struct group_run_stats *rs, struct thread_stat *ts,
292 int ddir)
293{
294 const char *ddir_str[] = { "read ", "write" };
295 unsigned long min, max, runt;
296 unsigned long long bw, iops;
297 double mean, dev;
298 char *io_p, *bw_p, *iops_p;
299 int i2p;
300
301 assert(ddir_rw(ddir));
302
303 if (!ts->runtime[ddir])
304 return;
305
306 i2p = is_power_of_2(rs->kb_base);
307 runt = ts->runtime[ddir];
308
309 bw = (1000 * ts->io_bytes[ddir]) / runt;
310 io_p = num2str(ts->io_bytes[ddir], 6, 1, i2p);
311 bw_p = num2str(bw, 6, 1, i2p);
312
313 iops = (1000 * (uint64_t)ts->total_io_u[ddir]) / runt;
314 iops_p = num2str(iops, 6, 1, 0);
315
316 log_info(" %s: io=%sB, bw=%sB/s, iops=%s, runt=%6llumsec\n",
317 ddir_str[ddir], io_p, bw_p, iops_p,
318 ts->runtime[ddir]);
319
320 free(io_p);
321 free(bw_p);
322 free(iops_p);
323
324 if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev)) {
325 const char *base = "(usec)";
326 char *minp, *maxp;
327
328 if (!usec_to_msec(&min, &max, &mean, &dev))
329 base = "(msec)";
330
331 minp = num2str(min, 6, 1, 0);
332 maxp = num2str(max, 6, 1, 0);
333
334 log_info(" slat %s: min=%s, max=%s, avg=%5.02f,"
335 " stdev=%5.02f\n", base, minp, maxp, mean, dev);
336
337 free(minp);
338 free(maxp);
339 }
340 if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev)) {
341 const char *base = "(usec)";
342 char *minp, *maxp;
343
344 if (!usec_to_msec(&min, &max, &mean, &dev))
345 base = "(msec)";
346
347 minp = num2str(min, 6, 1, 0);
348 maxp = num2str(max, 6, 1, 0);
349
350 log_info(" clat %s: min=%s, max=%s, avg=%5.02f,"
351 " stdev=%5.02f\n", base, minp, maxp, mean, dev);
352
353 free(minp);
354 free(maxp);
355 }
356 if (calc_lat(&ts->lat_stat[ddir], &min, &max, &mean, &dev)) {
357 const char *base = "(usec)";
358 char *minp, *maxp;
359
360 if (!usec_to_msec(&min, &max, &mean, &dev))
361 base = "(msec)";
362
363 minp = num2str(min, 6, 1, 0);
364 maxp = num2str(max, 6, 1, 0);
365
366 log_info(" lat %s: min=%s, max=%s, avg=%5.02f,"
367 " stdev=%5.02f\n", base, minp, maxp, mean, dev);
368
369 free(minp);
370 free(maxp);
371 }
372 if (ts->clat_percentiles) {
373 show_clat_percentiles(ts->io_u_plat[ddir],
374 ts->clat_stat[ddir].samples,
375 ts->percentile_list);
376 }
377 if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) {
378 double p_of_agg;
379
380 p_of_agg = mean * 100 / (double) rs->agg[ddir];
381 log_info(" bw (KB/s) : min=%5lu, max=%5lu, per=%3.2f%%,"
382 " avg=%5.02f, stdev=%5.02f\n", min, max, p_of_agg,
383 mean, dev);
384 }
385}
386
387static void show_lat(double *io_u_lat, int nr, const char **ranges,
388 const char *msg)
389{
390 int new_line = 1, i, line = 0;
391
392 for (i = 0; i < nr; i++) {
393 if (io_u_lat[i] <= 0.0)
394 continue;
395 if (new_line) {
396 if (line)
397 log_info("\n");
398 log_info(" lat (%s): ", msg);
399 new_line = 0;
400 line = 0;
401 }
402 if (line)
403 log_info(", ");
404 log_info("%s%3.2f%%", ranges[i], io_u_lat[i]);
405 line++;
406 if (line == 5)
407 new_line = 1;
408 }
409}
410
411static void show_lat_u(double *io_u_lat_u)
412{
413 const char *ranges[] = { "2=", "4=", "10=", "20=", "50=", "100=",
414 "250=", "500=", "750=", "1000=", };
415
416 show_lat(io_u_lat_u, FIO_IO_U_LAT_U_NR, ranges, "usec");
417}
418
419static void show_lat_m(double *io_u_lat_m)
420{
421 const char *ranges[] = { "2=", "4=", "10=", "20=", "50=", "100=",
422 "250=", "500=", "750=", "1000=", "2000=",
423 ">=2000=", };
424
425 show_lat(io_u_lat_m, FIO_IO_U_LAT_M_NR, ranges, "msec");
426}
427
428static void show_latencies(double *io_u_lat_u, double *io_u_lat_m)
429{
430 show_lat_u(io_u_lat_u);
431 log_info("\n");
432 show_lat_m(io_u_lat_m);
433 log_info("\n");
434}
435
436static void show_thread_status(struct thread_stat *ts,
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 double io_u_lat_u[FIO_IO_U_LAT_U_NR];
443 double io_u_lat_m[FIO_IO_U_LAT_M_NR];
444
445 if (!(ts->io_bytes[0] + ts->io_bytes[1]) &&
446 !(ts->total_io_u[0] + ts->total_io_u[1]))
447 return;
448
449 if (!ts->error) {
450 log_info("%s: (groupid=%d, jobs=%d): err=%2d: pid=%d\n",
451 ts->name, ts->groupid, ts->members,
452 ts->error, (int) ts->pid);
453 } else {
454 log_info("%s: (groupid=%d, jobs=%d): err=%2d (%s): pid=%d\n",
455 ts->name, ts->groupid, ts->members,
456 ts->error, ts->verror, (int) ts->pid);
457 }
458
459 if (ts->description)
460 log_info(" Description : [%s]\n", ts->description);
461
462 if (ts->io_bytes[DDIR_READ])
463 show_ddir_status(rs, ts, DDIR_READ);
464 if (ts->io_bytes[DDIR_WRITE])
465 show_ddir_status(rs, ts, DDIR_WRITE);
466
467 runtime = ts->total_run_time;
468 if (runtime) {
469 double runt = (double) runtime;
470
471 usr_cpu = (double) ts->usr_time * 100 / runt;
472 sys_cpu = (double) ts->sys_time * 100 / runt;
473 } else {
474 usr_cpu = 0;
475 sys_cpu = 0;
476 }
477
478 log_info(" cpu : usr=%3.2f%%, sys=%3.2f%%, ctx=%lu, majf=%lu,"
479 " minf=%lu\n", usr_cpu, sys_cpu, ts->ctx, ts->majf, ts->minf);
480
481 stat_calc_dist(ts->io_u_map, ts_total_io_u(ts), io_u_dist);
482 log_info(" IO depths : 1=%3.1f%%, 2=%3.1f%%, 4=%3.1f%%, 8=%3.1f%%,"
483 " 16=%3.1f%%, 32=%3.1f%%, >=64=%3.1f%%\n", io_u_dist[0],
484 io_u_dist[1], io_u_dist[2],
485 io_u_dist[3], io_u_dist[4],
486 io_u_dist[5], io_u_dist[6]);
487
488 stat_calc_dist(ts->io_u_submit, ts->total_submit, io_u_dist);
489 log_info(" submit : 0=%3.1f%%, 4=%3.1f%%, 8=%3.1f%%, 16=%3.1f%%,"
490 " 32=%3.1f%%, 64=%3.1f%%, >=64=%3.1f%%\n", io_u_dist[0],
491 io_u_dist[1], io_u_dist[2],
492 io_u_dist[3], io_u_dist[4],
493 io_u_dist[5], io_u_dist[6]);
494 stat_calc_dist(ts->io_u_complete, ts->total_complete, io_u_dist);
495 log_info(" complete : 0=%3.1f%%, 4=%3.1f%%, 8=%3.1f%%, 16=%3.1f%%,"
496 " 32=%3.1f%%, 64=%3.1f%%, >=64=%3.1f%%\n", io_u_dist[0],
497 io_u_dist[1], io_u_dist[2],
498 io_u_dist[3], io_u_dist[4],
499 io_u_dist[5], io_u_dist[6]);
500 log_info(" issued r/w/d: total=%lu/%lu/%lu, short=%lu/%lu/%lu\n",
501 ts->total_io_u[0], ts->total_io_u[1],
502 ts->total_io_u[2],
503 ts->short_io_u[0], ts->short_io_u[1],
504 ts->short_io_u[2]);
505 stat_calc_lat_u(ts, io_u_lat_u);
506 stat_calc_lat_m(ts, io_u_lat_m);
507 show_latencies(io_u_lat_u, io_u_lat_m);
508 if (ts->continue_on_error) {
509 log_info(" errors : total=%lu, first_error=%d/<%s>\n",
510 ts->total_err_count,
511 ts->first_error,
512 strerror(ts->first_error));
513 }
514}
515
516static void show_ddir_status_terse(struct thread_stat *ts,
517 struct group_run_stats *rs, int ddir)
518{
519 unsigned long min, max;
520 unsigned long long bw;
521 double mean, dev;
522
523 assert(ddir_rw(ddir));
524
525 bw = 0;
526 if (ts->runtime[ddir])
527 bw = ts->io_bytes[ddir] / ts->runtime[ddir];
528
529 log_info(";%llu;%llu;%llu", ts->io_bytes[ddir] >> 10, bw,
530 ts->runtime[ddir]);
531
532 if (calc_lat(&ts->slat_stat[ddir], &min, &max, &mean, &dev))
533 log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
534 else
535 log_info(";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0);
536
537 if (calc_lat(&ts->clat_stat[ddir], &min, &max, &mean, &dev))
538 log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
539 else
540 log_info(";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0);
541
542 if (calc_lat(&ts->lat_stat[ddir], &min, &max, &mean, &dev))
543 log_info(";%lu;%lu;%f;%f", min, max, mean, dev);
544 else
545 log_info(";%lu;%lu;%f;%f", 0UL, 0UL, 0.0, 0.0);
546
547 if (calc_lat(&ts->bw_stat[ddir], &min, &max, &mean, &dev)) {
548 double p_of_agg;
549
550 p_of_agg = mean * 100 / (double) rs->agg[ddir];
551 log_info(";%lu;%lu;%f%%;%f;%f", min, max, p_of_agg, mean, dev);
552 } else
553 log_info(";%lu;%lu;%f%%;%f;%f", 0UL, 0UL, 0.0, 0.0, 0.0);
554}
555
556#define FIO_TERSE_VERSION "2"
557
558static void show_thread_status_terse(struct thread_stat *ts,
559 struct group_run_stats *rs)
560{
561 double io_u_dist[FIO_IO_U_MAP_NR];
562 double io_u_lat_u[FIO_IO_U_LAT_U_NR];
563 double io_u_lat_m[FIO_IO_U_LAT_M_NR];
564 double usr_cpu, sys_cpu;
565 int i;
566
567 /* General Info */
568 log_info("%s;%s;%d;%d", FIO_TERSE_VERSION, ts->name, ts->groupid,
569 ts->error);
570 /* Log Read Status */
571 show_ddir_status_terse(ts, rs, 0);
572 /* Log Write Status */
573 show_ddir_status_terse(ts, rs, 1);
574
575 /* CPU Usage */
576 if (ts->total_run_time) {
577 double runt = (double) ts->total_run_time;
578
579 usr_cpu = (double) ts->usr_time * 100 / runt;
580 sys_cpu = (double) ts->sys_time * 100 / runt;
581 } else {
582 usr_cpu = 0;
583 sys_cpu = 0;
584 }
585
586 log_info(";%f%%;%f%%;%lu;%lu;%lu", usr_cpu, sys_cpu, ts->ctx, ts->majf,
587 ts->minf);
588
589 /* Calc % distribution of IO depths, usecond, msecond latency */
590 stat_calc_dist(ts->io_u_map, ts_total_io_u(ts), io_u_dist);
591 stat_calc_lat_u(ts, io_u_lat_u);
592 stat_calc_lat_m(ts, io_u_lat_m);
593
594 /* Only show fixed 7 I/O depth levels*/
595 log_info(";%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%;%3.1f%%",
596 io_u_dist[0], io_u_dist[1], io_u_dist[2], io_u_dist[3],
597 io_u_dist[4], io_u_dist[5], io_u_dist[6]);
598
599 /* Microsecond latency */
600 for (i = 0; i < FIO_IO_U_LAT_U_NR; i++)
601 log_info(";%3.2f%%", io_u_lat_u[i]);
602 /* Millisecond latency */
603 for (i = 0; i < FIO_IO_U_LAT_M_NR; i++)
604 log_info(";%3.2f%%", io_u_lat_m[i]);
605 /* Additional output if continue_on_error set - default off*/
606 if (ts->continue_on_error)
607 log_info(";%lu;%d", ts->total_err_count, ts->first_error);
608 log_info("\n");
609
610 /* Additional output if description is set */
611 if (ts->description)
612 log_info(";%s", ts->description);
613
614 log_info("\n");
615}
616
617static void sum_stat(struct io_stat *dst, struct io_stat *src, int nr)
618{
619 double mean, S;
620
621 if (src->samples == 0)
622 return;
623
624 dst->min_val = min(dst->min_val, src->min_val);
625 dst->max_val = max(dst->max_val, src->max_val);
626
627 /*
628 * Compute new mean and S after the merge
629 * <http://en.wikipedia.org/wiki/Algorithms_for_calculating_variance
630 * #Parallel_algorithm>
631 */
632 if (nr == 1) {
633 mean = src->mean;
634 S = src->S;
635 } else {
636 double delta = src->mean - dst->mean;
637
638 mean = ((src->mean * src->samples) +
639 (dst->mean * dst->samples)) /
640 (dst->samples + src->samples);
641
642 S = src->S + dst->S + pow(delta, 2.0) *
643 (dst->samples * src->samples) /
644 (dst->samples + src->samples);
645 }
646
647 dst->samples += src->samples;
648 dst->mean = mean;
649 dst->S = S;
650}
651
652void show_run_stats(void)
653{
654 struct group_run_stats *runstats, *rs;
655 struct thread_data *td;
656 struct thread_stat *threadstats, *ts;
657 int i, j, k, l, nr_ts, last_ts, idx;
658 int kb_base_warned = 0;
659
660 runstats = malloc(sizeof(struct group_run_stats) * (groupid + 1));
661
662 for (i = 0; i < groupid + 1; i++) {
663 rs = &runstats[i];
664
665 memset(rs, 0, sizeof(*rs));
666 rs->min_bw[0] = rs->min_run[0] = ~0UL;
667 rs->min_bw[1] = rs->min_run[1] = ~0UL;
668 }
669
670 /*
671 * find out how many threads stats we need. if group reporting isn't
672 * enabled, it's one-per-td.
673 */
674 nr_ts = 0;
675 last_ts = -1;
676 for_each_td(td, i) {
677 if (!td->o.group_reporting) {
678 nr_ts++;
679 continue;
680 }
681 if (last_ts == td->groupid)
682 continue;
683
684 last_ts = td->groupid;
685 nr_ts++;
686 }
687
688 threadstats = malloc(nr_ts * sizeof(struct thread_stat));
689
690 for (i = 0; i < nr_ts; i++) {
691 ts = &threadstats[i];
692
693 memset(ts, 0, sizeof(*ts));
694 for (j = 0; j <= DDIR_WRITE; j++) {
695 ts->lat_stat[j].min_val = -1UL;
696 ts->clat_stat[j].min_val = -1UL;
697 ts->slat_stat[j].min_val = -1UL;
698 ts->bw_stat[j].min_val = -1UL;
699 }
700 ts->groupid = -1;
701 }
702
703 j = 0;
704 last_ts = -1;
705 idx = 0;
706 for_each_td(td, i) {
707 if (idx && (!td->o.group_reporting ||
708 (td->o.group_reporting && last_ts != td->groupid))) {
709 idx = 0;
710 j++;
711 }
712
713 last_ts = td->groupid;
714
715 ts = &threadstats[j];
716
717 ts->clat_percentiles = td->o.clat_percentiles;
718 if (td->o.overwrite_plist)
719 ts->percentile_list = td->o.percentile_list;
720 else
721 ts->percentile_list = NULL;
722
723 idx++;
724 ts->members++;
725
726 if (ts->groupid == -1) {
727 /*
728 * These are per-group shared already
729 */
730 ts->name = td->o.name;
731 ts->description = td->o.description;
732 ts->groupid = td->groupid;
733
734 /*
735 * first pid in group, not very useful...
736 */
737 ts->pid = td->pid;
738
739 ts->kb_base = td->o.kb_base;
740 } else if (ts->kb_base != td->o.kb_base && !kb_base_warned) {
741 log_info("fio: kb_base differs for jobs in group, using"
742 " %u as the base\n", ts->kb_base);
743 kb_base_warned = 1;
744 }
745
746 ts->continue_on_error = td->o.continue_on_error;
747 ts->total_err_count += td->total_err_count;
748 ts->first_error = td->first_error;
749 if (!ts->error) {
750 if (!td->error && td->o.continue_on_error &&
751 td->first_error) {
752 ts->error = td->first_error;
753 ts->verror = td->verror;
754 } else if (td->error) {
755 ts->error = td->error;
756 ts->verror = td->verror;
757 }
758 }
759
760 for (l = 0; l <= DDIR_WRITE; l++) {
761 sum_stat(&ts->clat_stat[l], &td->ts.clat_stat[l], idx);
762 sum_stat(&ts->slat_stat[l], &td->ts.slat_stat[l], idx);
763 sum_stat(&ts->lat_stat[l], &td->ts.lat_stat[l], idx);
764 sum_stat(&ts->bw_stat[l], &td->ts.bw_stat[l], idx);
765
766 ts->stat_io_bytes[l] += td->ts.stat_io_bytes[l];
767 ts->io_bytes[l] += td->ts.io_bytes[l];
768
769 if (ts->runtime[l] < td->ts.runtime[l])
770 ts->runtime[l] = td->ts.runtime[l];
771 }
772
773 ts->usr_time += td->ts.usr_time;
774 ts->sys_time += td->ts.sys_time;
775 ts->ctx += td->ts.ctx;
776 ts->majf += td->ts.majf;
777 ts->minf += td->ts.minf;
778
779 for (k = 0; k < FIO_IO_U_MAP_NR; k++)
780 ts->io_u_map[k] += td->ts.io_u_map[k];
781 for (k = 0; k < FIO_IO_U_MAP_NR; k++)
782 ts->io_u_submit[k] += td->ts.io_u_submit[k];
783 for (k = 0; k < FIO_IO_U_MAP_NR; k++)
784 ts->io_u_complete[k] += td->ts.io_u_complete[k];
785 for (k = 0; k < FIO_IO_U_LAT_U_NR; k++)
786 ts->io_u_lat_u[k] += td->ts.io_u_lat_u[k];
787 for (k = 0; k < FIO_IO_U_LAT_M_NR; k++)
788 ts->io_u_lat_m[k] += td->ts.io_u_lat_m[k];
789
790
791 for (k = 0; k <= 2; k++) {
792 ts->total_io_u[k] += td->ts.total_io_u[k];
793 ts->short_io_u[k] += td->ts.short_io_u[k];
794 }
795
796 for (k = 0; k <= DDIR_WRITE; k++) {
797 int m;
798 for (m = 0; m < FIO_IO_U_PLAT_NR; m++)
799 ts->io_u_plat[k][m] += td->ts.io_u_plat[k][m];
800 }
801
802 ts->total_run_time += td->ts.total_run_time;
803 ts->total_submit += td->ts.total_submit;
804 ts->total_complete += td->ts.total_complete;
805 }
806
807 for (i = 0; i < nr_ts; i++) {
808 unsigned long long bw;
809
810 ts = &threadstats[i];
811 rs = &runstats[ts->groupid];
812 rs->kb_base = ts->kb_base;
813
814 for (j = 0; j <= DDIR_WRITE; j++) {
815 if (!ts->runtime[j])
816 continue;
817 if (ts->runtime[j] < rs->min_run[j] || !rs->min_run[j])
818 rs->min_run[j] = ts->runtime[j];
819 if (ts->runtime[j] > rs->max_run[j])
820 rs->max_run[j] = ts->runtime[j];
821
822 bw = 0;
823 if (ts->runtime[j]) {
824 unsigned long runt;
825
826 runt = ts->runtime[j];
827 bw = ts->io_bytes[j] / runt;
828 }
829 if (bw < rs->min_bw[j])
830 rs->min_bw[j] = bw;
831 if (bw > rs->max_bw[j])
832 rs->max_bw[j] = bw;
833
834 rs->io_kb[j] += ts->io_bytes[j] / rs->kb_base;
835 }
836 }
837
838 for (i = 0; i < groupid + 1; i++) {
839 unsigned long max_run[2];
840
841 rs = &runstats[i];
842 max_run[0] = rs->max_run[0];
843 max_run[1] = rs->max_run[1];
844
845 if (rs->max_run[0])
846 rs->agg[0] = (rs->io_kb[0] * 1000) / max_run[0];
847 if (rs->max_run[1])
848 rs->agg[1] = (rs->io_kb[1] * 1000) / max_run[1];
849 }
850
851 /*
852 * don't overwrite last signal output
853 */
854 if (!terse_output)
855 log_info("\n");
856
857 for (i = 0; i < nr_ts; i++) {
858 ts = &threadstats[i];
859 rs = &runstats[ts->groupid];
860
861 if (terse_output)
862 show_thread_status_terse(ts, rs);
863 else
864 show_thread_status(ts, rs);
865 }
866
867 if (!terse_output) {
868 for (i = 0; i < groupid + 1; i++)
869 show_group_stats(&runstats[i], i);
870
871 show_disk_util();
872 }
873
874 free(runstats);
875 free(threadstats);
876}
877
878static inline void add_stat_sample(struct io_stat *is, unsigned long data)
879{
880 double val = data;
881 double delta;
882
883 if (data > is->max_val)
884 is->max_val = data;
885 if (data < is->min_val)
886 is->min_val = data;
887
888 delta = val - is->mean;
889 if (delta) {
890 is->mean += delta / (is->samples + 1.0);
891 is->S += delta * (val - is->mean);
892 }
893
894 is->samples++;
895}
896
897static void __add_log_sample(struct io_log *iolog, unsigned long val,
898 enum fio_ddir ddir, unsigned int bs,
899 unsigned long t)
900{
901 const int nr_samples = iolog->nr_samples;
902
903 if (iolog->nr_samples == iolog->max_samples) {
904 int new_size = sizeof(struct io_sample) * iolog->max_samples*2;
905
906 iolog->log = realloc(iolog->log, new_size);
907 iolog->max_samples <<= 1;
908 }
909
910 iolog->log[nr_samples].val = val;
911 iolog->log[nr_samples].time = t;
912 iolog->log[nr_samples].ddir = ddir;
913 iolog->log[nr_samples].bs = bs;
914 iolog->nr_samples++;
915}
916
917static void add_log_sample(struct thread_data *td, struct io_log *iolog,
918 unsigned long val, enum fio_ddir ddir,
919 unsigned int bs)
920{
921 if (!ddir_rw(ddir))
922 return;
923
924 __add_log_sample(iolog, val, ddir, bs, mtime_since_now(&td->epoch));
925}
926
927void add_agg_sample(unsigned long val, enum fio_ddir ddir, unsigned int bs)
928{
929 struct io_log *iolog;
930
931 if (!ddir_rw(ddir))
932 return;
933
934 iolog = agg_io_log[ddir];
935 __add_log_sample(iolog, val, ddir, bs, mtime_since_genesis());
936}
937
938static void add_clat_percentile_sample(struct thread_stat *ts,
939 unsigned long usec, enum fio_ddir ddir)
940{
941 unsigned int idx = plat_val_to_idx(usec);
942 assert(idx < FIO_IO_U_PLAT_NR);
943
944 ts->io_u_plat[ddir][idx]++;
945}
946
947void add_clat_sample(struct thread_data *td, enum fio_ddir ddir,
948 unsigned long usec, unsigned int bs)
949{
950 struct thread_stat *ts = &td->ts;
951
952 if (!ddir_rw(ddir))
953 return;
954
955 add_stat_sample(&ts->clat_stat[ddir], usec);
956
957 if (ts->clat_log)
958 add_log_sample(td, ts->clat_log, usec, ddir, bs);
959
960 if (ts->clat_percentiles)
961 add_clat_percentile_sample(ts, usec, ddir);
962}
963
964void add_slat_sample(struct thread_data *td, enum fio_ddir ddir,
965 unsigned long usec, unsigned int bs)
966{
967 struct thread_stat *ts = &td->ts;
968
969 if (!ddir_rw(ddir))
970 return;
971
972 add_stat_sample(&ts->slat_stat[ddir], usec);
973
974 if (ts->slat_log)
975 add_log_sample(td, ts->slat_log, usec, ddir, bs);
976}
977
978void add_lat_sample(struct thread_data *td, enum fio_ddir ddir,
979 unsigned long usec, unsigned int bs)
980{
981 struct thread_stat *ts = &td->ts;
982
983 if (!ddir_rw(ddir))
984 return;
985
986 add_stat_sample(&ts->lat_stat[ddir], usec);
987
988 if (ts->lat_log)
989 add_log_sample(td, ts->lat_log, usec, ddir, bs);
990}
991
992void add_bw_sample(struct thread_data *td, enum fio_ddir ddir, unsigned int bs,
993 struct timeval *t)
994{
995 struct thread_stat *ts = &td->ts;
996 unsigned long spent, rate;
997
998 if (!ddir_rw(ddir))
999 return;
1000
1001 spent = mtime_since(&ts->stat_sample_time[ddir], t);
1002 if (spent < td->o.bw_avg_time)
1003 return;
1004
1005 rate = (td->this_io_bytes[ddir] - ts->stat_io_bytes[ddir]) *
1006 1000 / spent / 1024;
1007 add_stat_sample(&ts->bw_stat[ddir], rate);
1008
1009 if (ts->bw_log)
1010 add_log_sample(td, ts->bw_log, rate, ddir, bs);
1011
1012 fio_gettime(&ts->stat_sample_time[ddir], NULL);
1013 ts->stat_io_bytes[ddir] = td->this_io_bytes[ddir];
1014}
fio - Flexible IO Tester