Collect and show zone reset statistics
[fio.git] / stat.h
CommitLineData
a64e88da
JA
1#ifndef FIO_STAT_H
2#define FIO_STAT_H
3
ec41265e 4#include "iolog.h"
a666cab8 5#include "lib/output_buffer.h"
ec41265e 6
a64e88da 7struct group_run_stats {
6eaf09d6
SL
8 uint64_t max_run[DDIR_RWDIR_CNT], min_run[DDIR_RWDIR_CNT];
9 uint64_t max_bw[DDIR_RWDIR_CNT], min_bw[DDIR_RWDIR_CNT];
af7f87cb 10 uint64_t iobytes[DDIR_RWDIR_CNT];
6eaf09d6 11 uint64_t agg[DDIR_RWDIR_CNT];
a64e88da 12 uint32_t kb_base;
ad705bcb 13 uint32_t unit_base;
e883cb35 14 uint32_t sig_figs;
a64e88da 15 uint32_t groupid;
771e58be 16 uint32_t unified_rw_rep;
e5021b34 17} __attribute__((packed));
a64e88da
JA
18
19/*
20 * How many depth levels to log
21 */
22#define FIO_IO_U_MAP_NR 7
d6bb626e 23#define FIO_IO_U_LAT_N_NR 10
a64e88da
JA
24#define FIO_IO_U_LAT_U_NR 10
25#define FIO_IO_U_LAT_M_NR 12
26
37b08652
VF
27/*
28 * Constants for clat percentiles
29 */
30#define FIO_IO_U_PLAT_BITS 6
31#define FIO_IO_U_PLAT_VAL (1 << FIO_IO_U_PLAT_BITS)
32#define FIO_IO_U_PLAT_GROUP_NR 29
33#define FIO_IO_U_PLAT_NR (FIO_IO_U_PLAT_GROUP_NR * FIO_IO_U_PLAT_VAL)
34#define FIO_IO_U_LIST_MAX_LEN 20 /* The size of the default and user-specified
35 list of percentiles */
36
a64e88da
JA
37/*
38 * Aggregate clat samples to report percentile(s) of them.
39 *
40 * EXECUTIVE SUMMARY
41 *
42 * FIO_IO_U_PLAT_BITS determines the maximum statistical error on the
43 * value of resulting percentiles. The error will be approximately
44 * 1/2^(FIO_IO_U_PLAT_BITS+1) of the value.
45 *
46 * FIO_IO_U_PLAT_GROUP_NR and FIO_IO_U_PLAT_BITS determine the maximum
47 * range being tracked for latency samples. The maximum value tracked
37b08652 48 * accurately will be 2^(GROUP_NR + PLAT_BITS - 1) nanoseconds.
a64e88da
JA
49 *
50 * FIO_IO_U_PLAT_GROUP_NR and FIO_IO_U_PLAT_BITS determine the memory
51 * requirement of storing those aggregate counts. The memory used will
52 * be (FIO_IO_U_PLAT_GROUP_NR * 2^FIO_IO_U_PLAT_BITS) * sizeof(int)
53 * bytes.
54 *
55 * FIO_IO_U_PLAT_NR is the total number of buckets.
56 *
57 * DETAILS
58 *
59 * Suppose the clat varies from 0 to 999 (usec), the straightforward
60 * method is to keep an array of (999 + 1) buckets, in which a counter
61 * keeps the count of samples which fall in the bucket, e.g.,
62 * {[0],[1],...,[999]}. However this consumes a huge amount of space,
63 * and can be avoided if an approximation is acceptable.
64 *
65 * One such method is to let the range of the bucket to be greater
66 * than one. This method has low accuracy when the value is small. For
67 * example, let the buckets be {[0,99],[100,199],...,[900,999]}, and
68 * the represented value of each bucket be the mean of the range. Then
69 * a value 0 has an round-off error of 49.5. To improve on this, we
70 * use buckets with non-uniform ranges, while bounding the error of
71 * each bucket within a ratio of the sample value. A simple example
72 * would be when error_bound = 0.005, buckets are {
73 * {[0],[1],...,[99]}, {[100,101],[102,103],...,[198,199]},..,
74 * {[900,909],[910,919]...} }. The total range is partitioned into
75 * groups with different ranges, then buckets with uniform ranges. An
76 * upper bound of the error is (range_of_bucket/2)/value_of_bucket
77 *
78 * For better efficiency, we implement this using base two. We group
79 * samples by their Most Significant Bit (MSB), extract the next M bit
80 * of them as an index within the group, and discard the rest of the
81 * bits.
82 *
83 * E.g., assume a sample 'x' whose MSB is bit n (starting from bit 0),
84 * and use M bit for indexing
85 *
86 * | n | M bits | bit (n-M-1) ... bit 0 |
87 *
88 * Because x is at least 2^n, and bit 0 to bit (n-M-1) is at most
89 * (2^(n-M) - 1), discarding bit 0 to (n-M-1) makes the round-off
90 * error
91 *
92 * 2^(n-M)-1 2^(n-M) 1
93 * e <= --------- <= ------- = ---
94 * 2^n 2^n 2^M
95 *
96 * Furthermore, we use "mean" of the range to represent the bucket,
97 * the error e can be lowered by half to 1 / 2^(M+1). By using M bits
98 * as the index, each group must contains 2^M buckets.
99 *
100 * E.g. Let M (FIO_IO_U_PLAT_BITS) be 6
101 * Error bound is 1/2^(6+1) = 0.0078125 (< 1%)
102 *
103 * Group MSB #discarded range of #buckets
104 * error_bits value
105 * ----------------------------------------------------------------
106 * 0* 0~5 0 [0,63] 64
107 * 1* 6 0 [64,127] 64
108 * 2 7 1 [128,255] 64
109 * 3 8 2 [256,511] 64
110 * 4 9 3 [512,1023] 64
111 * ... ... ... [...,...] ...
37b08652 112 * 28 33 27 [8589934592,+inf]** 64
a64e88da
JA
113 *
114 * * Special cases: when n < (M-1) or when n == (M-1), in both cases,
115 * the value cannot be rounded off. Use all bits of the sample as
116 * index.
117 *
37b08652 118 * ** If a sample's MSB is greater than 33, it will be counted as 33.
a64e88da
JA
119 */
120
66347cfa
DE
121/*
122 * Trim cycle count measurements
123 */
124#define MAX_NR_BLOCK_INFOS 8192
125#define BLOCK_INFO_STATE_SHIFT 29
126#define BLOCK_INFO_TRIMS(block_info) \
127 ((block_info) & ((1 << BLOCK_INFO_STATE_SHIFT) - 1))
128#define BLOCK_INFO_STATE(block_info) \
129 ((block_info) >> BLOCK_INFO_STATE_SHIFT)
130#define BLOCK_INFO(state, trim_cycles) \
fc8d6d05 131 ((trim_cycles) | ((unsigned int) (state) << BLOCK_INFO_STATE_SHIFT))
66347cfa
DE
132#define BLOCK_INFO_SET_STATE(block_info, state) \
133 BLOCK_INFO(state, BLOCK_INFO_TRIMS(block_info))
134enum block_info_state {
135 BLOCK_STATE_UNINIT,
136 BLOCK_STATE_TRIMMED,
137 BLOCK_STATE_WRITTEN,
138 BLOCK_STATE_TRIM_FAILURE,
139 BLOCK_STATE_WRITE_FAILURE,
140 BLOCK_STATE_COUNT,
1d6d3455 141};
66347cfa 142
a64e88da
JA
143#define MAX_PATTERN_SIZE 512
144#define FIO_JOBNAME_SIZE 128
4e59d0f3 145#define FIO_JOBDESC_SIZE 256
a64e88da
JA
146#define FIO_VERROR_SIZE 128
147
148struct thread_stat {
149 char name[FIO_JOBNAME_SIZE];
150 char verror[FIO_VERROR_SIZE];
ddcc0b69 151 uint32_t error;
2f122b13 152 uint32_t thread_number;
ddcc0b69 153 uint32_t groupid;
a64e88da 154 uint32_t pid;
4e59d0f3 155 char description[FIO_JOBDESC_SIZE];
a64e88da 156 uint32_t members;
771e58be 157 uint32_t unified_rw_rep;
a64e88da
JA
158
159 /*
160 * bandwidth and latency stats
161 */
1e900e2d 162 struct io_stat sync_stat __attribute__((aligned(8)));/* fsync etc stats */
6eaf09d6
SL
163 struct io_stat clat_stat[DDIR_RWDIR_CNT]; /* completion latency */
164 struct io_stat slat_stat[DDIR_RWDIR_CNT]; /* submission latency */
165 struct io_stat lat_stat[DDIR_RWDIR_CNT]; /* total latency */
166 struct io_stat bw_stat[DDIR_RWDIR_CNT]; /* bandwidth stats */
167 struct io_stat iops_stat[DDIR_RWDIR_CNT]; /* IOPS stats */
a64e88da
JA
168
169 /*
170 * fio system usage accounting
171 */
172 uint64_t usr_time;
173 uint64_t sys_time;
174 uint64_t ctx;
175 uint64_t minf, majf;
176
177 /*
178 * IO depth and latency stats
179 */
b599759b
JA
180 uint32_t clat_percentiles;
181 uint32_t lat_percentiles;
435d195a 182 uint64_t percentile_precision;
802ad4a8 183 fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];
a64e88da 184
6cc0e5aa
AL
185 uint64_t io_u_map[FIO_IO_U_MAP_NR];
186 uint64_t io_u_submit[FIO_IO_U_MAP_NR];
187 uint64_t io_u_complete[FIO_IO_U_MAP_NR];
188 uint64_t io_u_lat_n[FIO_IO_U_LAT_N_NR];
189 uint64_t io_u_lat_u[FIO_IO_U_LAT_U_NR];
190 uint64_t io_u_lat_m[FIO_IO_U_LAT_M_NR];
191 uint64_t io_u_plat[DDIR_RWDIR_CNT][FIO_IO_U_PLAT_NR];
192 uint64_t io_u_sync_plat[FIO_IO_U_PLAT_NR];
3d0ebb30 193
7f3ecee2 194 uint64_t total_io_u[DDIR_RWDIR_SYNC_CNT];
de5cdfea
JA
195 uint64_t short_io_u[DDIR_RWDIR_CNT];
196 uint64_t drop_io_u[DDIR_RWDIR_CNT];
a64e88da
JA
197 uint64_t total_submit;
198 uint64_t total_complete;
199
6eaf09d6
SL
200 uint64_t io_bytes[DDIR_RWDIR_CNT];
201 uint64_t runtime[DDIR_RWDIR_CNT];
a64e88da
JA
202 uint64_t total_run_time;
203
204 /*
205 * IO Error related stats
206 */
3d0ebb30
GG
207 union {
208 uint16_t continue_on_error;
11e955e3 209 uint32_t pad2;
3d0ebb30 210 };
ddcc0b69 211 uint32_t first_error;
11e955e3 212 uint64_t total_err_count;
a64e88da 213
fd5d733f
BVA
214 /* ZBD stats */
215 uint64_t nr_zone_resets;
216
66347cfa
DE
217 uint64_t nr_block_infos;
218 uint32_t block_infos[MAX_NR_BLOCK_INFOS];
219
a64e88da 220 uint32_t kb_base;
ad705bcb 221 uint32_t unit_base;
3e260a46
JA
222
223 uint32_t latency_depth;
11e955e3 224 uint32_t pad3;
3e260a46
JA
225 uint64_t latency_target;
226 fio_fp64_t latency_percentile;
227 uint64_t latency_window;
16e56d25 228
e883cb35
JF
229 uint32_t sig_figs;
230
bb49c8bd
VF
231 uint64_t ss_dur;
232 uint32_t ss_state;
233 uint32_t ss_head;
bb49c8bd 234
bb49c8bd
VF
235 fio_fp64_t ss_limit;
236 fio_fp64_t ss_slope;
237 fio_fp64_t ss_deviation;
238 fio_fp64_t ss_criterion;
cb84f1fa
VF
239
240 union {
241 uint64_t *ss_iops_data;
11e955e3 242 uint64_t pad4;
cb84f1fa
VF
243 };
244
245 union {
246 uint64_t *ss_bw_data;
11e955e3 247 uint64_t pad5;
cb84f1fa 248 };
e5021b34 249} __attribute__((packed));
a64e88da 250
b75a394f
JA
251struct jobs_eta {
252 uint32_t nr_running;
253 uint32_t nr_ramp;
819db5f5 254
b75a394f 255 uint32_t nr_pending;
714e85f3 256 uint32_t nr_setting_up;
819db5f5 257
90eff1c9 258 uint64_t m_rate[DDIR_RWDIR_CNT], t_rate[DDIR_RWDIR_CNT];
90eff1c9 259 uint64_t rate[DDIR_RWDIR_CNT];
dbf285f7 260 uint32_t m_iops[DDIR_RWDIR_CNT], t_iops[DDIR_RWDIR_CNT];
6eaf09d6 261 uint32_t iops[DDIR_RWDIR_CNT];
b75a394f
JA
262 uint64_t elapsed_sec;
263 uint64_t eta_sec;
b7f05eb0 264 uint32_t is_pow2;
ad705bcb 265 uint32_t unit_base;
1d1f45ae 266
e883cb35
JF
267 uint32_t sig_figs;
268
dbf285f7
SW
269 uint32_t files_open;
270
1d1f45ae
JA
271 /*
272 * Network 'copy' of run_str[]
273 */
274 uint32_t nr_threads;
372aecb9 275 uint8_t run_str[];
e5021b34 276} __attribute__((packed));
b75a394f 277
65a4d15c
KC
278struct io_u_plat_entry {
279 struct flist_head list;
6cc0e5aa 280 uint64_t io_u_plat[FIO_IO_U_PLAT_NR];
65a4d15c
KC
281};
282
971caeb1 283extern struct fio_sem *stat_sem;
e5437a07 284
c5103619 285extern struct jobs_eta *get_jobs_eta(bool force, size_t *size);
723297c9 286
cef9175e
JA
287extern void stat_init(void);
288extern void stat_exit(void);
289
0279b880 290extern struct json_object * show_thread_status(struct thread_stat *ts, struct group_run_stats *rs, struct flist_head *, struct buf_output *);
a666cab8 291extern void show_group_stats(struct group_run_stats *rs, struct buf_output *);
8aa89d70 292extern bool calc_thread_status(struct jobs_eta *je, int force);
cf451d1e 293extern void display_thread_status(struct jobs_eta *je);
83f7b64e 294extern void __show_run_stats(void);
5ddc6707 295extern void __show_running_run_stats(void);
b852e7cf 296extern void show_running_run_stats(void);
06464907 297extern void check_for_running_stats(void);
fd595830 298extern void sum_thread_stats(struct thread_stat *dst, struct thread_stat *src, bool first);
37f0c1ae
JA
299extern void sum_group_stats(struct group_run_stats *dst, struct group_run_stats *src);
300extern void init_thread_stat(struct thread_stat *ts);
301extern void init_group_run_stat(struct group_run_stats *gs);
3e47bd25 302extern void eta_to_str(char *str, unsigned long eta_sec);
d6bb626e 303extern bool calc_lat(struct io_stat *is, unsigned long long *min, unsigned long long *max, double *mean, double *dev);
6cc0e5aa 304extern unsigned int calc_clat_percentiles(uint64_t *io_u_plat, unsigned long long nr, fio_fp64_t *plist, unsigned long long **output, unsigned long long *maxv, unsigned long long *minv);
fea76a0c 305extern void stat_calc_lat_n(struct thread_stat *ts, double *io_u_lat);
e5bd1347
JA
306extern void stat_calc_lat_m(struct thread_stat *ts, double *io_u_lat);
307extern void stat_calc_lat_u(struct thread_stat *ts, double *io_u_lat);
6cc0e5aa 308extern void stat_calc_dist(uint64_t *map, unsigned long total, double *io_u_dist);
6bb58215 309extern void reset_io_stats(struct thread_data *);
210dd0fc
JA
310extern void update_rusage_stat(struct thread_data *);
311extern void clear_rusage_stat(struct thread_data *);
2e33101f 312
d6bb626e 313extern void add_lat_sample(struct thread_data *, enum fio_ddir, unsigned long long,
5fff9543 314 unsigned long long, uint64_t);
d6bb626e 315extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long long,
5fff9543 316 unsigned long long, uint64_t);
cf8f852a 317extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long,
5fff9543
JF
318 unsigned long long, uint64_t);
319extern void add_agg_sample(union io_sample_data, enum fio_ddir, unsigned long long);
a47591e4
JA
320extern void add_iops_sample(struct thread_data *, struct io_u *,
321 unsigned int);
322extern void add_bw_sample(struct thread_data *, struct io_u *,
d6bb626e 323 unsigned int, unsigned long long);
b2b3eefe
JA
324extern void add_sync_clat_sample(struct thread_stat *ts,
325 unsigned long long nsec);
a47591e4 326extern int calc_log_samples(void);
cf8f852a
JA
327
328extern struct io_log *agg_io_log[DDIR_RWDIR_CNT];
329extern int write_bw_log;
330
74558486
JA
331static inline bool nsec_to_usec(unsigned long long *min,
332 unsigned long long *max, double *mean,
333 double *dev)
b29ad562 334{
d6bb626e 335 if (*min > 2000 && *max > 99999 && *dev > 1000.0) {
b29ad562
JA
336 *min /= 1000;
337 *max /= 1000;
338 *mean /= 1000.0;
339 *dev /= 1000.0;
8aa89d70 340 return true;
b29ad562
JA
341 }
342
8aa89d70 343 return false;
b29ad562 344}
d6bb626e 345
74558486
JA
346static inline bool nsec_to_msec(unsigned long long *min,
347 unsigned long long *max, double *mean,
348 double *dev)
d6bb626e
VF
349{
350 if (*min > 2000000 && *max > 99999999ULL && *dev > 1000000.0) {
351 *min /= 1000000;
352 *max /= 1000000;
353 *mean /= 1000000.0;
354 *dev /= 1000000.0;
355 return true;
356 }
357
358 return false;
359}
74558486 360
723297c9
JA
361/*
362 * Worst level condensing would be 1:5, so allow enough room for that
363 */
364#define __THREAD_RUNSTR_SZ(nr) ((nr) * 5)
11155857
JA
365#define THREAD_RUNSTR_SZ __THREAD_RUNSTR_SZ(thread_number)
366
66347cfa
DE
367uint32_t *io_u_block_info(struct thread_data *td, struct io_u *io_u);
368
a64e88da 369#endif