stat: ensure we align correctly
[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"
2a2fdab1
L
6#include "diskutil.h"
7#include "json.h"
ec41265e 8
a64e88da 9struct group_run_stats {
6eaf09d6
SL
10 uint64_t max_run[DDIR_RWDIR_CNT], min_run[DDIR_RWDIR_CNT];
11 uint64_t max_bw[DDIR_RWDIR_CNT], min_bw[DDIR_RWDIR_CNT];
af7f87cb 12 uint64_t iobytes[DDIR_RWDIR_CNT];
6eaf09d6 13 uint64_t agg[DDIR_RWDIR_CNT];
a64e88da 14 uint32_t kb_base;
ad705bcb 15 uint32_t unit_base;
e883cb35 16 uint32_t sig_figs;
a64e88da 17 uint32_t groupid;
771e58be 18 uint32_t unified_rw_rep;
e5021b34 19} __attribute__((packed));
a64e88da
JA
20
21/*
22 * How many depth levels to log
23 */
24#define FIO_IO_U_MAP_NR 7
d6bb626e 25#define FIO_IO_U_LAT_N_NR 10
a64e88da
JA
26#define FIO_IO_U_LAT_U_NR 10
27#define FIO_IO_U_LAT_M_NR 12
28
37b08652
VF
29/*
30 * Constants for clat percentiles
31 */
32#define FIO_IO_U_PLAT_BITS 6
33#define FIO_IO_U_PLAT_VAL (1 << FIO_IO_U_PLAT_BITS)
34#define FIO_IO_U_PLAT_GROUP_NR 29
35#define FIO_IO_U_PLAT_NR (FIO_IO_U_PLAT_GROUP_NR * FIO_IO_U_PLAT_VAL)
36#define FIO_IO_U_LIST_MAX_LEN 20 /* The size of the default and user-specified
37 list of percentiles */
38
a64e88da
JA
39/*
40 * Aggregate clat samples to report percentile(s) of them.
41 *
42 * EXECUTIVE SUMMARY
43 *
44 * FIO_IO_U_PLAT_BITS determines the maximum statistical error on the
45 * value of resulting percentiles. The error will be approximately
46 * 1/2^(FIO_IO_U_PLAT_BITS+1) of the value.
47 *
48 * FIO_IO_U_PLAT_GROUP_NR and FIO_IO_U_PLAT_BITS determine the maximum
49 * range being tracked for latency samples. The maximum value tracked
37b08652 50 * accurately will be 2^(GROUP_NR + PLAT_BITS - 1) nanoseconds.
a64e88da
JA
51 *
52 * FIO_IO_U_PLAT_GROUP_NR and FIO_IO_U_PLAT_BITS determine the memory
53 * requirement of storing those aggregate counts. The memory used will
54 * be (FIO_IO_U_PLAT_GROUP_NR * 2^FIO_IO_U_PLAT_BITS) * sizeof(int)
55 * bytes.
56 *
57 * FIO_IO_U_PLAT_NR is the total number of buckets.
58 *
59 * DETAILS
60 *
61 * Suppose the clat varies from 0 to 999 (usec), the straightforward
62 * method is to keep an array of (999 + 1) buckets, in which a counter
63 * keeps the count of samples which fall in the bucket, e.g.,
64 * {[0],[1],...,[999]}. However this consumes a huge amount of space,
65 * and can be avoided if an approximation is acceptable.
66 *
67 * One such method is to let the range of the bucket to be greater
68 * than one. This method has low accuracy when the value is small. For
69 * example, let the buckets be {[0,99],[100,199],...,[900,999]}, and
70 * the represented value of each bucket be the mean of the range. Then
71 * a value 0 has an round-off error of 49.5. To improve on this, we
72 * use buckets with non-uniform ranges, while bounding the error of
73 * each bucket within a ratio of the sample value. A simple example
74 * would be when error_bound = 0.005, buckets are {
75 * {[0],[1],...,[99]}, {[100,101],[102,103],...,[198,199]},..,
76 * {[900,909],[910,919]...} }. The total range is partitioned into
77 * groups with different ranges, then buckets with uniform ranges. An
78 * upper bound of the error is (range_of_bucket/2)/value_of_bucket
79 *
80 * For better efficiency, we implement this using base two. We group
81 * samples by their Most Significant Bit (MSB), extract the next M bit
82 * of them as an index within the group, and discard the rest of the
83 * bits.
84 *
85 * E.g., assume a sample 'x' whose MSB is bit n (starting from bit 0),
86 * and use M bit for indexing
87 *
88 * | n | M bits | bit (n-M-1) ... bit 0 |
89 *
90 * Because x is at least 2^n, and bit 0 to bit (n-M-1) is at most
91 * (2^(n-M) - 1), discarding bit 0 to (n-M-1) makes the round-off
92 * error
93 *
94 * 2^(n-M)-1 2^(n-M) 1
95 * e <= --------- <= ------- = ---
96 * 2^n 2^n 2^M
97 *
98 * Furthermore, we use "mean" of the range to represent the bucket,
99 * the error e can be lowered by half to 1 / 2^(M+1). By using M bits
100 * as the index, each group must contains 2^M buckets.
101 *
102 * E.g. Let M (FIO_IO_U_PLAT_BITS) be 6
103 * Error bound is 1/2^(6+1) = 0.0078125 (< 1%)
104 *
105 * Group MSB #discarded range of #buckets
106 * error_bits value
107 * ----------------------------------------------------------------
108 * 0* 0~5 0 [0,63] 64
109 * 1* 6 0 [64,127] 64
110 * 2 7 1 [128,255] 64
111 * 3 8 2 [256,511] 64
112 * 4 9 3 [512,1023] 64
113 * ... ... ... [...,...] ...
37b08652 114 * 28 33 27 [8589934592,+inf]** 64
a64e88da
JA
115 *
116 * * Special cases: when n < (M-1) or when n == (M-1), in both cases,
117 * the value cannot be rounded off. Use all bits of the sample as
118 * index.
119 *
37b08652 120 * ** If a sample's MSB is greater than 33, it will be counted as 33.
a64e88da
JA
121 */
122
66347cfa
DE
123/*
124 * Trim cycle count measurements
125 */
126#define MAX_NR_BLOCK_INFOS 8192
127#define BLOCK_INFO_STATE_SHIFT 29
128#define BLOCK_INFO_TRIMS(block_info) \
129 ((block_info) & ((1 << BLOCK_INFO_STATE_SHIFT) - 1))
130#define BLOCK_INFO_STATE(block_info) \
131 ((block_info) >> BLOCK_INFO_STATE_SHIFT)
132#define BLOCK_INFO(state, trim_cycles) \
fc8d6d05 133 ((trim_cycles) | ((unsigned int) (state) << BLOCK_INFO_STATE_SHIFT))
66347cfa
DE
134#define BLOCK_INFO_SET_STATE(block_info, state) \
135 BLOCK_INFO(state, BLOCK_INFO_TRIMS(block_info))
136enum block_info_state {
137 BLOCK_STATE_UNINIT,
138 BLOCK_STATE_TRIMMED,
139 BLOCK_STATE_WRITTEN,
140 BLOCK_STATE_TRIM_FAILURE,
141 BLOCK_STATE_WRITE_FAILURE,
142 BLOCK_STATE_COUNT,
1d6d3455 143};
66347cfa 144
a64e88da
JA
145#define MAX_PATTERN_SIZE 512
146#define FIO_JOBNAME_SIZE 128
4e59d0f3 147#define FIO_JOBDESC_SIZE 256
a64e88da
JA
148#define FIO_VERROR_SIZE 128
149
150struct thread_stat {
151 char name[FIO_JOBNAME_SIZE];
152 char verror[FIO_VERROR_SIZE];
ddcc0b69 153 uint32_t error;
2f122b13 154 uint32_t thread_number;
ddcc0b69 155 uint32_t groupid;
a64e88da 156 uint32_t pid;
4e59d0f3 157 char description[FIO_JOBDESC_SIZE];
a64e88da 158 uint32_t members;
771e58be 159 uint32_t unified_rw_rep;
a64e88da
JA
160
161 /*
162 * bandwidth and latency stats
163 */
1e900e2d 164 struct io_stat sync_stat __attribute__((aligned(8)));/* fsync etc stats */
6eaf09d6
SL
165 struct io_stat clat_stat[DDIR_RWDIR_CNT]; /* completion latency */
166 struct io_stat slat_stat[DDIR_RWDIR_CNT]; /* submission latency */
167 struct io_stat lat_stat[DDIR_RWDIR_CNT]; /* total latency */
168 struct io_stat bw_stat[DDIR_RWDIR_CNT]; /* bandwidth stats */
169 struct io_stat iops_stat[DDIR_RWDIR_CNT]; /* IOPS stats */
a64e88da
JA
170
171 /*
172 * fio system usage accounting
173 */
174 uint64_t usr_time;
175 uint64_t sys_time;
176 uint64_t ctx;
177 uint64_t minf, majf;
178
179 /*
180 * IO depth and latency stats
181 */
b599759b
JA
182 uint32_t clat_percentiles;
183 uint32_t lat_percentiles;
435d195a 184 uint64_t percentile_precision;
802ad4a8 185 fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];
a64e88da 186
6cc0e5aa
AL
187 uint64_t io_u_map[FIO_IO_U_MAP_NR];
188 uint64_t io_u_submit[FIO_IO_U_MAP_NR];
189 uint64_t io_u_complete[FIO_IO_U_MAP_NR];
190 uint64_t io_u_lat_n[FIO_IO_U_LAT_N_NR];
191 uint64_t io_u_lat_u[FIO_IO_U_LAT_U_NR];
192 uint64_t io_u_lat_m[FIO_IO_U_LAT_M_NR];
193 uint64_t io_u_plat[DDIR_RWDIR_CNT][FIO_IO_U_PLAT_NR];
194 uint64_t io_u_sync_plat[FIO_IO_U_PLAT_NR];
3d0ebb30 195
7f3ecee2 196 uint64_t total_io_u[DDIR_RWDIR_SYNC_CNT];
de5cdfea
JA
197 uint64_t short_io_u[DDIR_RWDIR_CNT];
198 uint64_t drop_io_u[DDIR_RWDIR_CNT];
a64e88da
JA
199 uint64_t total_submit;
200 uint64_t total_complete;
201
6eaf09d6
SL
202 uint64_t io_bytes[DDIR_RWDIR_CNT];
203 uint64_t runtime[DDIR_RWDIR_CNT];
a64e88da
JA
204 uint64_t total_run_time;
205
206 /*
207 * IO Error related stats
208 */
3d0ebb30
GG
209 union {
210 uint16_t continue_on_error;
11e955e3 211 uint32_t pad2;
3d0ebb30 212 };
ddcc0b69 213 uint32_t first_error;
11e955e3 214 uint64_t total_err_count;
a64e88da 215
fd5d733f
BVA
216 /* ZBD stats */
217 uint64_t nr_zone_resets;
218
66347cfa
DE
219 uint64_t nr_block_infos;
220 uint32_t block_infos[MAX_NR_BLOCK_INFOS];
221
a64e88da 222 uint32_t kb_base;
ad705bcb 223 uint32_t unit_base;
3e260a46
JA
224
225 uint32_t latency_depth;
11e955e3 226 uint32_t pad3;
3e260a46
JA
227 uint64_t latency_target;
228 fio_fp64_t latency_percentile;
229 uint64_t latency_window;
16e56d25 230
e883cb35
JF
231 uint32_t sig_figs;
232
bb49c8bd
VF
233 uint64_t ss_dur;
234 uint32_t ss_state;
235 uint32_t ss_head;
bb49c8bd 236
bb49c8bd
VF
237 fio_fp64_t ss_limit;
238 fio_fp64_t ss_slope;
239 fio_fp64_t ss_deviation;
240 fio_fp64_t ss_criterion;
cb84f1fa 241
4ee9c375 242 uint64_t io_u_plat_high_prio[DDIR_RWDIR_CNT][FIO_IO_U_PLAT_NR] __attribute__((aligned(8)));;
b2a432bf 243 uint64_t io_u_plat_prio[DDIR_RWDIR_CNT][FIO_IO_U_PLAT_NR];
4ee9c375 244 struct io_stat clat_high_prio_stat[DDIR_RWDIR_CNT] __attribute__((aligned(8)));
b2a432bf
PC
245 struct io_stat clat_prio_stat[DDIR_RWDIR_CNT];
246
cb84f1fa
VF
247 union {
248 uint64_t *ss_iops_data;
11e955e3 249 uint64_t pad4;
cb84f1fa
VF
250 };
251
252 union {
253 uint64_t *ss_bw_data;
11e955e3 254 uint64_t pad5;
cb84f1fa 255 };
96563db9
JA
256
257 uint64_t cachehit;
258 uint64_t cachemiss;
e5021b34 259} __attribute__((packed));
a64e88da 260
4c515ab4
BVA
261#define JOBS_ETA { \
262 uint32_t nr_running; \
263 uint32_t nr_ramp; \
264 \
265 uint32_t nr_pending; \
266 uint32_t nr_setting_up; \
267 \
cd8920a4
JA
268 uint64_t m_rate[DDIR_RWDIR_CNT]; \
269 uint64_t t_rate[DDIR_RWDIR_CNT]; \
4c515ab4 270 uint64_t rate[DDIR_RWDIR_CNT]; \
cd8920a4
JA
271 uint32_t m_iops[DDIR_RWDIR_CNT]; \
272 uint32_t t_iops[DDIR_RWDIR_CNT]; \
273 uint32_t iops[DDIR_RWDIR_CNT]; \
274 uint32_t pad; \
275 uint64_t elapsed_sec; \
276 uint64_t eta_sec; \
4c515ab4
BVA
277 uint32_t is_pow2; \
278 uint32_t unit_base; \
279 \
280 uint32_t sig_figs; \
281 \
282 uint32_t files_open; \
283 \
284 /* \
285 * Network 'copy' of run_str[] \
286 */ \
287 uint32_t nr_threads; \
cd8920a4 288 uint32_t pad2; \
4c515ab4
BVA
289 uint8_t run_str[]; \
290}
291
292struct jobs_eta JOBS_ETA;
293struct jobs_eta_packed JOBS_ETA __attribute__((packed));
b75a394f 294
65a4d15c
KC
295struct io_u_plat_entry {
296 struct flist_head list;
6cc0e5aa 297 uint64_t io_u_plat[FIO_IO_U_PLAT_NR];
65a4d15c
KC
298};
299
971caeb1 300extern struct fio_sem *stat_sem;
e5437a07 301
c5103619 302extern struct jobs_eta *get_jobs_eta(bool force, size_t *size);
723297c9 303
cef9175e
JA
304extern void stat_init(void);
305extern void stat_exit(void);
306
0279b880 307extern struct json_object * show_thread_status(struct thread_stat *ts, struct group_run_stats *rs, struct flist_head *, struct buf_output *);
a666cab8 308extern void show_group_stats(struct group_run_stats *rs, struct buf_output *);
8aa89d70 309extern bool calc_thread_status(struct jobs_eta *je, int force);
cf451d1e 310extern void display_thread_status(struct jobs_eta *je);
83f7b64e 311extern void __show_run_stats(void);
5ddc6707 312extern void __show_running_run_stats(void);
b852e7cf 313extern void show_running_run_stats(void);
06464907 314extern void check_for_running_stats(void);
fd595830 315extern void sum_thread_stats(struct thread_stat *dst, struct thread_stat *src, bool first);
37f0c1ae
JA
316extern void sum_group_stats(struct group_run_stats *dst, struct group_run_stats *src);
317extern void init_thread_stat(struct thread_stat *ts);
318extern void init_group_run_stat(struct group_run_stats *gs);
3e47bd25 319extern void eta_to_str(char *str, unsigned long eta_sec);
d6bb626e 320extern bool calc_lat(struct io_stat *is, unsigned long long *min, unsigned long long *max, double *mean, double *dev);
6cc0e5aa 321extern 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 322extern void stat_calc_lat_n(struct thread_stat *ts, double *io_u_lat);
e5bd1347
JA
323extern void stat_calc_lat_m(struct thread_stat *ts, double *io_u_lat);
324extern void stat_calc_lat_u(struct thread_stat *ts, double *io_u_lat);
6cc0e5aa 325extern void stat_calc_dist(uint64_t *map, unsigned long total, double *io_u_dist);
6bb58215 326extern void reset_io_stats(struct thread_data *);
210dd0fc
JA
327extern void update_rusage_stat(struct thread_data *);
328extern void clear_rusage_stat(struct thread_data *);
2e33101f 329
d6bb626e 330extern void add_lat_sample(struct thread_data *, enum fio_ddir, unsigned long long,
b2a432bf 331 unsigned long long, uint64_t, uint8_t);
d6bb626e 332extern void add_clat_sample(struct thread_data *, enum fio_ddir, unsigned long long,
b2a432bf 333 unsigned long long, uint64_t, uint8_t);
cf8f852a 334extern void add_slat_sample(struct thread_data *, enum fio_ddir, unsigned long,
b2a432bf
PC
335 unsigned long long, uint64_t, uint8_t);
336extern void add_agg_sample(union io_sample_data, enum fio_ddir, unsigned long long bs,
337 uint8_t priority_bit);
a47591e4
JA
338extern void add_iops_sample(struct thread_data *, struct io_u *,
339 unsigned int);
340extern void add_bw_sample(struct thread_data *, struct io_u *,
d6bb626e 341 unsigned int, unsigned long long);
b2b3eefe 342extern void add_sync_clat_sample(struct thread_stat *ts,
2a2fdab1 343 unsigned long long nsec);
a47591e4 344extern int calc_log_samples(void);
cf8f852a 345
2a2fdab1
L
346extern void print_disk_util(struct disk_util_stat *, struct disk_util_agg *, int terse, struct buf_output *);
347extern void json_array_add_disk_util(struct disk_util_stat *dus,
348 struct disk_util_agg *agg, struct json_array *parent);
349
cf8f852a 350extern struct io_log *agg_io_log[DDIR_RWDIR_CNT];
f1867a7f 351extern bool write_bw_log;
cf8f852a 352
74558486
JA
353static inline bool nsec_to_usec(unsigned long long *min,
354 unsigned long long *max, double *mean,
355 double *dev)
b29ad562 356{
d6bb626e 357 if (*min > 2000 && *max > 99999 && *dev > 1000.0) {
b29ad562
JA
358 *min /= 1000;
359 *max /= 1000;
360 *mean /= 1000.0;
361 *dev /= 1000.0;
8aa89d70 362 return true;
b29ad562
JA
363 }
364
8aa89d70 365 return false;
b29ad562 366}
d6bb626e 367
74558486
JA
368static inline bool nsec_to_msec(unsigned long long *min,
369 unsigned long long *max, double *mean,
370 double *dev)
d6bb626e
VF
371{
372 if (*min > 2000000 && *max > 99999999ULL && *dev > 1000000.0) {
373 *min /= 1000000;
374 *max /= 1000000;
375 *mean /= 1000000.0;
376 *dev /= 1000000.0;
377 return true;
378 }
379
380 return false;
381}
74558486 382
723297c9
JA
383/*
384 * Worst level condensing would be 1:5, so allow enough room for that
385 */
386#define __THREAD_RUNSTR_SZ(nr) ((nr) * 5)
11155857
JA
387#define THREAD_RUNSTR_SZ __THREAD_RUNSTR_SZ(thread_number)
388
66347cfa
DE
389uint32_t *io_u_block_info(struct thread_data *td, struct io_u *io_u);
390
a64e88da 391#endif