-
Notifications
You must be signed in to change notification settings - Fork 767
/
Copy pathvalkey-benchmark.c
2077 lines (1926 loc) · 82.3 KB
/
valkey-benchmark.c
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
/* Server benchmark utility.
*
* Copyright (c) 2009-2012, Redis Ltd.
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* * Redistributions of source code must retain the above copyright notice,
* this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* * Neither the name of Redis nor the names of its contributors may be used
* to endorse or promote products derived from this software without
* specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
* LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
* CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
* SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
* INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
* CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
* ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
* POSSIBILITY OF SUCH DAMAGE.
*/
#include "fmacros.h"
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <unistd.h>
#include <errno.h>
#include <time.h>
#include <sys/time.h>
#include <signal.h>
#include <assert.h>
#include <math.h>
#include <pthread.h>
#include <stdatomic.h>
#include <sdscompat.h> /* Use hiredis' sds compat header that maps sds calls to their hi_ variants */
#include <sds.h> /* Use hiredis sds. */
#include "ae.h"
#include <hiredis.h>
#ifdef USE_OPENSSL
#include <openssl/ssl.h>
#include <openssl/err.h>
#include <hiredis_ssl.h>
#endif
#include "adlist.h"
#include "dict.h"
#include "zmalloc.h"
#include "crc16_slottable.h"
#include "hdr_histogram.h"
#include "cli_common.h"
#include "mt19937-64.h"
#define UNUSED(V) ((void)V)
#define RANDPTR_INITIAL_SIZE 8
#define DEFAULT_LATENCY_PRECISION 3
#define MAX_LATENCY_PRECISION 4
#define MAX_THREADS 500
#define CLUSTER_SLOTS 16384
#define CONFIG_LATENCY_HISTOGRAM_MIN_VALUE 10L /* >= 10 usecs */
#define CONFIG_LATENCY_HISTOGRAM_MAX_VALUE 3000000L /* <= 3 secs(us precision) */
#define CONFIG_LATENCY_HISTOGRAM_INSTANT_MAX_VALUE 3000000L /* <= 3 secs(us precision) */
#define SHOW_THROUGHPUT_INTERVAL 250 /* 250ms */
#define CLIENT_GET_EVENTLOOP(c) (c->thread_id >= 0 ? config.threads[c->thread_id]->el : config.el)
struct benchmarkThread;
struct clusterNode;
struct serverConfig;
/* Read from replica options */
typedef enum readFromReplica {
FROM_PRIMARY_ONLY = 0, /* default option */
FROM_REPLICA_ONLY,
FROM_ALL
} readFromReplica;
static struct config {
aeEventLoop *el;
cliConnInfo conn_info;
const char *hostsocket;
int tls;
struct cliSSLconfig sslconfig;
int numclients;
_Atomic int liveclients;
int requests;
_Atomic int requests_issued;
_Atomic int requests_finished;
_Atomic int previous_requests_finished;
int last_printed_bytes;
long long previous_tick;
int keysize;
int datasize;
int replacekeys;
int keyspacelen;
int sequential_replacement;
int keepalive;
int pipeline;
long long start;
long long totlatency;
const char *title;
list *clients;
int quiet;
int csv;
int loop;
int idlemode;
sds input_dbnumstr;
char *tests;
int stdinarg; /* get last arg from stdin. (-x option) */
int precision;
int num_threads;
struct benchmarkThread **threads;
int cluster_mode;
readFromReplica read_from_replica;
int cluster_node_count;
struct clusterNode **cluster_nodes;
struct serverConfig *redis_config;
struct hdr_histogram *latency_histogram;
struct hdr_histogram *current_sec_latency_histogram;
_Atomic int is_fetching_slots;
_Atomic int is_updating_slots;
_Atomic int slots_last_update;
int enable_tracking;
int num_functions;
int num_keys_in_fcall;
pthread_mutex_t liveclients_mutex;
pthread_mutex_t is_updating_slots_mutex;
int resp3; /* use RESP3 */
} config;
typedef struct _client {
redisContext *context;
sds obuf;
char **randptr; /* Pointers to :rand: strings inside the command buf */
size_t randlen; /* Number of pointers in client->randptr */
size_t randfree; /* Number of unused pointers in client->randptr */
char **stagptr; /* Pointers to slot hashtags (cluster mode only) */
size_t staglen; /* Number of pointers in client->stagptr */
size_t stagfree; /* Number of unused pointers in client->stagptr */
size_t written; /* Bytes of 'obuf' already written */
long long start; /* Start time of a request */
long long latency; /* Request latency */
int pending; /* Number of pending requests (replies to consume) */
int prefix_pending; /* If non-zero, number of pending prefix commands. Commands
such as auth and select are prefixed to the pipeline of
benchmark commands and discarded after the first send. */
int prefixlen; /* Size in bytes of the pending prefix commands */
int thread_id;
struct clusterNode *cluster_node;
int slots_last_update;
} *client;
/* Threads. */
typedef struct benchmarkThread {
int index;
pthread_t thread;
aeEventLoop *el;
} benchmarkThread;
/* Cluster. */
typedef struct clusterNode {
char *ip;
int port;
sds name;
int flags;
sds replicate; /* Primary ID if node is a replica */
int *slots;
int slots_count;
int *updated_slots; /* Used by updateClusterSlotsConfiguration */
int updated_slots_count; /* Used by updateClusterSlotsConfiguration */
int replicas_count;
struct serverConfig *redis_config;
} clusterNode;
typedef struct serverConfig {
sds save;
sds appendonly;
} serverConfig;
/* Prototypes */
static void writeHandler(aeEventLoop *el, int fd, void *privdata, int mask);
static void createMissingClients(client c);
static benchmarkThread *createBenchmarkThread(int index);
static void freeBenchmarkThread(benchmarkThread *thread);
static void freeBenchmarkThreads(void);
static void *execBenchmarkThread(void *ptr);
static clusterNode *createClusterNode(char *ip, int port);
static serverConfig *getServerConfig(const char *ip, int port, const char *hostsocket);
static redisContext *getRedisContext(const char *ip, int port, const char *hostsocket);
static void freeServerConfig(serverConfig *cfg);
static int fetchClusterSlotsConfiguration(client c);
static void updateClusterSlotsConfiguration(void);
static long long showThroughput(struct aeEventLoop *eventLoop, long long id, void *clientData);
/* Dict callbacks */
static uint64_t dictSdsHash(const void *key);
static int dictSdsKeyCompare(const void *key1, const void *key2);
/* Implementation */
static long long ustime(void) {
struct timeval tv;
long long ust;
gettimeofday(&tv, NULL);
ust = ((long long)tv.tv_sec) * 1000000;
ust += tv.tv_usec;
return ust;
}
static long long mstime(void) {
return ustime() / 1000;
}
static uint64_t dictSdsHash(const void *key) {
return dictGenHashFunction((unsigned char *)key, sdslen((char *)key));
}
static int dictSdsKeyCompare(const void *key1, const void *key2) {
int l1, l2;
l1 = sdslen((sds)key1);
l2 = sdslen((sds)key2);
if (l1 != l2) return 0;
return memcmp(key1, key2, l1) == 0;
}
static dictType dtype = {
dictSdsHash, /* hash function */
NULL, /* key dup */
dictSdsKeyCompare, /* key compare */
NULL, /* key destructor */
NULL, /* val destructor */
NULL /* allow to expand */
};
static redisContext *getRedisContext(const char *ip, int port, const char *hostsocket) {
redisContext *ctx = NULL;
redisReply *reply = NULL;
struct timeval tv = {0};
if (hostsocket == NULL)
ctx = redisConnectWrapper(ip, port, tv, 0);
else
ctx = redisConnectUnixWrapper(hostsocket, tv, 0);
if (ctx == NULL || ctx->err) {
fprintf(stderr, "Could not connect to server at ");
char *err = (ctx != NULL ? ctx->errstr : "");
if (hostsocket == NULL)
fprintf(stderr, "%s:%d: %s\n", ip, port, err);
else
fprintf(stderr, "%s: %s\n", hostsocket, err);
goto cleanup;
}
if (config.tls == 1) {
const char *err = NULL;
if (cliSecureConnection(ctx, config.sslconfig, &err) == REDIS_ERR && err) {
fprintf(stderr, "Could not negotiate a TLS connection: %s\n", err);
goto cleanup;
}
}
if (config.conn_info.auth == NULL) return ctx;
if (config.conn_info.user == NULL)
reply = redisCommand(ctx, "AUTH %s", config.conn_info.auth);
else
reply = redisCommand(ctx, "AUTH %s %s", config.conn_info.user, config.conn_info.auth);
if (reply != NULL) {
if (reply->type == REDIS_REPLY_ERROR) {
if (hostsocket == NULL)
fprintf(stderr, "Node %s:%d replied with error:\n%s\n", ip, port, reply->str);
else
fprintf(stderr, "Node %s replied with error:\n%s\n", hostsocket, reply->str);
freeReplyObject(reply);
redisFree(ctx);
exit(1);
}
freeReplyObject(reply);
return ctx;
}
fprintf(stderr, "ERROR: failed to fetch reply from ");
if (hostsocket == NULL)
fprintf(stderr, "%s:%d\n", ip, port);
else
fprintf(stderr, "%s\n", hostsocket);
cleanup:
freeReplyObject(reply);
redisFree(ctx);
return NULL;
}
static serverConfig *getServerConfig(const char *ip, int port, const char *hostsocket) {
serverConfig *cfg = zcalloc(sizeof(*cfg));
if (!cfg) return NULL;
redisContext *c = NULL;
redisReply *reply = NULL, *sub_reply = NULL;
c = getRedisContext(ip, port, hostsocket);
if (c == NULL) {
freeServerConfig(cfg);
exit(1);
}
redisAppendCommand(c, "CONFIG GET %s", "save");
redisAppendCommand(c, "CONFIG GET %s", "appendonly");
int abort_test = 0;
int i = 0;
void *r = NULL;
for (; i < 2; i++) {
int res = redisGetReply(c, &r);
if (reply) freeReplyObject(reply);
reply = res == REDIS_OK ? ((redisReply *)r) : NULL;
if (res != REDIS_OK || !r) goto fail;
if (reply->type == REDIS_REPLY_ERROR) {
goto fail;
}
if (reply->type != REDIS_REPLY_ARRAY || reply->elements < 2) goto fail;
sub_reply = reply->element[1];
char *value = sub_reply->str;
if (!value) value = "";
switch (i) {
case 0: cfg->save = sdsnew(value); break;
case 1: cfg->appendonly = sdsnew(value); break;
}
}
freeReplyObject(reply);
redisFree(c);
return cfg;
fail:
if (reply && reply->type == REDIS_REPLY_ERROR && !strncmp(reply->str, "NOAUTH", 6)) {
if (hostsocket == NULL)
fprintf(stderr, "Node %s:%d replied with error:\n%s\n", ip, port, reply->str);
else
fprintf(stderr, "Node %s replied with error:\n%s\n", hostsocket, reply->str);
abort_test = 1;
}
freeReplyObject(reply);
redisFree(c);
freeServerConfig(cfg);
if (abort_test) exit(1);
return NULL;
}
static void freeServerConfig(serverConfig *cfg) {
if (cfg->save) sdsfree(cfg->save);
if (cfg->appendonly) sdsfree(cfg->appendonly);
zfree(cfg);
}
static void freeClient(client c) {
aeEventLoop *el = CLIENT_GET_EVENTLOOP(c);
listNode *ln;
aeDeleteFileEvent(el, c->context->fd, AE_WRITABLE);
aeDeleteFileEvent(el, c->context->fd, AE_READABLE);
if (c->thread_id >= 0) {
int requests_finished = atomic_load_explicit(&config.requests_finished, memory_order_relaxed);
if (requests_finished >= config.requests) {
aeStop(el);
}
}
redisFree(c->context);
sdsfree(c->obuf);
zfree(c->randptr);
zfree(c->stagptr);
zfree(c);
if (config.num_threads) pthread_mutex_lock(&(config.liveclients_mutex));
config.liveclients--;
ln = listSearchKey(config.clients, c);
assert(ln != NULL);
listDelNode(config.clients, ln);
if (config.num_threads) pthread_mutex_unlock(&(config.liveclients_mutex));
}
static void freeAllClients(void) {
listNode *ln = config.clients->head, *next;
while (ln) {
next = ln->next;
freeClient(ln->value);
ln = next;
}
}
static void resetClient(client c) {
aeEventLoop *el = CLIENT_GET_EVENTLOOP(c);
aeDeleteFileEvent(el, c->context->fd, AE_WRITABLE);
aeDeleteFileEvent(el, c->context->fd, AE_READABLE);
aeCreateFileEvent(el, c->context->fd, AE_WRITABLE, writeHandler, c);
c->written = 0;
c->pending = config.pipeline;
}
static void generateClientKey(client c) {
static _Atomic size_t seq_key = 0;
for (size_t i = 0; i < c->randlen; i++) {
char *p = c->randptr[i] + 11;
size_t key = 0;
if (config.keyspacelen != 0) {
if (config.sequential_replacement) {
key = atomic_fetch_add_explicit(&seq_key, 1, memory_order_relaxed);
} else {
key = random();
}
key %= config.keyspacelen;
}
for (size_t j = 0; j < 12; j++) {
*p = '0' + key % 10;
key /= 10;
p--;
}
}
}
static void setClusterKeyHashTag(client c) {
assert(c->thread_id >= 0);
clusterNode *node = c->cluster_node;
assert(node);
int is_updating_slots = atomic_load_explicit(&config.is_updating_slots, memory_order_relaxed);
/* If updateClusterSlotsConfiguration is updating the slots array,
* call updateClusterSlotsConfiguration is order to block the thread
* since the mutex is locked. When the slots will be updated by the
* thread that's actually performing the update, the execution of
* updateClusterSlotsConfiguration won't actually do anything, since
* the updated_slots_count array will be already NULL. */
if (is_updating_slots) updateClusterSlotsConfiguration();
int slot = node->slots[rand() % node->slots_count];
const char *tag = crc16_slot_table[slot];
int taglen = strlen(tag);
size_t i;
for (i = 0; i < c->staglen; i++) {
char *p = c->stagptr[i] + 1;
p[0] = tag[0];
p[1] = (taglen >= 2 ? tag[1] : '}');
p[2] = (taglen == 3 ? tag[2] : '}');
}
}
static void clientDone(client c) {
int requests_finished = atomic_load_explicit(&config.requests_finished, memory_order_relaxed);
if (requests_finished >= config.requests) {
freeClient(c);
if (!config.num_threads && config.el) aeStop(config.el);
return;
}
if (config.keepalive) {
resetClient(c);
} else {
if (config.num_threads) pthread_mutex_lock(&(config.liveclients_mutex));
config.liveclients--;
createMissingClients(c);
config.liveclients++;
if (config.num_threads) pthread_mutex_unlock(&(config.liveclients_mutex));
freeClient(c);
}
}
static void readHandler(aeEventLoop *el, int fd, void *privdata, int mask) {
client c = privdata;
void *reply = NULL;
UNUSED(el);
UNUSED(fd);
UNUSED(mask);
/* Calculate latency only for the first read event. This means that the
* server already sent the reply and we need to parse it. Parsing overhead
* is not part of the latency, so calculate it only once, here. */
if (c->latency < 0) c->latency = ustime() - (c->start);
if (redisBufferRead(c->context) != REDIS_OK) {
fprintf(stderr, "Error: %s\n", c->context->errstr);
exit(1);
} else {
while (c->pending) {
if (redisGetReply(c->context, &reply) != REDIS_OK) {
fprintf(stderr, "Error: %s\n", c->context->errstr);
exit(1);
}
if (reply != NULL) {
if (reply == (void *)REDIS_REPLY_ERROR) {
fprintf(stderr, "Unexpected error reply, exiting...\n");
exit(1);
}
redisReply *r = reply;
if (r->type == REDIS_REPLY_ERROR) {
/* Try to update slots configuration if reply error is
* MOVED/ASK/CLUSTERDOWN and the key(s) used by the command
* contain(s) the slot hash tag.
* If the error is not topology-update related then we
* immediately exit to avoid false results. */
if (c->cluster_node && c->staglen) {
int fetch_slots = 0, do_wait = 0;
if (!strncmp(r->str, "MOVED", 5) || !strncmp(r->str, "ASK", 3))
fetch_slots = 1;
else if (!strncmp(r->str, "CLUSTERDOWN", 11)) {
/* Usually the cluster is able to recover itself after
* a CLUSTERDOWN error, so try to sleep one second
* before requesting the new configuration. */
fetch_slots = 1;
do_wait = 1;
fprintf(stderr, "Error from server %s:%d: %s.\n", c->cluster_node->ip,
c->cluster_node->port, r->str);
}
if (do_wait) sleep(1);
if (fetch_slots && !fetchClusterSlotsConfiguration(c)) exit(1);
} else {
if (c->cluster_node) {
fprintf(stderr, "Error from server %s:%d: %s\n", c->cluster_node->ip, c->cluster_node->port,
r->str);
} else
fprintf(stderr, "Error from server: %s\n", r->str);
exit(1);
}
}
freeReplyObject(reply);
/* This is an OK for prefix commands such as auth and select.*/
if (c->prefix_pending > 0) {
c->prefix_pending--;
c->pending--;
/* Discard prefix commands on first response.*/
if (c->prefixlen > 0) {
size_t j;
sdsrange(c->obuf, c->prefixlen, -1);
/* We also need to fix the pointers to the strings
* we need to randomize. */
for (j = 0; j < c->randlen; j++) c->randptr[j] -= c->prefixlen;
/* Fix the pointers to the slot hash tags */
for (j = 0; j < c->staglen; j++) c->stagptr[j] -= c->prefixlen;
c->prefixlen = 0;
}
continue;
}
int requests_finished = atomic_fetch_add_explicit(&config.requests_finished, 1, memory_order_relaxed);
if (requests_finished < config.requests) {
if (config.num_threads == 0) {
hdr_record_value(config.latency_histogram, // Histogram to record to
(long)c->latency <= CONFIG_LATENCY_HISTOGRAM_MAX_VALUE
? (long)c->latency
: CONFIG_LATENCY_HISTOGRAM_MAX_VALUE); // Value to record
hdr_record_value(config.current_sec_latency_histogram, // Histogram to record to
(long)c->latency <= CONFIG_LATENCY_HISTOGRAM_INSTANT_MAX_VALUE
? (long)c->latency
: CONFIG_LATENCY_HISTOGRAM_INSTANT_MAX_VALUE); // Value to record
} else {
hdr_record_value_atomic(config.latency_histogram, // Histogram to record to
(long)c->latency <= CONFIG_LATENCY_HISTOGRAM_MAX_VALUE
? (long)c->latency
: CONFIG_LATENCY_HISTOGRAM_MAX_VALUE); // Value to record
hdr_record_value_atomic(config.current_sec_latency_histogram, // Histogram to record to
(long)c->latency <= CONFIG_LATENCY_HISTOGRAM_INSTANT_MAX_VALUE
? (long)c->latency
: CONFIG_LATENCY_HISTOGRAM_INSTANT_MAX_VALUE); // Value to record
}
}
c->pending--;
if (c->pending == 0) {
clientDone(c);
break;
}
} else {
break;
}
}
}
}
static void writeHandler(aeEventLoop *el, int fd, void *privdata, int mask) {
client c = privdata;
UNUSED(el);
UNUSED(fd);
UNUSED(mask);
/* Initialize request when nothing was written. */
if (c->written == 0) {
/* Enforce upper bound to number of requests. */
int requests_issued = atomic_fetch_add_explicit(&config.requests_issued, config.pipeline, memory_order_relaxed);
if (requests_issued >= config.requests) {
return;
}
/* Really initialize: replace keys and set start time. */
if (config.replacekeys) generateClientKey(c);
if (config.cluster_mode && c->staglen > 0) setClusterKeyHashTag(c);
c->slots_last_update = atomic_load_explicit(&config.slots_last_update, memory_order_relaxed);
c->start = ustime();
c->latency = -1;
}
const ssize_t buflen = sdslen(c->obuf);
const ssize_t writeLen = buflen - c->written;
if (writeLen > 0) {
void *ptr = c->obuf + c->written;
while (1) {
/* Optimistically try to write before checking if the file descriptor
* is actually writable. At worst we get EAGAIN. */
const ssize_t nwritten = cliWriteConn(c->context, ptr, writeLen);
if (nwritten != writeLen) {
if (nwritten == -1 && errno != EAGAIN) {
if (errno != EPIPE) fprintf(stderr, "Error writing to the server: %s\n", strerror(errno));
freeClient(c);
return;
} else if (nwritten > 0) {
c->written += nwritten;
return;
}
} else {
aeDeleteFileEvent(el, c->context->fd, AE_WRITABLE);
aeCreateFileEvent(el, c->context->fd, AE_READABLE, readHandler, c);
return;
}
}
}
}
/* Create a benchmark client, configured to send the command passed as 'cmd' of
* 'len' bytes.
*
* The command is copied N times in the client output buffer (that is reused
* again and again to send the request to the server) accordingly to the configured
* pipeline size.
*
* Also an initial SELECT command is prepended in order to make sure the right
* database is selected, if needed. The initial SELECT will be discarded as soon
* as the first reply is received.
*
* To create a client from scratch, the 'from' pointer is set to NULL. If instead
* we want to create a client using another client as reference, the 'from' pointer
* points to the client to use as reference. In such a case the following
* information is take from the 'from' client:
*
* 1) The command line to use.
* 2) The offsets of the __rand_int__ elements inside the command line, used
* for arguments randomization.
*
* Even when cloning another client, prefix commands are applied if needed.*/
static client createClient(char *cmd, size_t len, client from, int thread_id) {
int j;
int is_cluster_client = (config.cluster_mode && thread_id >= 0);
client c = zmalloc(sizeof(struct _client));
const char *ip = NULL;
int port = 0;
struct timeval tv = {0};
c->cluster_node = NULL;
if (config.hostsocket == NULL || is_cluster_client) {
if (!is_cluster_client) {
ip = config.conn_info.hostip;
port = config.conn_info.hostport;
} else {
int node_idx = 0;
if (config.num_threads < config.cluster_node_count)
node_idx = config.liveclients % config.cluster_node_count;
else
node_idx = thread_id % config.cluster_node_count;
clusterNode *node = config.cluster_nodes[node_idx];
assert(node != NULL);
ip = (const char *)node->ip;
port = node->port;
c->cluster_node = node;
}
c->context = redisConnectWrapper(ip, port, tv, 1);
} else {
c->context = redisConnectUnixWrapper(config.hostsocket, tv, 1);
}
if (c->context->err) {
fprintf(stderr, "Could not connect to server at ");
if (config.hostsocket == NULL || is_cluster_client)
fprintf(stderr, "%s:%d: %s\n", ip, port, c->context->errstr);
else
fprintf(stderr, "%s: %s\n", config.hostsocket, c->context->errstr);
exit(1);
}
if (config.tls == 1) {
const char *err = NULL;
if (cliSecureConnection(c->context, config.sslconfig, &err) == REDIS_ERR && err) {
fprintf(stderr, "Could not negotiate a TLS connection: %s\n", err);
exit(1);
}
}
c->thread_id = thread_id;
/* Suppress hiredis cleanup of unused buffers for max speed. */
c->context->reader->maxbuf = 0;
/* Build the request buffer:
* Queue N requests accordingly to the pipeline size, or simply clone
* the example client buffer. */
c->obuf = sdsempty();
/* Prefix the request buffer with AUTH and/or SELECT commands, if applicable.
* These commands are discarded after the first response, so if the client is
* reused the commands will not be used again. */
c->prefix_pending = 0;
if (config.conn_info.auth) {
char *buf = NULL;
int len;
if (config.conn_info.user == NULL)
len = redisFormatCommand(&buf, "AUTH %s", config.conn_info.auth);
else
len = redisFormatCommand(&buf, "AUTH %s %s", config.conn_info.user, config.conn_info.auth);
c->obuf = sdscatlen(c->obuf, buf, len);
free(buf);
c->prefix_pending++;
}
if (config.enable_tracking) {
char *buf = NULL;
int len = redisFormatCommand(&buf, "CLIENT TRACKING on");
c->obuf = sdscatlen(c->obuf, buf, len);
free(buf);
c->prefix_pending++;
}
/* If a DB number different than zero is selected, prefix our request
* buffer with the SELECT command, that will be discarded the first
* time the replies are received, so if the client is reused the
* SELECT command will not be used again. */
if (config.conn_info.input_dbnum != 0 && !is_cluster_client) {
c->obuf = sdscatprintf(c->obuf, "*2\r\n$6\r\nSELECT\r\n$%d\r\n%s\r\n", (int)sdslen(config.input_dbnumstr),
config.input_dbnumstr);
c->prefix_pending++;
}
if (config.resp3) {
char *buf = NULL;
int len = redisFormatCommand(&buf, "HELLO 3");
c->obuf = sdscatlen(c->obuf, buf, len);
free(buf);
c->prefix_pending++;
}
if (config.cluster_mode && (config.read_from_replica == FROM_REPLICA_ONLY || config.read_from_replica == FROM_ALL)) {
char *buf = NULL;
int len;
len = redisFormatCommand(&buf, "READONLY");
c->obuf = sdscatlen(c->obuf, buf, len);
free(buf);
c->prefix_pending++;
}
c->prefixlen = sdslen(c->obuf);
/* Append the request itself. */
if (from) {
c->obuf = sdscatlen(c->obuf, from->obuf + from->prefixlen, sdslen(from->obuf) - from->prefixlen);
} else {
for (j = 0; j < config.pipeline; j++) c->obuf = sdscatlen(c->obuf, cmd, len);
}
c->written = 0;
c->pending = config.pipeline + c->prefix_pending;
c->randptr = NULL;
c->randlen = 0;
c->stagptr = NULL;
c->staglen = 0;
/* Find substrings in the output buffer that need to be replaced. */
if (config.replacekeys) {
if (from) {
c->randlen = from->randlen;
c->randfree = 0;
c->randptr = zmalloc(sizeof(char *) * c->randlen);
/* copy the offsets. */
for (j = 0; j < (int)c->randlen; j++) {
c->randptr[j] = c->obuf + (from->randptr[j] - from->obuf);
/* Adjust for the different select prefix length. */
c->randptr[j] += c->prefixlen - from->prefixlen;
}
} else {
char *p = c->obuf;
c->randlen = 0;
c->randfree = RANDPTR_INITIAL_SIZE;
c->randptr = zmalloc(sizeof(char *) * c->randfree);
while ((p = strstr(p, "__rand_int__")) != NULL) {
if (c->randfree == 0) {
c->randptr = zrealloc(c->randptr, sizeof(char *) * c->randlen * 2);
c->randfree += c->randlen;
}
c->randptr[c->randlen++] = p;
c->randfree--;
p += 12; /* 12 is strlen("__rand_int__). */
}
}
}
/* If cluster mode is enabled, set slot hashtags pointers. */
if (config.cluster_mode) {
if (from) {
c->staglen = from->staglen;
c->stagfree = 0;
c->stagptr = zmalloc(sizeof(char *) * c->staglen);
/* copy the offsets. */
for (j = 0; j < (int)c->staglen; j++) {
c->stagptr[j] = c->obuf + (from->stagptr[j] - from->obuf);
/* Adjust for the different select prefix length. */
c->stagptr[j] += c->prefixlen - from->prefixlen;
}
} else {
char *p = c->obuf;
c->staglen = 0;
c->stagfree = RANDPTR_INITIAL_SIZE;
c->stagptr = zmalloc(sizeof(char *) * c->stagfree);
while ((p = strstr(p, "{tag}")) != NULL) {
if (c->stagfree == 0) {
c->stagptr = zrealloc(c->stagptr, sizeof(char *) * c->staglen * 2);
c->stagfree += c->staglen;
}
c->stagptr[c->staglen++] = p;
c->stagfree--;
p += 5; /* 5 is strlen("{tag}"). */
}
}
}
aeEventLoop *el = NULL;
if (thread_id < 0)
el = config.el;
else {
benchmarkThread *thread = config.threads[thread_id];
el = thread->el;
}
if (config.idlemode == 0)
aeCreateFileEvent(el, c->context->fd, AE_WRITABLE, writeHandler, c);
else
/* In idle mode, clients still need to register readHandler for catching errors */
aeCreateFileEvent(el, c->context->fd, AE_READABLE, readHandler, c);
listAddNodeTail(config.clients, c);
atomic_fetch_add_explicit(&config.liveclients, 1, memory_order_relaxed);
c->slots_last_update = atomic_load_explicit(&config.slots_last_update, memory_order_relaxed);
return c;
}
static void createMissingClients(client c) {
int n = 0;
while (config.liveclients < config.numclients) {
int thread_id = -1;
if (config.num_threads) thread_id = config.liveclients % config.num_threads;
createClient(NULL, 0, c, thread_id);
/* Listen backlog is quite limited on most systems */
if (++n > 64) {
usleep(50000);
n = 0;
}
}
}
static void showLatencyReport(void) {
const float reqpersec = (float)config.requests_finished / ((float)config.totlatency / 1000.0f);
const float p0 = ((float)hdr_min(config.latency_histogram)) / 1000.0f;
const float p50 = hdr_value_at_percentile(config.latency_histogram, 50.0) / 1000.0f;
const float p95 = hdr_value_at_percentile(config.latency_histogram, 95.0) / 1000.0f;
const float p99 = hdr_value_at_percentile(config.latency_histogram, 99.0) / 1000.0f;
const float p100 = ((float)hdr_max(config.latency_histogram)) / 1000.0f;
const float avg = hdr_mean(config.latency_histogram) / 1000.0f;
if (!config.quiet && !config.csv) {
printf("%*s\r", config.last_printed_bytes, " "); // ensure there is a clean line
printf("====== %s ======\n", config.title);
printf(" %d requests completed in %.2f seconds\n", config.requests_finished, (float)config.totlatency / 1000);
printf(" %d parallel clients\n", config.numclients);
printf(" %d bytes payload\n", config.datasize);
printf(" keep alive: %d\n", config.keepalive);
if (config.cluster_mode) {
const char *node_roles = NULL;
if (config.read_from_replica == FROM_ALL) {
node_roles = "cluster";
} else if (config.read_from_replica == FROM_REPLICA_ONLY) {
node_roles = "replica";
} else {
node_roles = "primary";
}
printf(" cluster mode: yes (%d %s)\n", config.cluster_node_count, node_roles);
int m;
for (m = 0; m < config.cluster_node_count; m++) {
clusterNode *node = config.cluster_nodes[m];
serverConfig *cfg = node->redis_config;
if (cfg == NULL) continue;
printf(" node [%d] configuration:\n", m);
printf(" save: %s\n", sdslen(cfg->save) ? cfg->save : "NONE");
printf(" appendonly: %s\n", cfg->appendonly);
}
} else {
if (config.redis_config) {
printf(" host configuration \"save\": %s\n", config.redis_config->save);
printf(" host configuration \"appendonly\": %s\n", config.redis_config->appendonly);
}
}
printf(" multi-thread: %s\n", (config.num_threads ? "yes" : "no"));
if (config.num_threads) printf(" threads: %d\n", config.num_threads);
printf("\n");
printf("Latency by percentile distribution:\n");
struct hdr_iter iter;
long long previous_cumulative_count = -1;
const long long total_count = config.latency_histogram->total_count;
hdr_iter_percentile_init(&iter, config.latency_histogram, 1);
struct hdr_iter_percentiles *percentiles = &iter.specifics.percentiles;
while (hdr_iter_next(&iter)) {
const double value = iter.highest_equivalent_value / 1000.0f;
const double percentile = percentiles->percentile;
const long long cumulative_count = iter.cumulative_count;
if (previous_cumulative_count != cumulative_count || cumulative_count == total_count) {
printf("%3.3f%% <= %.3f milliseconds (cumulative count %lld)\n", percentile, value, cumulative_count);
}
previous_cumulative_count = cumulative_count;
}
printf("\n");
printf("Cumulative distribution of latencies:\n");
previous_cumulative_count = -1;
hdr_iter_linear_init(&iter, config.latency_histogram, 100);
while (hdr_iter_next(&iter)) {
const double value = iter.highest_equivalent_value / 1000.0f;
const long long cumulative_count = iter.cumulative_count;
const double percentile = ((double)cumulative_count / (double)total_count) * 100.0;
if (previous_cumulative_count != cumulative_count || cumulative_count == total_count) {
printf("%3.3f%% <= %.3f milliseconds (cumulative count %lld)\n", percentile, value, cumulative_count);
}
/* After the 2 milliseconds latency to have percentages split
* by decimals will just add a lot of noise to the output. */
if (iter.highest_equivalent_value > 2000) {
hdr_iter_linear_set_value_units_per_bucket(&iter, 1000);
}
previous_cumulative_count = cumulative_count;
}
printf("\n");
printf("Summary:\n");
printf(" throughput summary: %.2f requests per second\n", reqpersec);
printf(" latency summary (msec):\n");
printf(" %9s %9s %9s %9s %9s %9s\n", "avg", "min", "p50", "p95", "p99", "max");
printf(" %9.3f %9.3f %9.3f %9.3f %9.3f %9.3f\n", avg, p0, p50, p95, p99, p100);
} else if (config.csv) {
printf("\"%s\",\"%.2f\",\"%.3f\",\"%.3f\",\"%.3f\",\"%.3f\",\"%.3f\",\"%.3f\"\n", config.title, reqpersec, avg,
p0, p50, p95, p99, p100);
} else {
printf("%*s\r", config.last_printed_bytes, " "); // ensure there is a clean line
printf("%s: %.2f requests per second, p50=%.3f msec\n", config.title, reqpersec, p50);
}
}
static void initBenchmarkThreads(void) {
int i;
if (config.threads) freeBenchmarkThreads();
config.threads = zmalloc(config.num_threads * sizeof(benchmarkThread *));
for (i = 0; i < config.num_threads; i++) {
benchmarkThread *thread = createBenchmarkThread(i);
config.threads[i] = thread;
}
}
static void startBenchmarkThreads(void) {
int i;
for (i = 0; i < config.num_threads; i++) {
benchmarkThread *t = config.threads[i];
if (pthread_create(&(t->thread), NULL, execBenchmarkThread, t)) {
fprintf(stderr, "FATAL: Failed to start thread %d.\n", i);
exit(1);
}
}
for (i = 0; i < config.num_threads; i++) pthread_join(config.threads[i]->thread, NULL);
}
static void benchmark(const char *title, char *cmd, int len) {
client c;
config.title = title;
config.requests_issued = 0;
config.requests_finished = 0;
config.previous_requests_finished = 0;
config.last_printed_bytes = 0;
hdr_init(CONFIG_LATENCY_HISTOGRAM_MIN_VALUE, // Minimum value
CONFIG_LATENCY_HISTOGRAM_MAX_VALUE, // Maximum value
config.precision, // Number of significant figures
&config.latency_histogram); // Pointer to initialise
hdr_init(CONFIG_LATENCY_HISTOGRAM_MIN_VALUE, // Minimum value
CONFIG_LATENCY_HISTOGRAM_INSTANT_MAX_VALUE, // Maximum value
config.precision, // Number of significant figures
&config.current_sec_latency_histogram); // Pointer to initialise
if (config.num_threads) initBenchmarkThreads();
int thread_id = config.num_threads > 0 ? 0 : -1;
c = createClient(cmd, len, NULL, thread_id);
createMissingClients(c);
config.start = mstime();
if (!config.num_threads)
aeMain(config.el);
else
startBenchmarkThreads();
config.totlatency = mstime() - config.start;
showLatencyReport();
freeAllClients();
if (config.threads) freeBenchmarkThreads();
if (config.current_sec_latency_histogram) hdr_close(config.current_sec_latency_histogram);
if (config.latency_histogram) hdr_close(config.latency_histogram);