forked from shaygalon/memcache-perf
-
Notifications
You must be signed in to change notification settings - Fork 0
/
mcperf.cc
1621 lines (1361 loc) · 47.9 KB
/
mcperf.cc
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
#include <arpa/inet.h>
#include <assert.h>
#include <pthread.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/socket.h>
#include <sys/types.h>
#include <time.h>
#include <unistd.h>
#include <queue>
#include <string>
#include <vector>
#include <event2/buffer.h>
#include <event2/bufferevent.h>
#include <event2/dns.h>
#include <event2/event.h>
#include <event2/thread.h>
#include <event2/util.h>
#include "config.h"
#ifdef HAVE_LIBZMQ
#include <zmq.hpp>
#endif
#include "AdaptiveSampler.h"
#include "AgentStats.h"
#ifndef HAVE_PTHREAD_BARRIER_INIT
#include "barrier.h"
#endif
#include "cmdline.h"
#include "Connection.h"
#include "ConnectionOptions.h"
#include "log.h"
#include "mcperf.h"
#include "util.h"
#include "cpu_stat_thread.h"
#define MIN(a,b) ((a) < (b) ? (a) : (b))
using namespace std;
gengetopt_args_info args;
char random_char[2 * 1024 * 1024]; // Buffer used to generate random values.
#ifdef HAVE_LIBZMQ
vector<zmq::socket_t*> agent_sockets;
zmq::context_t context(1);
#endif
struct thread_data {
const vector<string> *servers;
options_t *options;
bool master; // Thread #0, not to be confused with agent master.
#ifdef HAVE_LIBZMQ
zmq::socket_t *socket;
#endif
};
// struct evdns_base *evdns;
pthread_barrier_t barrier;
double boot_time;
void init_random_stuff();
void go(const vector<string> &servers, options_t &options,
ConnectionStats &stats
#ifdef HAVE_LIBZMQ
, zmq::socket_t* socket = NULL
#endif
);
char const *command_string[] = {
"start_trace",
"stop_trace",
"shutdown"
};
void do_mcperf(const vector<string> &servers, options_t &options,
ConnectionStats &stats, bool master = true
#ifdef HAVE_LIBZMQ
, zmq::socket_t* socket = NULL
#endif
);
void args_to_options(options_t* options);
void* thread_main(void *arg);
#ifdef HAVE_LIBZMQ
static unsigned int max_poll_time=0;
static unsigned int poll_interval_s=1;
//static long poll_interval_ns=100000;
void setup_socket_timers() {
if (args.poll_freq_given)
poll_interval_s=args.poll_freq_arg;
if (args.poll_max_given)
max_poll_time=args.poll_max_arg;
}
#if defined(ZMQ_NOBLOCK)
int noblock_flag=ZMQ_NOBLOCK;
#elif defined(ZMQ_DONTWAIT)
int noblock_flag=ZMQ_DONTWAIT;
#endif
void tokenize(const string& str,
vector<string>& tokens,
const string& delimiters = ",")
{
// Skip delimiters at beginning.
string::size_type lastPos = str.find_first_not_of(delimiters, 0);
// Find first "non-delimiter".
string::size_type pos = str.find_first_of(delimiters, lastPos);
while (string::npos != pos || string::npos != lastPos)
{
// Found a token, add it to the vector.
tokens.push_back(str.substr(lastPos, pos - lastPos));
// Skip delimiters. Note the "not_of"
lastPos = str.find_first_not_of(delimiters, pos);
// Find next "non-delimiter"
pos = str.find_first_of(delimiters, lastPos);
}
}
void deTokenize(string& str,
const vector<string>& tokens,
const string& delimiter = ",")
{
vector<string>::const_iterator it;
//collect all tokens to a single string
it=tokens.begin();
str=*it;
it++;
for ( ;
it != tokens.end() ;
it++)
{
str += delimiter;
str += *it;
}
}
bool poll_recv(zmq::socket_t &socket, zmq::message_t *msg) {
unsigned int timer=max_poll_time;
bool status=false;
D("- recv");
//if no timeout, just block
if (timer == 0) {
return socket.recv(msg);
}
//otherwise, recv non blocking until timeout passed
int itid=0;
struct timespec st;
st.tv_sec=poll_interval_s;
st.tv_nsec=0;
do {
itid++;
status=socket.recv(msg,noblock_flag);
if (status)
return status;
// if failed, check errno for fail reason
if (zmq_errno () == EAGAIN) {
// if failed just bcs msg non blocking, try again after sleeping for poll interval
nanosleep(&st,NULL);
timer-=poll_interval_s;
} else {
W("ERROR in socket! [%d]",zmq_errno());
break;
}
if ((itid & 0xff) == 0) {
V("Socket recv multi iterate...");//TODO:DBG
}
} while (timer > poll_interval_s);
W("Failed to recv within requested time limit. Aborting recv.");
return false;
}
bool poll_send(zmq::socket_t &socket, zmq::message_t &msg) {
unsigned int timer=max_poll_time;
bool status=false;
D("- send");
//if no timeout, just block
#if 1
return socket.send(msg);
#else
if (timer == 0) {
return socket.send(msg);
}
//otherwise, recv non blocking until timeout passed
do {
status=socket.send(msg,noblock_flag);
if (status == 0)
return status;
// if failed, check errno for fail reason
if (zmq_errno () == EAGAIN) {
// if failed just bcs msg non blocking, try again after sleeping for poll interval
usleep(poll_interval_us);
timer-=poll_interval_us;
} else {
W("ERROR in socket! [%d]",zmq_errno());
break;
}
} while (timer > poll_interval_us);
W("Failed to send within requested time limit. Aborting send.");
return false;
#endif
}
static std::string s_recv (zmq::socket_t &socket) {
zmq::message_t message;
//socket.recv(&message);
bool status=poll_recv(socket,&message);
if (status)
return std::string(static_cast<char*>(message.data()), message.size());
else
return std::string("FAIL-RECV");
}
// Convert string to 0MQ string and send to socket
static bool s_send (zmq::socket_t &socket, const std::string &string) {
zmq::message_t message(string.size());
memcpy(message.data(), string.data(), string.size());
//return socket.send(message);
return poll_send(socket,message);
}
/*
* Agent protocol
*
* PREPARATION PHASE
*
* 1. Master -> Agent: options_t
*
* options_t contains most of the information needed to drive the
* client, including the aggregate QPS that has been requested.
* However, neither the master nor the agent know at this point how
* many total connections will be made to the memcached server.
*
* 2. Agent -> Master: int num = (--threads) * (--lambda_mul)
*
* The agent sends a number to the master indicating how many threads
* this mcperf agent will spawn, and a mutiplier that weights how
* many QPS this agent's connections will send relative to unweighted
* connections (i.e. we can request that a purely load-generating
* agent or an agent on a really fast network connection be more
* aggressive than other agents or the master).
*
* 3. Master -> Agent: lambda_denom
*
* The master aggregates all of the numbers collected in (2) and
* computes a global "lambda_denom". Which is essentially a count of
* the total number of Connections across all mcperf instances,
* weighted by lambda_mul if necessary. It broadcasts this number to
* all agents.
*
* Each instance of mcperf at this point adjusts the lambda in
* options_t sent in (1) to account for lambda_denom. Note that
* lambda_mul is specific to each instance of mcperf
* (i.e. --lambda_mul X) and not sent as part of options_t.
*
* lambda = qps / lambda_denom * args.lambda_mul;
*
* RUN PHASE
*
* After the PREP phase completes, everyone executes do_mcperf().
* All clients spawn threads, open connections, load the DB, and wait
* for all connections to become IDLE. Following that, they
* synchronize and finally do the heavy lifting.
*
* [IF WARMUP] -1: Master <-> Agent: Synchronize
* [IF WARMUP] 0: Everyone: RUN for options.warmup seconds.
* 1. Master <-> Agent: Synchronize
* 2. Everyone: RUN for options.time seconds.
* 3. Master -> Agent: Dummy message
* 4. Agent -> Master: Send AgentStats [w/ RX/TX bytes, # gets/sets]
*
* The master then aggregates AgentStats across all agents with its
* own ConnectionStats to compute overall statistics.
*/
void agent() {
zmq::context_t context(1);
zmq::socket_t socket(context, ZMQ_REP);
socket.bind((string("tcp://*:")+string(args.agent_port_arg)).c_str());
int lid=0;
while (true) {
zmq::message_t request;
socket.recv(&request);
lid++;
zmq::message_t num(sizeof(int));
*((int *) num.data()) = args.threads_arg * args.lambda_mul_arg;
socket.send(num);
V("sent num %d",lid);
options_t options;
memcpy(&options, request.data(), sizeof(options));
V("Got options: %d %s",options.connections,options.loadonly ? "loadonly" : options.noload ? "noload" : "");
//get a string containing the servers, and parse it to extract all servers
string server_opt=s_recv(socket);
vector<string> servers;
tokenize(server_opt,servers);
s_send(socket, "ack");
V("sent ack");
vector<string>::iterator i;
for (i= servers.begin(); i!=servers.end(); i++) {
V("Got server = %s", i->c_str());
}
options.threads = args.threads_arg;
socket.recv(&request);
options.lambda_denom = *((int *) request.data());
s_send(socket, "THANKS");
V("sent tnx");
// V("AGENT SLEEPS"); sleep(1);
options.lambda = (double) options.qps / options.lambda_denom * args.lambda_mul_arg;
V("lambda_denom = %d, lambda = %f, qps = %d",
options.lambda_denom, options.lambda, options.qps);
// if (options.threads > 1)
pthread_barrier_init(&barrier, NULL, options.threads);
ConnectionStats stats = ConnectionStats();
V("launching go");
go(servers, options, stats, &socket);
V("Done run.");
AgentStats as;
as.rx_bytes = stats.rx_bytes;
as.tx_bytes = stats.tx_bytes;
as.gets = stats.gets;
as.sets = stats.sets;
as.get_misses = stats.get_misses;
as.start = stats.start;
as.stop = stats.stop;
as.skips = stats.skips;
#ifdef LOGSAMPLER_BINS
for (int i=0; i<LOGSAMPLER_BINS; i++)
as.get_bins[i]=stats.get_sampler.bins[i];
as.get_sum = stats.get_sampler.sum;
as.get_sum_sq = stats.get_sampler.sum_sq;
#endif
string req = s_recv(socket);
V("req = %s", req.c_str());
request.rebuild(sizeof(as));
memcpy(request.data(), &as, sizeof(as));
socket.send(request);
V("send = %s", req.c_str());
if (log_level > DEBUG) {
stats.print_header(false);
printf(" QPS\n");
stats.print_stats("read", stats.get_sampler,false);
printf(" %8.1f\n", stats.get_qps());
}
}
}
void prep_agent(const vector<string>& servers, options_t& options) {
int sum = options.lambda_denom;
int aid=0;
if (args.measure_connections_given)
sum = args.measure_connections_arg * options.server_given * options.threads;
int master_sum = sum;
if (args.measure_qps_given) {
sum = 0;
if (options.qps) options.qps -= args.measure_qps_arg;
}
vector<zmq::socket_t*>::iterator its;
for ( its=agent_sockets.begin(); its!=agent_sockets.end(); its++ ) {
zmq::socket_t *s=*its;
zmq::message_t message(sizeof(options_t));
bool status;
aid++;
V("Agent %d prep ", aid);
memcpy((void *) message.data(), &options, sizeof(options_t));
status=poll_send(*s,message);
D("Agent %d prep send = %s", aid, status?"true":"false");
zmq::message_t rep;
status=poll_recv(*s,&rep);
D("Agent %d prep recv= %s", aid, status?"true":"false");
if (!status) { // problem communicating with this agent, don't use it!
W("Agent failure detected, skip agent %d!",aid);
its=agent_sockets.erase(its); // remove from list of active agents
delete(s);
its--; // adjust the iterator since we did not iterate over the next agent
continue;
}
unsigned int num = *((int *) rep.data());
sum += options.connections * (options.roundrobin ?
(servers.size() > num ? servers.size() : num) :
(servers.size() * num));
int itid=0;
//collect all servers to a single msg
string all_servers;
deTokenize(all_servers,servers);
//send servers msg to agent and wait for ack
s_send(*s, all_servers);
string response = s_recv(*s);
if (response.compare("FAIL-RECV") == 0) {
itid=-1;
}
// in case communication with agent broke down, remove from active list
if (itid<0) {
W("Agent failure detected, skip agent %d!",aid);
its=agent_sockets.erase(its); // remove from list of active agents
delete(s);
its--; // adjust the iterator since we did not iterate over the next agent
}
}
// Adjust options_t according to --measure_* arguments.
options.lambda_denom = sum;
options.lambda = (double) options.qps / options.lambda_denom *
args.lambda_mul_arg;
V("lambda_denom = %d", sum);
if (args.measure_qps_given) {
double master_lambda = (double) args.measure_qps_arg / master_sum;
if (options.qps && master_lambda > options.lambda)
V("warning: master_lambda (%f) > options.lambda (%f)",
master_lambda, options.lambda);
options.lambda = master_lambda;
}
if (args.measure_depth_given) options.depth = args.measure_depth_arg;
for ( its=agent_sockets.begin(); its!=agent_sockets.end(); its++ ) {
zmq::socket_t *s=*its;
zmq::message_t message(sizeof(sum));
*((int *) message.data()) = sum;
poll_send(*s,message);
string rep = s_recv(*s);
if (rep.compare("FAIL-RECV") == 0) {
W("Agent failure detected, skip agent %d!",aid);
its=agent_sockets.erase(its); // remove from list of active agents
delete(s);
its--; // adjust the iterator since we did not iterate over the next agent
}
}
// Master sleeps here to give agents a chance to connect to
// memcached server before the master, so that the master is never
// the very first set of connections. Is this reasonable or
// necessary? Most probably not.
V("MASTER SLEEPS"); sleep_time(1.5);
}
void finish_agent(ConnectionStats &stats) {
int aid=0;
//for (auto s: agent_sockets) {
vector<zmq::socket_t*>::iterator its;
for ( its=agent_sockets.begin(); its!=agent_sockets.end(); its++ ) {
zmq::socket_t *s=*its;
aid++;
bool status;
status=s_send(*s, "stats");
D("Agent %d finish send = %s", aid, status?"true":"false");
AgentStats as;
zmq::message_t message;
status=poll_recv(*s,&message);
D("Agent %d finish recv = %s", aid, status?"true":"false");
memcpy(&as, message.data(), sizeof(as));
stats.accumulate(as);
}
}
/*
* This synchronization routine is ridiculous because the master only
* has a ZMQ_REQ socket to the agents, but it needs to wait for a
* message from each agent before it releases them. In order to get
* the ZMQ socket into a state where it'll allow the agent to send it
* a message, it must first send a message ("sync_req"). In order to
* not leave the socket dangling with an incomplete transaction, the
* agent must send a reply ("ack").
*
* Without this stupid complication it would be:
*
* For each agent:
* Agent -> Master: sync
* For each agent:
* Master -> Agent: proceed
*
* In this way, all agents must arrive at the barrier and the master
* must receive a message from each of them before it continues. It
* then broadcasts the message to proceed, which reasonably limits
* skew.
*/
int sync_agent(zmq::socket_t* socket) {
V("agent: synchronizing");
int aid=0;
int errors=0;
if (args.agent_given) {
vector<zmq::socket_t*>::iterator its;
for ( its=agent_sockets.begin(); its!=agent_sockets.end(); its++ ) {
zmq::socket_t *s=*its;
s_send(*s, "sync_req");
D("Sent sync_req to agent %d",++aid);
}
aid=0;
/* The real sync */
for ( its=agent_sockets.begin(); its!=agent_sockets.end(); its++ ) {
zmq::socket_t *s=*its;aid++;
string rep = s_recv(*s);
if (rep.compare(string("sync")) != 0) {
W("sync_agent[M]: out of sync [1] for agent %d expected sync got %s",aid,rep.c_str());
errors++;
if (rep.compare("FAIL-RECV") == 0) {
W("Agent failure detected, skip agent %d!",aid);
its=agent_sockets.erase(its); // remove from list of active agents
delete(s);
its--; // adjust the iterator since we did not iterate over the next agent
}
}
}
aid=0;
for ( its=agent_sockets.begin(); its!=agent_sockets.end(); its++ ) {
zmq::socket_t *s=*its;
s_send(*s, "proceed");
D("Sent proceed to agent %d",++aid);
}
/* End sync */
aid=0;
for ( its=agent_sockets.begin(); its!=agent_sockets.end(); its++ ) {
zmq::socket_t *s=*its;aid++;
string rep = s_recv(*s);
if (rep.compare(string("ack")) != 0) {
W("sync_agent[M]: out of sync [2] for agent %d expected ack got %s",aid,rep.c_str());
errors++;
if (rep.compare("FAIL-RECV") == 0) {
W("Agent failure detected, skip agent %d!",aid);
its=agent_sockets.erase(its); // remove from list of active agents
delete(s);
its--; // adjust the iterator since we did not iterate over the next agent
}
}
}
} else if (args.agentmode_given) {
string rep = s_recv(*socket);
if (rep.compare(string("sync_req")) != 0) {
W("sync_agent[A]: out of sync [1] got %s expected sync_req",rep.c_str());
errors++;
}
/* The real sync */
s_send(*socket, "sync");
rep = s_recv(*socket);
if (rep.compare(string("proceed")) != 0) {
W("sync_agent[A]: out of sync [2] got %s expected proceed",rep.c_str());
errors++;
}
/* End sync */
s_send(*socket, "ack");
}
V("agent: synchronized with %d errors",errors);
return errors;
}
#endif
string name_to_ipaddr(string host, int addport=1) {
char *s_copy = new char[host.length() + 1];
strcpy(s_copy, host.c_str());
char *saveptr = NULL; // For reentrant strtok().
char *h_ptr = strtok_r(s_copy, "@", &saveptr);
char *p_ptr = strtok_r(NULL, "@", &saveptr);
char ipaddr[16];
if (h_ptr == NULL)
DIE("strtok(.., \":\") failed to parse %s", host.c_str());
string hostname = h_ptr;
string port = "11211";
if (p_ptr) port = p_ptr;
struct evutil_addrinfo hints;
struct evutil_addrinfo *answer = NULL;
int err;
/* Build the hints to tell getaddrinfo how to act. */
memset(&hints, 0, sizeof(hints));
hints.ai_family = AF_UNSPEC; /* v4 or v6 is fine. */
hints.ai_socktype = SOCK_STREAM;
hints.ai_protocol = IPPROTO_TCP; /* We want a TCP socket */
/* Only return addresses we can use. */
hints.ai_flags = EVUTIL_AI_ADDRCONFIG;
/* Look up the hostname. */
err = evutil_getaddrinfo(h_ptr, NULL, &hints, &answer);
if (err < 0) {
DIE("Error while resolving '%s': %s",
host.c_str(), evutil_gai_strerror(err));
}
if (answer == NULL) DIE("No DNS answer.");
void *ptr = NULL;
switch (answer->ai_family) {
case AF_INET:
ptr = &((struct sockaddr_in *) answer->ai_addr)->sin_addr;
break;
case AF_INET6:
ptr = &((struct sockaddr_in6 *) answer->ai_addr)->sin6_addr;
break;
}
inet_ntop (answer->ai_family, ptr, ipaddr, 16);
D("Resolved %s to %s", h_ptr, (string(ipaddr) + ":" + string(port)).c_str());
delete[] s_copy;
if (addport)
return string(ipaddr) + "@" + string(port);
else
return string(ipaddr);
}
int main(int argc, char **argv) {
if (cmdline_parser(argc, argv, &args) != 0) exit(-1);
for (unsigned int i = 0; i < args.verbose_given; i++)
log_level = (log_level_t) ((int) log_level - 1);
if (args.quiet_given) log_level = QUIET;
if (args.depth_arg < 1) DIE("--depth must be >= 1");
if (args.qps_arg < 0) DIE("--qps must be >= 0");
if (args.update_arg < 0.0 || args.update_arg > 1.0)
DIE("--update must be >= 0.0 and <= 1.0");
if (args.time_arg < 1) DIE("--time must be >= 1");
if (args.connections_arg < 1 || args.connections_arg > MAXIMUM_CONNECTIONS)
DIE("--connections must be between [1,%d]", MAXIMUM_CONNECTIONS);
if (!args.server_given && !args.agentmode_given)
DIE("--server or --agentmode must be specified.");
// TODO: Discover peers, share arguments.
init_random_stuff();
boot_time = get_time();
setvbuf(stdout, NULL, _IONBF, 0);
// struct event_base *base;
// if ((base = event_base_new()) == NULL) DIE("event_base_new() fail");
// evthread_use_pthreads();
// if ((evdns = evdns_base_new(base, 1)) == 0) DIE("evdns");
options_t options;
bzero(&options, sizeof(options_t));
args_to_options(&options);
#ifdef HAVE_LIBZMQ
if (args.agentmode_given) {
agent();
return 0;
} else if (args.agent_given) {
int status;
setup_socket_timers();
for (unsigned int i = 0; i < args.agent_given; i++) {
zmq::socket_t *s = new zmq::socket_t(context, ZMQ_REQ);
if (s==NULL) {
DIE("Could not open socket! %s",zmq_strerror(zmq_errno()));
}
// Support multiple agents on the same host.
// =========================================
// The -a host[@agent_port] option extends the agent
// connection string to allow multiple agents running
// on the same host with different ports. When agent_port
// is part of agent option, the agent_port option is ignored.
// It is still desireable to use agent_port option in the agent mode
// when starting agent to listen on a specific zmq port.
string aname=args.agent_arg[i];
string host;
size_t port_pos = aname.find("@");
if (port_pos != string::npos) {
string hstr = aname.substr(0, port_pos);
string ip_addr = name_to_ipaddr(hstr, 0);
string port = aname.substr(port_pos + 1, aname.length());
host = string("tcp://") + ip_addr + string(":") + port;
} else {
host = string("tcp://") + name_to_ipaddr(args.agent_arg[i],0) +
string(":") + string(args.agent_port_arg);
}
D("Add %s as agent\n",host.c_str());
// setup socket to handle as many connections as we will need
int nconns = args.measure_connections_given ? args.measure_connections_arg :
options.connections;
int nthreads = args.threads_given ? args.threads_arg :
options.threads;
int total_conn = 2 * (nconns+1) * (nthreads+1);
if (total_conn<100)
total_conn=100;
s->setsockopt(ZMQ_BACKLOG,&total_conn,sizeof(total_conn));
int linger=10000;
s->setsockopt(ZMQ_LINGER,&linger,sizeof(linger));
//then connect
try {
s->connect(host.c_str());
agent_sockets.push_back(s);
} catch (...) {
DIE("Agent not available at %s! Please make sure that the agent process is running, and the ports are open.\n",host.c_str());
}
//s->connect(host.c_str());
//agent_sockets.push_back(s);
}
}
#endif
pthread_barrier_init(&barrier, NULL, options.threads);
cpu_info_t cpustat;
pthread_create(&(cpustat.tid), 0, cpu_stat_thread, &cpustat);
vector<string> servers;
for (unsigned int s = 0; s < args.server_given; s++) {
string sname=args.server_arg[s];
size_t port_pos=sname.find("@");
if ((port_pos != string::npos) &&
(sname.find("-", port_pos) != string::npos)) { //parse range of ports on machine in case more then one instance is running
char *saveptr = NULL; // For reentrant strtok().
char *cname=strdup(sname.c_str());
char *h_ptr = strtok_r(cname, "@", &saveptr);
char *start_port = strtok_r(NULL, "-", &saveptr);
char *end_port = strtok_r(NULL, "-", &saveptr);
string ip_addr=name_to_ipaddr(h_ptr,0);
int sp=atoi(start_port);
int ep=atoi(end_port);
while (sp <= ep) {
char buf[48];
string s_instance=ip_addr + ":" + string( std::to_string(sp));
servers.push_back(s_instance);
sp++;
}
} else {
servers.push_back(name_to_ipaddr(string(args.server_arg[s])));
}
}
ConnectionStats stats;
if (args.plot_all_given)
stats.plotall=true;
double peak_qps = 0.0;
bool avgseek=false;
if (args.search_given) {
char *n_ptr = strtok(args.search_arg, ":");
char *x_ptr = strtok(NULL, ":");
if (n_ptr == NULL || x_ptr == NULL) DIE("Invalid --search argument");
if (strstr("avg",n_ptr)!=NULL) {
avgseek=true;
}
int n = atoi(n_ptr);
int x = atoi(x_ptr);
if (avgseek) I("Search-mode. Find QPS @ %dus avg latency.", x, n);
else I("Search-mode. Find QPS @ %dus %dth percentile.", x, n);
//first determine max qps without paying attention to latency.
int high_qps = 10000000;
int low_qps = 1; // 5000;
double nth;
int cur_qps;
go(servers, options, stats);
if (avgseek) nth=stats.get_avg();
else nth = stats.get_nth(n);
peak_qps = stats.get_qps();
high_qps = stats.get_qps();
cur_qps = stats.get_qps();
I("peak qps = % 8d, %s = %.1f", high_qps, n_ptr, nth);
//if latency at peak more then requested, search for requested latency setting to middle point between current and a point with known good latency
//continue searching until the boundary between known good and current is ~5%
if (nth > x) {
while ((high_qps > low_qps * 1.02) && cur_qps > (peak_qps * .01)) {
cur_qps = (high_qps + low_qps) / 2;
args_to_options(&options);
options.qps = cur_qps;
options.lambda = (double) options.qps / (double) options.lambda_denom * args.lambda_mul_arg;
stats = ConnectionStats();
go(servers, options, stats);
if (avgseek) nth=stats.get_avg();
else nth = stats.get_nth(n);
I(". target = % 8d, %s = %.1f, high_qps = %d, low_qps = %d, qps = %.0f", cur_qps,n_ptr,nth, high_qps, low_qps, stats.get_qps());
if (nth > x /*|| cur_qps > stats.get_qps() * 1.05*/) high_qps = cur_qps;
else low_qps = cur_qps;
}
// now if last value found at conversion has latency over the limit, use the last value before latency came within bounds,
// and decrease requested qps by 1% until latency comes within bounds, or we go below 90% of previously found value.
while (nth > x && cur_qps > (peak_qps * .01) && cur_qps > (low_qps * 0.90)) {
cur_qps = cur_qps * 99 / 100;
args_to_options(&options);
options.qps = cur_qps;
options.lambda = (double) options.qps / (double) options.lambda_denom * args.lambda_mul_arg;
stats = ConnectionStats();
go(servers, options, stats);
if (avgseek) nth=stats.get_avg();
else nth = stats.get_nth(n);
I(". target = % 8d, %s = %.1f, high_qps = %d, low_qps = %d, qps = %.0f", cur_qps,n_ptr,nth, high_qps, low_qps, stats.get_qps());
}
}
} else if (args.scan_given) {
char *min_ptr = strtok(args.scan_arg, ":");
char *max_ptr = strtok(NULL, ":");
char *step_ptr = strtok(NULL, ":");
if (min_ptr == NULL || min_ptr == NULL || step_ptr == NULL)
DIE("Invalid --scan argument");
int min = atoi(min_ptr);
int max = atoi(max_ptr);
int step = atoi(step_ptr);
stats.print_header(false);
printf("%8s %8s\n", "QPS", "target");
for (int q = min; q <= max; q += step) {
args_to_options(&options);
options.qps = q;
options.lambda = (double) options.qps / (double) options.lambda_denom * args.lambda_mul_arg;
stats = ConnectionStats();
reset_cpu_stats();
go(servers, options, stats);
D("CPU Usage Stats (avg/min/max): %.2Lf%%,%.2Lf%%,%.2Lf%%\n",cpustat.avg,cpustat.min,cpustat.max);
stats.print_stats("read", stats.get_sampler, false);
printf(" %8.1f", stats.get_qps());
printf(" %8d\n", q);
}
} else {
go(servers, options, stats);
}
if (!args.scan_given && !args.loadonly_given) {
stats.print_header();
stats.print_stats("read", stats.get_sampler, true, true);
stats.print_stats("update", stats.set_sampler);
stats.print_stats("op_q", stats.op_sampler);
float total = (float)(stats.gets + stats.sets);
printf("\nTotal QPS = %.1f (%.0f / %.1fs)\n",
total / (stats.stop - stats.start),
total, stats.stop - stats.start);
if (args.search_given && peak_qps > 0.0)
printf("Peak QPS = %.1f\n", peak_qps);
printf("\n");
printf("Total connections = %d\n", options.connections * options.server_given * options.threads);
printf("Misses = %" PRIu64 " (%.1f%%)\n", stats.get_misses,
(double) stats.get_misses/stats.gets*100);
printf("Skipped TXs = %" PRIu64 " (%.1f%%)\n\n", stats.skips,
(double) stats.skips / total * 100);
printf("RX %10" PRIu64 " bytes : %6.1f MB/s\n",
stats.rx_bytes,
(double) stats.rx_bytes / 1024 / 1024 / (stats.stop - stats.start));
printf("TX %10" PRIu64 " bytes : %6.1f MB/s\n",
stats.tx_bytes,
(double) stats.tx_bytes / 1024 / 1024 / (stats.stop - stats.start));
if (args.save_given) {
printf("Saving latency samples to %s.\n", args.save_arg);
FILE *file;
if ((file = fopen(args.save_arg, "w")) == NULL)
DIE("--save: failed to open %s: %s", args.save_arg, strerror(errno));
std::vector<Operation>::const_iterator i;
for ( i= stats.get_sampler.samples.begin(); i!=stats.get_sampler.samples.end(); i++) {
fprintf(file, "%f %f\n", i->start_time - boot_time, i->time());
}
}
}
stop_cpu_stats();
pthread_join(cpustat.tid,0);
if (!args.loadonly_given)
printf("CPU Usage Stats (avg/min/max): %.2Lf%%,%.2Lf%%,%.2Lf%%\n",cpustat.avg,cpustat.min,cpustat.max);
// if (args.threads_arg > 1)
pthread_barrier_destroy(&barrier);
#ifdef HAVE_LIBZMQ
if (args.agent_given) {
vector<zmq::socket_t*>::iterator its;
for ( its=agent_sockets.begin(); its!=agent_sockets.end(); its++ ) {
zmq::socket_t *s=*its;
//for (auto i: agent_sockets) delete i;
delete s;
}
}
#endif
// evdns_base_free(evdns, 0);
// event_base_free(base);
cmdline_parser_free(&args);
return 0;
}
void go(const vector<string>& servers, options_t& options,
ConnectionStats &stats
#ifdef HAVE_LIBZMQ
, zmq::socket_t* socket
#endif
) {
#ifdef HAVE_LIBZMQ
if (args.agent_given > 0) {
V("agent given");
prep_agent(servers, options);
V("Agent prep done.");
}
#endif
if (options.threads > 1) {
pthread_t pt[options.threads];
struct thread_data td[options.threads];
#ifdef __clang__
vector<string>* ts = static_cast<vector<string>*>(alloca(sizeof(vector<string>) * options.threads));
#else
vector<string> ts[options.threads];
#endif
int current_cpu = -1;
//options.qps/=options.threads;
//options.lambda/=(double)options.threads;
D("Starting %d threads.", options.threads);
for (int t = 0; t < options.threads; t++) {
td[t].options = &options;
#ifdef HAVE_LIBZMQ