aboutsummaryrefslogtreecommitdiffhomepage
path: root/test/cpp/qps/client.h
blob: 73f91eed2d84b83335d3b6587730eb6bdc82d0ff (plain)
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
/*
 *
 * Copyright 2015 gRPC authors.
 *
 * Licensed under the Apache License, Version 2.0 (the "License");
 * you may not use this file except in compliance with the License.
 * You may obtain a copy of the License at
 *
 *     http://www.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 *
 */

#ifndef TEST_QPS_CLIENT_H
#define TEST_QPS_CLIENT_H

#include <stdlib.h>

#include <condition_variable>
#include <mutex>
#include <unordered_map>
#include <vector>

#include <grpc/support/log.h>
#include <grpc/support/time.h>
#include <grpcpp/channel.h>
#include <grpcpp/support/byte_buffer.h>
#include <grpcpp/support/channel_arguments.h>
#include <grpcpp/support/slice.h>

#include "src/proto/grpc/testing/benchmark_service.grpc.pb.h"
#include "src/proto/grpc/testing/payloads.pb.h"

#include "src/core/lib/gpr/env.h"
#include "src/cpp/util/core_stats.h"
#include "test/cpp/qps/histogram.h"
#include "test/cpp/qps/interarrival.h"
#include "test/cpp/qps/qps_worker.h"
#include "test/cpp/qps/server.h"
#include "test/cpp/qps/usage_timer.h"
#include "test/cpp/util/create_test_channel.h"
#include "test/cpp/util/test_credentials_provider.h"

#define INPROC_NAME_PREFIX "qpsinproc:"

namespace grpc {
namespace testing {

template <class RequestType>
class ClientRequestCreator {
 public:
  ClientRequestCreator(RequestType* req, const PayloadConfig&) {
    // this template must be specialized
    // fail with an assertion rather than a compile-time
    // check since these only happen at the beginning anyway
    GPR_ASSERT(false);
  }
};

template <>
class ClientRequestCreator<SimpleRequest> {
 public:
  ClientRequestCreator(SimpleRequest* req,
                       const PayloadConfig& payload_config) {
    if (payload_config.has_bytebuf_params()) {
      GPR_ASSERT(false);  // not appropriate for this specialization
    } else if (payload_config.has_simple_params()) {
      req->set_response_type(grpc::testing::PayloadType::COMPRESSABLE);
      req->set_response_size(payload_config.simple_params().resp_size());
      req->mutable_payload()->set_type(
          grpc::testing::PayloadType::COMPRESSABLE);
      int size = payload_config.simple_params().req_size();
      std::unique_ptr<char[]> body(new char[size]);
      req->mutable_payload()->set_body(body.get(), size);
    } else if (payload_config.has_complex_params()) {
      GPR_ASSERT(false);  // not appropriate for this specialization
    } else {
      // default should be simple proto without payloads
      req->set_response_type(grpc::testing::PayloadType::COMPRESSABLE);
      req->set_response_size(0);
      req->mutable_payload()->set_type(
          grpc::testing::PayloadType::COMPRESSABLE);
    }
  }
};

template <>
class ClientRequestCreator<ByteBuffer> {
 public:
  ClientRequestCreator(ByteBuffer* req, const PayloadConfig& payload_config) {
    if (payload_config.has_bytebuf_params()) {
      size_t req_sz =
          static_cast<size_t>(payload_config.bytebuf_params().req_size());
      std::unique_ptr<char[]> buf(new char[req_sz]);
      memset(buf.get(), 0, req_sz);
      Slice slice(buf.get(), req_sz);
      *req = ByteBuffer(&slice, 1);
    } else {
      GPR_ASSERT(false);  // not appropriate for this specialization
    }
  }
};

class HistogramEntry final {
 public:
  HistogramEntry() : value_used_(false), status_used_(false) {}
  bool value_used() const { return value_used_; }
  double value() const { return value_; }
  void set_value(double v) {
    value_used_ = true;
    value_ = v;
  }
  bool status_used() const { return status_used_; }
  int status() const { return status_; }
  void set_status(int status) {
    status_used_ = true;
    status_ = status;
  }

 private:
  bool value_used_;
  double value_;
  bool status_used_;
  int status_;
};

typedef std::unordered_map<int, int64_t> StatusHistogram;

inline void MergeStatusHistogram(const StatusHistogram& from,
                                 StatusHistogram* to) {
  for (StatusHistogram::const_iterator it = from.begin(); it != from.end();
       ++it) {
    (*to)[it->first] += it->second;
  }
}

class Client {
 public:
  Client()
      : timer_(new UsageTimer),
        interarrival_timer_(),
        started_requests_(false),
        last_reset_poll_count_(0) {
    gpr_event_init(&start_requests_);
  }
  virtual ~Client() {}

  ClientStats Mark(bool reset) {
    Histogram latencies;
    StatusHistogram statuses;
    UsageTimer::Result timer_result;

    MaybeStartRequests();

    int cur_poll_count = GetPollCount();
    int poll_count = cur_poll_count - last_reset_poll_count_;
    if (reset) {
      std::vector<Histogram> to_merge(threads_.size());
      std::vector<StatusHistogram> to_merge_status(threads_.size());

      for (size_t i = 0; i < threads_.size(); i++) {
        threads_[i]->BeginSwap(&to_merge[i], &to_merge_status[i]);
      }
      std::unique_ptr<UsageTimer> timer(new UsageTimer);
      timer_.swap(timer);
      for (size_t i = 0; i < threads_.size(); i++) {
        latencies.Merge(to_merge[i]);
        MergeStatusHistogram(to_merge_status[i], &statuses);
      }
      timer_result = timer->Mark();
      last_reset_poll_count_ = cur_poll_count;
    } else {
      // merge snapshots of each thread histogram
      for (size_t i = 0; i < threads_.size(); i++) {
        threads_[i]->MergeStatsInto(&latencies, &statuses);
      }
      timer_result = timer_->Mark();
    }

    // Print the median latency per interval for one thread.
    // If the number of warmup seconds is x, then the first x + 1 numbers in the
    // vector are from the warmup period and should be discarded.
    if (median_latency_collection_interval_seconds_ > 0) {
      std::vector<double> medians_per_interval =
          threads_[0]->GetMedianPerIntervalList();
      gpr_log(GPR_INFO, "Num threads: %ld", threads_.size());
      gpr_log(GPR_INFO, "Number of medians: %ld", medians_per_interval.size());
      for (size_t j = 0; j < medians_per_interval.size(); j++) {
        gpr_log(GPR_INFO, "%f", medians_per_interval[j]);
      }
    }

    grpc_stats_data core_stats;
    grpc_stats_collect(&core_stats);

    ClientStats stats;
    latencies.FillProto(stats.mutable_latencies());
    for (StatusHistogram::const_iterator it = statuses.begin();
         it != statuses.end(); ++it) {
      RequestResultCount* rrc = stats.add_request_results();
      rrc->set_status_code(it->first);
      rrc->set_count(it->second);
    }
    stats.set_time_elapsed(timer_result.wall);
    stats.set_time_system(timer_result.system);
    stats.set_time_user(timer_result.user);
    stats.set_cq_poll_count(poll_count);
    CoreStatsToProto(core_stats, stats.mutable_core_stats());
    return stats;
  }

  // Must call AwaitThreadsCompletion before destructor to avoid a race
  // between destructor and invocation of virtual ThreadFunc
  void AwaitThreadsCompletion() {
    gpr_atm_rel_store(&thread_pool_done_, static_cast<gpr_atm>(true));
    DestroyMultithreading();
    std::unique_lock<std::mutex> g(thread_completion_mu_);
    while (threads_remaining_ != 0) {
      threads_complete_.wait(g);
    }
  }

  // Returns the interval (in seconds) between collecting latency medians. If 0,
  // no periodic median latencies will be collected.
  double GetLatencyCollectionIntervalInSeconds() {
    return median_latency_collection_interval_seconds_;
  }

  virtual int GetPollCount() {
    // For sync client.
    return 0;
  }

 protected:
  bool closed_loop_;
  gpr_atm thread_pool_done_;
  double median_latency_collection_interval_seconds_;  // In seconds

  void StartThreads(size_t num_threads) {
    gpr_atm_rel_store(&thread_pool_done_, static_cast<gpr_atm>(false));
    threads_remaining_ = num_threads;
    for (size_t i = 0; i < num_threads; i++) {
      threads_.emplace_back(new Thread(this, i));
    }
  }

  void EndThreads() {
    MaybeStartRequests();
    threads_.clear();
  }

  virtual void DestroyMultithreading() = 0;

  void SetupLoadTest(const ClientConfig& config, size_t num_threads) {
    // Set up the load distribution based on the number of threads
    const auto& load = config.load_params();

    std::unique_ptr<RandomDistInterface> random_dist;
    switch (load.load_case()) {
      case LoadParams::kClosedLoop:
        // Closed-loop doesn't use random dist at all
        break;
      case LoadParams::kPoisson:
        random_dist.reset(
            new ExpDist(load.poisson().offered_load() / num_threads));
        break;
      default:
        GPR_ASSERT(false);
    }

    // Set closed_loop_ based on whether or not random_dist is set
    if (!random_dist) {
      closed_loop_ = true;
    } else {
      closed_loop_ = false;
      // set up interarrival timer according to random dist
      interarrival_timer_.init(*random_dist, num_threads);
      const auto now = gpr_now(GPR_CLOCK_MONOTONIC);
      for (size_t i = 0; i < num_threads; i++) {
        next_time_.push_back(gpr_time_add(
            now,
            gpr_time_from_nanos(interarrival_timer_.next(i), GPR_TIMESPAN)));
      }
    }
  }

  gpr_timespec NextIssueTime(int thread_idx) {
    const gpr_timespec result = next_time_[thread_idx];
    next_time_[thread_idx] =
        gpr_time_add(next_time_[thread_idx],
                     gpr_time_from_nanos(interarrival_timer_.next(thread_idx),
                                         GPR_TIMESPAN));
    return result;
  }
  std::function<gpr_timespec()> NextIssuer(int thread_idx) {
    return closed_loop_ ? std::function<gpr_timespec()>()
                        : std::bind(&Client::NextIssueTime, this, thread_idx);
  }

  class Thread {
   public:
    Thread(Client* client, size_t idx)
        : client_(client), idx_(idx), impl_(&Thread::ThreadFunc, this) {}

    ~Thread() { impl_.join(); }

    void BeginSwap(Histogram* n, StatusHistogram* s) {
      std::lock_guard<std::mutex> g(mu_);
      n->Swap(&histogram_);
      s->swap(statuses_);
    }

    void MergeStatsInto(Histogram* hist, StatusHistogram* s) {
      std::unique_lock<std::mutex> g(mu_);
      hist->Merge(histogram_);
      MergeStatusHistogram(statuses_, s);
    }

    std::vector<double> GetMedianPerIntervalList() {
      return medians_each_interval_list_;
    }

    void UpdateHistogram(HistogramEntry* entry) {
      std::lock_guard<std::mutex> g(mu_);
      if (entry->value_used()) {
        histogram_.Add(entry->value());
        if (client_->GetLatencyCollectionIntervalInSeconds() > 0) {
          histogram_per_interval_.Add(entry->value());
          double now = UsageTimer::Now();
          if ((now - interval_start_time_) >=
              client_->GetLatencyCollectionIntervalInSeconds()) {
            // Record the median latency of requests from the last interval.
            // Divide by 1e3 to get microseconds.
            medians_each_interval_list_.push_back(
                histogram_per_interval_.Percentile(50) / 1e3);
            histogram_per_interval_.Reset();
            interval_start_time_ = now;
          }
        }
      }
      if (entry->status_used()) {
        statuses_[entry->status()]++;
      }
    }

   private:
    Thread(const Thread&);
    Thread& operator=(const Thread&);

    void ThreadFunc() {
      int wait_loop = 0;
      while (!gpr_event_wait(
          &client_->start_requests_,
          gpr_time_add(gpr_now(GPR_CLOCK_REALTIME),
                       gpr_time_from_seconds(20, GPR_TIMESPAN)))) {
        gpr_log(GPR_INFO, "%" PRIdPTR ": Waiting for benchmark to start (%d)",
                idx_, wait_loop);
        wait_loop++;
      }

      client_->ThreadFunc(idx_, this);
      client_->CompleteThread();
    }

    std::mutex mu_;
    Histogram histogram_;
    StatusHistogram statuses_;
    Client* client_;
    const size_t idx_;
    std::thread impl_;
    // The following are used only if
    // median_latency_collection_interval_seconds_ is greater than 0
    Histogram histogram_per_interval_;
    std::vector<double> medians_each_interval_list_;
    double interval_start_time_;
  };

  bool ThreadCompleted() {
    return static_cast<bool>(gpr_atm_acq_load(&thread_pool_done_));
  }

  virtual void ThreadFunc(size_t thread_idx, Client::Thread* t) = 0;

  std::vector<std::unique_ptr<Thread>> threads_;
  std::unique_ptr<UsageTimer> timer_;

  InterarrivalTimer interarrival_timer_;
  std::vector<gpr_timespec> next_time_;

  std::mutex thread_completion_mu_;
  size_t threads_remaining_;
  std::condition_variable threads_complete_;

  gpr_event start_requests_;
  bool started_requests_;

  int last_reset_poll_count_;

  void MaybeStartRequests() {
    if (!started_requests_) {
      started_requests_ = true;
      gpr_event_set(&start_requests_, (void*)1);
    }
  }

  void CompleteThread() {
    std::lock_guard<std::mutex> g(thread_completion_mu_);
    threads_remaining_--;
    if (threads_remaining_ == 0) {
      threads_complete_.notify_all();
    }
  }
};

template <class StubType, class RequestType>
class ClientImpl : public Client {
 public:
  ClientImpl(const ClientConfig& config,
             std::function<std::unique_ptr<StubType>(std::shared_ptr<Channel>)>
                 create_stub)
      : cores_(gpr_cpu_num_cores()), create_stub_(create_stub) {
    for (int i = 0; i < config.client_channels(); i++) {
      channels_.emplace_back(
          config.server_targets(i % config.server_targets_size()), config,
          create_stub_, i);
    }
    WaitForChannelsToConnect();
    median_latency_collection_interval_seconds_ =
        config.median_latency_collection_interval_millis() / 1e3;
    ClientRequestCreator<RequestType> create_req(&request_,
                                                 config.payload_config());
  }
  virtual ~ClientImpl() {}

  void WaitForChannelsToConnect() {
    int connect_deadline_seconds = 10;
    /* Allow optionally overriding connect_deadline in order
     * to deal with benchmark environments in which the server
     * can take a long time to become ready. */
    char* channel_connect_timeout_str =
        gpr_getenv("QPS_WORKER_CHANNEL_CONNECT_TIMEOUT");
    if (channel_connect_timeout_str != nullptr &&
        strcmp(channel_connect_timeout_str, "") != 0) {
      connect_deadline_seconds = atoi(channel_connect_timeout_str);
    }
    gpr_log(GPR_INFO,
            "Waiting for up to %d seconds for all channels to connect",
            connect_deadline_seconds);
    gpr_free(channel_connect_timeout_str);
    gpr_timespec connect_deadline = gpr_time_add(
        gpr_now(GPR_CLOCK_REALTIME),
        gpr_time_from_seconds(connect_deadline_seconds, GPR_TIMESPAN));
    CompletionQueue cq;
    size_t num_remaining = 0;
    for (auto& c : channels_) {
      if (!c.is_inproc()) {
        Channel* channel = c.get_channel();
        grpc_connectivity_state last_observed = channel->GetState(true);
        if (last_observed == GRPC_CHANNEL_READY) {
          gpr_log(GPR_INFO, "Channel %p connected!", channel);
        } else {
          num_remaining++;
          channel->NotifyOnStateChange(last_observed, connect_deadline, &cq,
                                       channel);
        }
      }
    }
    while (num_remaining > 0) {
      bool ok = false;
      void* tag = nullptr;
      cq.Next(&tag, &ok);
      Channel* channel = static_cast<Channel*>(tag);
      if (!ok) {
        gpr_log(GPR_ERROR, "Channel %p failed to connect within the deadline",
                channel);
        abort();
      } else {
        grpc_connectivity_state last_observed = channel->GetState(true);
        if (last_observed == GRPC_CHANNEL_READY) {
          gpr_log(GPR_INFO, "Channel %p connected!", channel);
          num_remaining--;
        } else {
          channel->NotifyOnStateChange(last_observed, connect_deadline, &cq,
                                       channel);
        }
      }
    }
  }

 protected:
  const int cores_;
  RequestType request_;

  class ClientChannelInfo {
   public:
    ClientChannelInfo(
        const grpc::string& target, const ClientConfig& config,
        std::function<std::unique_ptr<StubType>(std::shared_ptr<Channel>)>
            create_stub,
        int shard) {
      ChannelArguments args;
      args.SetInt("shard_to_ensure_no_subchannel_merges", shard);
      set_channel_args(config, &args);

      grpc::string type;
      if (config.has_security_params() &&
          config.security_params().cred_type().empty()) {
        type = kTlsCredentialsType;
      } else {
        type = config.security_params().cred_type();
      }

      grpc::string inproc_pfx(INPROC_NAME_PREFIX);
      if (target.find(inproc_pfx) != 0) {
        channel_ = CreateTestChannel(
            target, type, config.security_params().server_host_override(),
            !config.security_params().use_test_ca(),
            std::shared_ptr<CallCredentials>(), args);
        gpr_log(GPR_INFO, "Connecting to %s", target.c_str());
        is_inproc_ = false;
      } else {
        grpc::string tgt = target;
        tgt.erase(0, inproc_pfx.length());
        int srv_num = std::stoi(tgt);
        channel_ = (*g_inproc_servers)[srv_num]->InProcessChannel(args);
        is_inproc_ = true;
      }
      stub_ = create_stub(channel_);
    }
    Channel* get_channel() { return channel_.get(); }
    StubType* get_stub() { return stub_.get(); }
    bool is_inproc() { return is_inproc_; }

   private:
    void set_channel_args(const ClientConfig& config, ChannelArguments* args) {
      for (const auto& channel_arg : config.channel_args()) {
        if (channel_arg.value_case() == ChannelArg::kStrValue) {
          args->SetString(channel_arg.name(), channel_arg.str_value());
        } else if (channel_arg.value_case() == ChannelArg::kIntValue) {
          args->SetInt(channel_arg.name(), channel_arg.int_value());
        } else {
          gpr_log(GPR_ERROR, "Empty channel arg value.");
        }
      }
    }

    std::shared_ptr<Channel> channel_;
    std::unique_ptr<StubType> stub_;
    bool is_inproc_;
  };
  std::vector<ClientChannelInfo> channels_;
  std::function<std::unique_ptr<StubType>(const std::shared_ptr<Channel>&)>
      create_stub_;
};

std::unique_ptr<Client> CreateSynchronousClient(const ClientConfig& args);
std::unique_ptr<Client> CreateAsyncClient(const ClientConfig& args);
std::unique_ptr<Client> CreateCallbackClient(const ClientConfig& args);
std::unique_ptr<Client> CreateGenericAsyncStreamingClient(
    const ClientConfig& args);

}  // namespace testing
}  // namespace grpc

#endif