Function thread_worker

tools/cpp_examples/demo-client/src/bert_service.cpp:123–167

}

void thread_worker(PredictorApi* api,
                   int thread_id,
                   int batch_size,
                   int server_concurrency,
                   const std::vector<std::string>& data_list) {
  Request req;
  Response res;
  api->thrd_initialize();
  std::string line;
  int turns = 0;
  while (turns < max_turn) {
    timeval start;
    gettimeofday(&start, NULL);
    api->thrd_clear();
    Predictor* predictor = api->fetch_predictor("bert_service");
    if (!predictor) {
      LOG(ERROR) << "Failed fetch predictor: bert_service";
      return;
    }
    req.Clear();
    res.Clear();
    while (g_concurrency.load() >= server_concurrency) {
    }
    g_concurrency++;
    LOG(INFO) << "Current concurrency " << g_concurrency.load();
    int data_index = turns * batch_size;
    if (create_req(&req, data_list, data_index, batch_size) != 0) {
      return;
    }
    if (predictor->inference(&req, &res) != 0) {
      LOG(ERROR) << "failed call predictor with req:" << req.ShortDebugString();
      return;
    }
    timeval end;
    gettimeofday(&end, NULL);
    uint64_t elapse_ms = (end.tv_sec * 1000 + end.tv_usec / 1000) -
                         (start.tv_sec * 1000 + start.tv_usec / 1000);
    response_time[thread_id].push_back(elapse_ms);
    print_res(req, res, predictor->tag(), elapse_ms);
    g_concurrency--;
    LOG(INFO) << "Done. Current concurrency " << g_concurrency.load();
    turns++;
  }
  api->thrd_finalize();
}

void calc_time(int server_concurrency, int batch_size) {
  std::vector<int> time_list;