update benchmark

Author: qicosmos

include/ylt/coro_io/ibverbs/ib_socket.hpp

@@ -206,7 +206,7 @@ struct ib_socket_shared_state_t
           sr.opcode = IBV_WR_SEND;
           sr.send_flags |= IBV_SEND_SIGNALED;
           std::error_code err;
-          if (self->qp_ == nullptr) {
+          if (self->has_close_) {
             err = std::make_error_code(std::errc::operation_canceled);
           }
           // post the receive request to the RQ

src/coro_rpc/benchmark/bench.cpp

@@ -71,7 +71,7 @@ inline std::string_view echo() {
   return g_resp_str;
 }
 
-ylt::metric::summary_t g_latency{"rpc call latency(us)", "help",
+ylt::metric::summary_t g_latency{"Latency(us) of rpc call", "help",
                                  std::vector{0.5, 0.9, 0.95, 0.99},
                                  std::chrono::seconds{60}};
 
@@ -94,13 +94,15 @@ async_simple::coro::Lazy<void> watcher(const bench_config& conf) {
     std::cout << "qps " << c / conf.send_data_len << ", Throughput:" << val
               << " Gb/s\n";
   }
+
+  std::cout << "# Benchmark result \n";
   double val = (8.0 * total) / (1000'000'000ll * conf.duration);
   std::cout << "avg qps " << total / (conf.send_data_len * conf.duration)
             << " and throughput:" << val << " Gb/s in duration "
             << conf.duration << "\n";
   std::string str_rate;
   g_latency.serialize(str_rate);
-  std::cout << "latency: " << str_rate << "\n";
+  std::cout << str_rate << "\n";
   co_return;
 }
 
@@ -127,7 +129,7 @@ async_simple::coro::Lazy<std::error_code> request(const bench_config& conf) {
             auto start = std::chrono::steady_clock::now();
             auto result = co_await client.call<echo>();
             if (!result.has_value()) {
-              ELOG_ERROR << result.error().msg;
+              ELOG_WARN << result.error().msg;
               co_return false;
             }
             auto now = std::chrono::steady_clock::now();

website/docs/zh/coro_rpc/coro_rpc_client.md

@@ -336,3 +336,46 @@ Lazy<void> call() {
 单个coro_rpc_client在多个线程同时调用时需要注意，只有部分成员函数是线程安全的，包括`send_request()`,`close()`,`connect()`,`get_executor()`,`get_pipeline_size()`,`get_client_id()`,`get_config()`等。如果用户没有重新调用connect()函数并传入endpoint或hostname，那么`get_port()`,`get_host()`函数也是线程安全的。
 
 需要注意，`call`,`get_resp_attachment`,`set_req_attachment`,`release_resp_attachment`和`init_config`函数均不是线程安全的，禁止多个线程同时调用。此时只能使用`send_request`实现多个线程并发请求同一个连接。
+
+# coro_rpc 性能测试
+先编译coro_rpc 的bench再执行bench命令行，命令行详情：
+```
+options:
+  -u, --url                   url (string [=0.0.0.0:9000])
+  -s, --send_data_len         send data length (unsigned long [=8388608])
+  -c, --client_concurrency    total number of http clients (unsigned int [=0])
+  -m, --max_request_count     max request count (unsigned long [=100000000])
+  -p, --port                  server port (unsigned short [=9000])
+  -r, --resp_len              response data length (unsigned long [=13])
+  -b, --buffer_size           buffer size (unsigned int [=2097152])
+  -o, --log_level             Severity::INFO 1 as default, WARN is 4 (int [=1])
+  -i, --enable_ib             enable ib (bool [=1])
+  -d, --duration              duration seconds (unsigned int [=100000])
+  -?, --help                  print this message
+```
+
+server侧命令行：
+`./bench -p 9004 -r 13 -o 5`
+
+含义：启动服务端，response的数据长度为13，日志级别为error(日志级别和easylog的日志级别对应)。
+
+client测命令行：
+`./bench -u 0.0.0.0:9004 -c 100 -s 8388608 -o 5 -d 30`
+
+含义：启动100个client去压测，每次请求发送8MB数据，日志级别为error，持续时间30秒。
+
+默认会启用ibverbs，如果希望只测试tcp则添加`-i 0`
+
+性能测试完成之后将输出qps，吞吐和latency数据，测试结果类似于：
+```
+# Benchmark result
+avg qps 2757 and throughput:185.04 Gb/s in duration 30
+# HELP Latency(us) of rpc call help
+# TYPE Latency(us) of rpc call summary
+rpc call latency(us){quantile="0.500000"} 3616.000000
+rpc call latency(us){quantile="0.900000"} 3712.000000
+rpc call latency(us){quantile="0.950000"} 3776.000000
+rpc call latency(us){quantile="0.990000"} 3872.000000
+rpc call latency(us)_sum 298707968.000000
+rpc call latency(us)_count 82761
+```