负载测试

由于 Socket.IO 有自己的协议，包括握手、心跳和自定义数据包编码，负载测试 Socket.IO 服务器最简单的方法是使用 Socket.IO 客户端库并创建大量客户端。

有两种经典的解决方案可以做到这一点：

Artillery

Artillery 是负载测试应用程序的绝佳工具。它允许创建连接、发送事件和检查确认。

文档可以在这里找到。

重要提示：默认安装带有 v2 客户端，与 v3/v4 服务器不兼容。您需要为此安装自定义引擎：https://github.com/ptejada/artillery-engine-socketio-v3

安装:

  1. $ npm install artillery artillery-engine-socketio-v3

示例场景：

  1. # my-scenario.yml
  2. config:
  3.   target: "http://localhost:3000"
  4.   phases:
  5.     - duration: 60
  6.       arrivalRate: 10
  7.   engines:
  8.    socketio-v3: {}
  10. scenarios:
  11.   - name: My sample scenario
  12.     engine: socketio-v3
  13.     flow:
  14.       # wait for the WebSocket upgrade (optional)
  15.       - think: 1
  17.       # basic emit
  18.       - emit:
  19.           channel: "hello"
  20.           data: "world"
  22.       # emit an object
  23.       - emit:
  24.           channel: "hello"
  25.           data:
  26.             id: 42
  27.             status: "in progress"
  28.             tags:
  29.               - "tag1"
  30.               - "tag2"
  32.       # emit in a custom namespace
  33.       - namespace: "/my-namespace"
  34.         emit:
  35.           channel: "hello"
  36.           data: "world"
  38.       # emit with acknowledgement
  39.       - emit:
  40.           channel: "ping"
  41.         acknowledge:
  42.           match:
  43.             value: "pong"
  45.       # do nothing for 30 seconds then disconnect
  46.       - think: 30

要运行此方案：

  1. $ npx artillery run my-scenario.yml

Artillery 还附带了许多很棒的功能，例如能够将指标发布到各种端点 或 从 AWS 运行测试。

它唯一的限制是您无法轻松测试服务器到客户端的事件，因为 Artillery DSL 更适合经典的客户端到服务器通信。这将我们带到下一节.

手动创建客户端

这是创建一千个 Socket.IO 客户端并监控每秒接收的数据包数量的基本脚本：

  1. const { io } = require("socket.io-client");
  3. const URL = process.env.URL || "http://localhost:3000";
  4. const MAX_CLIENTS = 1000;
  5. const POLLING_PERCENTAGE = 0.05;
  6. const CLIENT_CREATION_INTERVAL_IN_MS = 10;
  7. const EMIT_INTERVAL_IN_MS = 1000;
  9. let clientCount = 0;
  10. let lastReport = new Date().getTime();
  11. let packetsSinceLastReport = 0;
  13. const createClient = () => {
  14.   // for demonstration purposes, some clients stay stuck in HTTP long-polling
  15.   const transports =
  16.     Math.random() < POLLING_PERCENTAGE ? ["polling"] : ["polling", "websocket"];
  18.   const socket = io(URL, {
  19.     transports,
  20.   });
  22.   setInterval(() => {
  23.     socket.emit("client to server event");
  24.   }, EMIT_INTERVAL_IN_MS);
  26.   socket.on("server to client event", () => {
  27.     packetsSinceLastReport++;
  28.   });
  30.   socket.on("disconnect", (reason) => {
  31.     console.log(`disconnect due to ${reason}`);
  32.   });
  34.   if (++clientCount < MAX_CLIENTS) {
  35.     setTimeout(createClient, CLIENT_CREATION_INTERVAL_IN_MS);
  36.   }
  37. };
  39. createClient();
  41. const printReport = () => {
  42.   const now = new Date().getTime();
  43.   const durationSinceLastReport = (now - lastReport) / 1000;
  44.   const packetsPerSeconds = (
  45.     packetsSinceLastReport / durationSinceLastReport
  46.   ).toFixed(2);
  48.   console.log(
  49.     `client count: ${clientCount} ; average packets received per second: ${packetsPerSeconds}`
  50.   );
  52.   packetsSinceLastReport = 0;
  53.   lastReport = now;
  54. };
  56. setInterval(printReport, 5000);

您可以将其用作负载测试您自己的应用程序的起点。