Cloud Run部署Flask-SocketIO服务遇Polling 400错误求助

我使用Python-socketio结合Flask监听GCP Pub/Sub主题，获取消息后推送给React应用。本地运行完全正常，但将服务部署到GCP Cloud Run后，客户端轮询时出现400错误。客户端初始启动时显示已连接，但向Pub/Sub主题发送消息时触发该错误。

服务端代码（Flask-SocketIO） #

from flask import Flask, request, jsonify
from flask_socketio import SocketIO
import base64
import json
import humps
import os

app = Flask(__name__)
socketio = SocketIO(app, cors_allowed_origins="*", cors_allowed_methods="*")

@app.route('/', methods=['POST'])
def receive_pubsub_message():
    publish_message = request.get_json()

    if "message" in publish_message and "data" in publish_message["message"]:
        # Decode Pub/Sub message
        message_data = base64.b64decode(publish_message["message"]["data"]).decode("utf-8")
        request_json = humps.decamelize(json.loads(message_data))
    else:
        # Handle regular JSON payload (if necessary)
        request_json = humps.decamelize(request.get_json())
    print("request: ", request_json)
    
    # Emit the decoded message to all connected clients
    socketio.emit('new_message', request_json)
    print("Emitting new_message event to clients: ", request_json)

    # Acknowledge the Pub/Sub message by returning a 200 response
    return jsonify({}), 200


if __name__ == '__main__':
    socketio.run(app, debug=True, host='0.0.0.0', port=int(os.environ.get('PORT', 8081)))

客户端代码（React） #

import React, { useEffect, useState } from "react";
import io from "socket.io-client";

function App() {
  const [messages, setMessages] = useState([]);

  useEffect(() => {

    // const socket = io("http://localhost:8081");

    const socket = io(
      "wss://xxx-uc.a.run.app"
    );

    socket.on("connect", () => {
      console.log("Connected to server: ", { socket });
    });

    // Listen for incoming messages
    socket.on("new_message", (msg) => {
      console.log("new message: ", { msg });
      // Use the callback function form of setMessages to ensure state updates properly
      setMessages((prevMessages) => [...prevMessages, JSON.stringify(msg)]);
    });

    return () => {
      socket.disconnect();
    };
  }, []); // Empty dependency array to run once on component mount

  return (
    <div className="App">
      <h1>Chat Application</h1>
      <div className="message-list">
        {messages.map((msg, idx) => (
          <div className="message" key={idx}>
            {msg}
          </div>
        ))}
      </div>
    </div>
  );
}

export default App;

错误现象 #

部署到Cloud Run后，客户端发起的轮询请求返回400状态码，请求地址类似：https://xxx-uc.a.run.app/socket.io/?EIO=4&transport=polling&t=Oiwn41V&sid=KenZXFDBdawUL3opAAAA

排查方案 #

• 开启Cloud Run会话亲和性：SocketIO需要客户端始终连接到同一服务实例，Cloud Run默认关闭会话亲和性，导致后续轮询请求被路由到其他实例，实例无法识别之前生成的会话ID（sid），返回400。进入Cloud Run服务设置，找到会话亲和性选项并开启。
• 适配生产级异步模式：本地debug模式使用Flask自带服务器，部署到Cloud Run需使用生产级异步引擎。确保requirements.txt包含eventlet，修改启动代码关闭debug并指定异步模式：

  if __name__ == '__main__':
      socketio.run(app, debug=False, host='0.0.0.0', port=int(os.environ.get('PORT', 8081)), async_mode='eventlet')
  

• 优化CORS配置：避免使用通配符*作为cors_allowed_origins，显式指定React应用的域名，同时开启凭证支持（如果需要）：

  socketio = SocketIO(app, cors_allowed_origins=["https://your-react-app-domain.com"], allow_credentials=True)
  

• 查看Cloud Run日志：检查服务端日志，若出现Invalid session ID类错误，可确认是会话亲和性导致的问题；若有其他错误信息，可进一步定位根源。
• 验证Pub/Sub推送的实例一致性：如果开启会话亲和性后仍有问题，需确认Pub/Sub的推送请求是否路由到了客户端连接的实例，不过该场景下的400错误优先关联客户端轮询的会话问题。

内容的提问来源于stack exchange，提问作者GLP