Node.js负载测试出现ECONNREFUSED/ETIMEDOUT问题求助

开发环境负载测试场景：客户端向Node.js服务器并行发送多API请求，处理10万条记录（每条需发起20-25次API调用），初期运行正常，成功处理7.5-8万条后频繁出现ECONNREFUSED/ETIMEDOUT错误。已使用单例axios实例，实现退避重试机制但效果不佳，CPU使用率未达上限。当前axios配置如下：

const Agent = require('agentkeepalive');
const HttpsAgent = require('agentkeepalive').HttpsAgent;
import axios from 'axios';

const Cacheable = require('cacheable-lookup');

let cacheable = new Cacheable.default({ maxTtl: 300 });

let instance;
const keepAliveAgent = new Agent({
  maxSockets: 1500,
  maxFreeSockets: 256,
  timeout: 5 * 60000,
  freeSocketTimeout: 5000,
  keepAlive: true,
  keepAliveMsecs: 60000,
});

// HTTPS agent
const httpsKeepAliveAgent = new HttpsAgent({
  maxSockets: 1500,
  maxFreeSockets: 256,
  timeout: 5 * 60000,
  freeSocketTimeout: 5000,
  keepAlive: true,
  keepAliveMsecs: 60000,
});

// Attach cacheable-lookup to the agents
cacheable.install(keepAliveAgent);
cacheable.install(httpsKeepAliveAgent);


const axiosInstanceGenerator = () => {
  if (!instance) {
    instance = axios.create({
      timeout: 30 * 60000,
      httpAgent: keepAliveAgent,
      httpsAgent: httpsKeepAliveAgent,
    });
  }

  return instance;
};

export const axiosInstance = axiosInstanceGenerator();

优化方案 #

一、调整连接池参数，避免目标服务器过载 #

当前maxSockets设置过高（1500），可能超出目标服务器的连接承载上限；maxFreeSockets偏低导致闲置连接快速回收，新建连接开销累积。调整参数如下：

const keepAliveAgent = new Agent({
  maxSockets: 500, // 降低并发连接数，匹配目标服务器能力
  maxFreeSockets: 1000, // 增加空闲连接保留数，减少新建连接频率
  timeout: 5 * 60000,
  freeSocketTimeout: 30000, // 延长空闲连接超时，避免频繁销毁重建
  keepAlive: true,
  keepAliveMsecs: 60000,
});

const httpsKeepAliveAgent = new HttpsAgent({
  maxSockets: 500,
  maxFreeSockets: 1000,
  timeout: 5 * 60000,
  freeSocketTimeout: 30000,
  keepAlive: true,
  keepAliveMsecs: 60000,
});

二、优化DNS缓存注入逻辑 #

cacheable-lookup绑定到agent的方式存在兼容性风险，改为直接注入axios的lookup配置：

// 简化Cacheable实例化方式
const cacheable = new Cacheable({ maxTtl: 300 });

const axiosInstanceGenerator = () => {
  if (!instance) {
    instance = axios.create({
      timeout: 5 * 60000, // 与agent超时对齐，避免逻辑冲突
      httpAgent: keepAliveAgent,
      httpsAgent: httpsKeepAliveAgent,
      lookup: cacheable.lookup.bind(cacheable) // 直接注入DNS缓存
    });
  }
  return instance;
};

三、实现智能重试+流量限流 #

现有重试机制未区分错误类型，且无流量控制，导致无效重试加剧连接压力。结合axios-retry和p-limit优化：

1. 安装依赖：

npm install axios-retry p-limit

2. 修改代码：

import axiosRetry from 'axios-retry';
import pLimit from 'p-limit';

// 限制并发请求数，避免瞬间打满连接池
const requestLimit = pLimit(400);

const axiosInstanceGenerator = () => {
  if (!instance) {
    instance = axios.create({
      timeout: 5 * 60000,
      httpAgent: keepAliveAgent,
      httpsAgent: httpsKeepAliveAgent,
      lookup: cacheable.lookup.bind(cacheable)
    });

    // 配置差异化重试规则
    axiosRetry(instance, {
      retries: 3,
      retryDelay: (retryCount) => Math.pow(2, retryCount) * 1000, // 指数退避：1s→2s→4s
      retryCondition: (error) => {
        // 仅对连接类错误重试，跳过业务错误
        return axiosRetry.isNetworkError(error) || 
               error.code === 'ECONNREFUSED' || 
               error.code === 'ETIMEDOUT';
      },
      shouldResetTimeout: true // 重试时重置超时时间
    });
  }
  return instance;
};

// 用限流包装请求
export const limitedAxiosRequest = (config) => {
  return requestLimit(() => axiosInstance.request(config));
};

四、添加连接池状态监控 #

定时打印连接池状态，排查是否存在连接泄漏：

// 每10秒打印一次连接池状态
setInterval(() => {
  console.log('HTTP Agent 状态:', {
    总连接数: keepAliveAgent.totalSockets,
    空闲连接数: keepAliveAgent.freeSockets,
    活跃连接数: Object.keys(keepAliveAgent.sockets).reduce((sum, key) => sum + keepAliveAgent.sockets[key].length, 0)
  });
  console.log('HTTPS Agent 状态:', {
    总连接数: httpsKeepAliveAgent.totalSockets,
    空闲连接数: httpsKeepAliveAgent.freeSockets,
    活跃连接数: Object.keys(httpsKeepAliveAgent.sockets).reduce((sum, key) => sum + httpsKeepAliveAgent.sockets[key].length, 0)
  });
}, 10000);

若发现总连接数持续增长、空闲连接未回收，需检查请求是否未正确捕获错误（未释放连接）。

内容的提问来源于stack exchange，提问作者WizRd AM