无需重新启动即可升级Node.JS Web应用程序

时间:2014-05-09 12:09:18

标签: node.js high-availability

有没有办法在不重新启动的情况下更改Node.JS应用程序的代码。

例如,如果当时正在从服务器下载文件,是否有办法在不中断文件传输的情况下升级服务器?

1 个答案:

答案 0 :(得分:8)

是的,可以使用cluster模块。

基本上,您将启动应用程序的多个从属实例,通过主进程共享同一端口。主设备将传入的连接分配给从设备。

当你想重新启动时,主人将按顺序:

  • 停止将连接传递给从属
  • 轻轻地告诉它重启
  • 等待它重生

这里有一些代码可以帮助您入门(您需要npm install async underscore)。

注意:这只是为了让您入门。您应该准备好在真实设置中处理意外的从站故障和超时。

var cluster = require('cluster');
var http = require('http');
var _ = require('underscore');
var async = require('async');
var numCPUs = require('os').cpus().length;

if (cluster.isMaster) {
  console.log('starting master with PID', process.pid);
  // Fork workers.
  var slaves = {};

  for (var i = 0; i < numCPUs; i++) {
    cluster.fork();
  }

  cluster.on('fork', function (worker) {
    slaves[worker.id] = worker;
  });

  cluster.on('exit', function (worker, code, signal) {
    console.log('worker ' + worker.process.pid + ' died');
    delete slaves[worker.id];

    // restart the worker
    cluster.fork();
  });

  process.on('SIGHUP', function restartApp() {
    console.log('restarting all slaves');
    // important: force node.js to reload all js sources
    delete require.cache;

    var toRestart = _(slaves).values();

    async.eachSeries(toRestart, function (slave, done) {
      slave.kill('SIGTERM');

      // when the new worker starts, proceed to the next instance
      cluster.once('listening', function () {
        done();
      });
    });
  });
} else {
  console.log('- starting slave with PID', process.pid);
  // Workers can share any TCP connection
  // In this case its a HTTP server
  var server = http.createServer(function(req, res) {
    res.writeHead(200);
    res.end("hello world\n");
  });

  process.on('SIGTERM', function () {
    // finish all current connections, then stop
    server.close(function () {
      process.exit(0);
    });
  });

  server.listen(8000);
}

当你调用`kill -HUP时,你会看到它按顺序重启所有从属:

worker 5156 died
- starting slave with PID 5164
worker 5157 died
- starting slave with PID 5165
worker 5158 died
- starting slave with PID 5166
worker 5159 died
- starting slave with PID 5167

重新启动时,总​​有3个正在运行的从站来处理传入连接。

最后一件事:如果您尝试,console.log在从属SIGTERM处理程序中不执行任何操作,因为它已经与主服务器断开连接。