runner.js

"use strict";

var cluster = require("cluster");

if (process.env.NODE_ENV == 'production' && cluster.isMaster) {
    // this is the master control process
    console.log("Control process running: PID=" + process.pid);

    // fork as many times as we have CPUs
    var numCPUs = process.env.NPROCS || require("os").cpus().length;

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

    // handle unwanted worker exits
    cluster.on("exit", function(worker, code, signal) {
        var exit_code = worker.process.exitCode;
        if (!worker.suicide) {
            console.log("Error: Worker crashed with exit code: " + code + "/" + exit_code + ", signal: " + signal + ". Spawning a replacement.");
            cluster.fork();
        }
    });

    // I'm using the SIGUSR2 signal to listen for reload requests
    // you could, instead, use file watcher logic, or anything else
    process.on("SIGUSR2", function() {
        console.log("SIGUSR2 received, reloading workers");

        // delete the cached module, so we can reload the app
        delete require.cache[require.resolve("./app")];

        // only reload one worker at a time
        // otherwise, we'll have a time when no request handlers are running
        var i = 0;
        var workers = Object.keys(cluster.workers);
        var f = function() {
            if (i == workers.length) return; 

            console.log("Killing " + workers[i]);
            var worker = cluster.workers[workers[i]];

            worker.disconnect();
            var disconnect_received = false;
            worker.on("disconnect", function() {
                disconnect_received = true;
                console.log("Shutdown complete");
                var dead = false;
                var timer = setTimeout(function () {
                    if (!dead) {
                        console.log("Worker " + workers[i] + " failed to shutdown. Killing.");
                        worker.kill();
                    }
                }, 30000);
                worker.on("exit", function () { dead = true; clearTimeout(timer); });
            });
            var timer = setTimeout(function () {
                if (!disconnect_received) {
                    console.log("Disconnect never received by worker. Killing.");
                    worker.kill();
                }
            }, 30000);
            var newWorker = cluster.fork();
            newWorker.on("listening", function() {
                console.log("Replacement worker online.");
                i++;
                f();
            });
        }
        f();
    });
} else {
    var app = require("./app");
}