Cluster script that I wrote for both Vungle and Webflow to cluster node apps. This code is running on hundreds of nodes with no issues. Supports worker disconnect timeout and a daily process restart as well.

cluster.js

var cluster = require('cluster');
var os = require('os');
var http = require('http');
var workers = [];
var startCount = 0;

var TIMEOUT = 10000;    // Wait 10 seconds for workers to disconnect
var RESTART = 86400000; // 24 hours
var intervalId;

function initInt() {
  clearInterval(intervalId);
  intervalId = setInterval(function () {
    console.log("Restarting...");
    process.kill(process.pid, "SIGUSR2");
  }, RESTART);
}

if (cluster.isMaster) {
  console.log('Master starting... CPUs: %d', os.cpus().length);
  os.cpus().forEach(function() {
    workers.push(cluster.fork());
  });

  initInt();

  process.on('SIGUSR2',function() {
    console.log("Got SIGUSR2 with " + workers.length + " workers");
    initInt();

    // Blank the worker list, but keep a copy for ourselves:
    var copyWorkers = workers.slice();
    workers = [];

    copyWorkers.forEach(function (worker) {
      console.log("Disconnecting " + worker.id);
      worker.disconnect();
      var savedTimeout = setTimeout(function () {
        console.log(worker.id + " not responding, destroying.");
        worker.forced_death = true;
        worker.destroy();
        clearTimeout(savedTimeout);
      }, TIMEOUT);

      worker.on('disconnect', function () {
        if (worker.forced_death) { 
          console.log("Forced death on disconnect " + worker.id);
          return; 
        }

        console.log(worker.id + " done.");
        worker.destroy();
        clearTimeout(savedTimeout);
      });

      var newWorker = cluster.fork();
      console.log('Pushing new worker: ' + newWorker.id);
      workers.push(newWorker);
    });
  });

  cluster.on('exit', function(worker, code, signal) {
    console.log('worker exit:', worker.id, code, signal);
    if (worker.suicide) { 
      console.log('exit - suicided worker: ' + worker.id);
      return; 
    }

    workers = workers.filter(function(existingWorker) { 
      return existingWorker.id !== worker.id; 
    });
    if (!worker.reloaded) {
      var newWorker = cluster.fork();
      console.log(worker.id + ' died - forking to: ' + newWorker.id);
      workers.push(newWorker);
    }
  });

} else {
  console.log('worker #' + cluster.worker.id + ' starting...');
  var server = require('./server');
  server.listen(3000);
}