Cluster solution based on node's cluster module.

#Features

• Restarts workers that die
• Zero downtime reloading of application when receiving SIGHUP
• Gracefully dies when receiving SIGQUIT
• Correctly reloads code when deployed with capistrano

Hacks

• Overrides net._createServerHandle to track sockets open by master so they can be closed during graceful shutdown. If they are not closed then new clients will hang instead of receiving connection refused.
• Overrides net.Server.prototype.listen in worker to ensure an error is generated if a port cannot be bound to

#Issues

• Not enough testing
• Nodejs's cluster module doesn't appear to be fair (possibly why people have created their own nodejs http proxy balancers)

#Related

• http://learnboost.github.com/cluster/
• https://github.com/LearnBoost/up
• https://github.com/LearnBoost/distribute

#Usage

Example app.js

var ncluster = require('ncluster');
ncluster('./server.js', {workers: 5});

Example server.js

var http = require('http');
var http_server = http.createServer(function(req, res) {
  res.writeHead(200, {'Content-Type': 'text/plain'});
  res.end('Hello From Worker\n');
});

http_server.listen(3000);


module.exports = http_server;

Default Configuration

var defaults = {
  workers: os.cpus().length,
  dir: path.dirname(process.argv[1]),
  heartbeat_timeout: 10 * 1000,
  startup_timeout: 60 * 1000,
  log_dir: "log",
  kill_wait_timeout: 30 * 1000,
  heartbeat_interval: 500
};

Examples

https://github.com/benmmurphy/nodejs_vagrant_helloworld