Simple Node Cluster

After much testing of our internal API, we made it available for several sites to use & got terrible performance.

Of course, we got better performance with Node than we did with PHP, thanks to async parallelizing functions, but we didn’t think too much about Node’s single-threaded nature.

We already had several sites using the API with more being added daily, all causing performance to quickly dwindle.

In a rush, we discovered Node’s cluster module, which I don’t remember from v0.6, but is pretty reliable in v0.10.

Per the cluster documentation:

A single instance of Node runs in a single thread. To take advantage of multi-core systems the user will sometimes want to launch a cluster of Node processes to handle the load.

The cluster module allows you to easily create child processes that all share server ports.

Implementation was dead easy.

1. Setup our Express application for easy use via require:

// app/server.js
var app = require('./app')
var http = require('http')

http.createServer(app).listen(app.get('port'), function () {
  console.log('Express server listening on port ' + app.get('port'))
})

For development, we will run node app/server:

$ node app/server.js
Express server listening on port 3000

2. Create cluster.js

// app/cluster.js

var cluster = require('cluster')

if (cluster.isMaster) {
  // Count the machine's CPUs
  var cpuCount = require('os').cpus().length

  // Create a worker for each CPU
  for (var i = 0; i < cpuCount; i += 1) {
    cluster.fork()
  }

  // Listen for dying workers & start 'em back up!
  cluster.on('exit', function () {
    cluster.fork()
  })
} else {
  require('./server')
}

3. Update package.json to use cluster.js

// package.json
{
  "scripts": {
    "start": "node app/cluster"
  }
}

4. Done!

Now on production, npm start will run node app/cluster and spawn up 4+ instances:

$ npm start

> api@1.4.2 start /Users/deploy/api
> node app/cluster

Express server listening on port 3000
Express server listening on port 3000
Express server listening on port 3000

cluster