Error handling, domains, and crash-only design

The Node community has embraced a crash-only design pattern, which simply means this: if you get an uncaught exception, catch it, log it, and restart the process. Crash-only design and domains work quite well as a pattern, particularly if your application is using cluster. Let's make a change to our cluster module, ./lib/cluster/index.js, on vision-core. Here, we include the domain module; instead of simply including our module to run in a cluster, we create a domain and call the run method. We then include a domain-based error handler that logs and then closes the process via process.exit(1). The cluster exit handler will pick this up and fork a new process:

var cluster = require('cluster')
, http = require('http')
, numCPUs = require('os').cpus().length
, logger = require('../logger')
, domain = require('domain');

function Cluster() {}

Cluster.prototype.run = function(module) {
  if (cluster.isMaster) {
    for (var i = 0; i < numCPUs...