Error Handling

Suggestions

// instead of:
if (util.isError(err) && (err.message !== ''))
// use just:
if (err)

// instead of:
next(err)
// use:
return next(err)    // ditto for other callbacks

// either throw or call back, but never both:
callback(err);
throw err;

// instead of anonymous functions:
function(cb) {}
// always name functions for better traceability:
function loadConfig(cb) {}

// instead of:
init(function startServer(db) {
// always send (err) as the first argument to any callback:
init(function startServer(err, db) {

Even further...

• Domains: you probably don't want to use this
• Wrapping: https://github.com/davepacheco/node-verror#werror-wrap-layered-errors

Performance

(& flow control)

Useful flow modules

• vasync (https://github.com/davepacheco/node-vasync)
• connect-parallel (https://github.com/Applifier/connect-parallel)
• promises (https://github.com/then/promise)

In general, parallel-ize wherever possible.

"Iceberg" organization

// this is the business logic you care about
vasync.parallel([ foo, bar, biz, baz ], onComplete);

// forces another good convention: no anonymous functions!
function foo(done) { }
function bar(done) { }
function biz(done) { }
function baz(done) { }

function onComplete(err, results) { }

Cluster

• Always cluster (with 1...N workers)
• Consider a tested worker-forking module (https://github.com/hunterloftis/throng ... recently hacked together for DevCenter)
  (there are a million others)

Raise maxSockets

  http.globalAgent.maxSockets = Infinity;

console.log == slow

consider 'verbose' or 'quiet' env vars (investigating streaming to stdout atm)

Testing

Useful Modules

• Mocha
• var assert = require('chai').assert
• Nock
• Supertest

Other suggestions

• Consider virtual machines / vagrant
• Add a .jshintrc to the module root
• Smaller modules = more likely to be tested
• "Pyramid" testing - isolate business logic at the base

Large App Patterns

The Wrapped App

• Almost there already (t1-modules)
• Goal is to separate business concerns & http concerns
• https://github.com/hunterloftis/node-production/blob/master/lib/web/index.js#L18

Worker Queues

• https://devcenter.heroku.com/articles/asynchronous-web-worker-model-using-rabbitmq-in-node
• Redis, RabbitMQ, IronWorker...

Message Brokers

• Fred George - good explanation of microservices
• Use a message broker (eg, amqp) to communicate between lots of independent services via messages. Page renderer -> "I need options for car X", Car Configurator -> "Options for car X: blah" ... Page renderer has a timeout and renders with whatever data is available after, say, 300ms.