# Node.js Performance

# Enable gzip

const http = require('http')
const fs   = require('fs')
const zlib = require('zlib')

http.createServer((request, response) => {
  const stream          = fs.createReadStream('index.html')
  const acceptsEncoding = request.headers['accept-encoding']

  let encoder = {
    hasEncoder     : false,
    contentEncoding: {},
    createEncoder  : () => throw 'There is no encoder'
  }

  if (!acceptsEncoding) {
    acceptsEncoding = ''
  }

  if (acceptsEncoding.match(/\bdeflate\b/)) {
    encoder = {
      hasEncoder     : true,
      contentEncoding: { 'content-encoding': 'deflate' },
      createEncoder  : zlib.createDeflate
    }
  } else if (acceptsEncoding.match(/\bgzip\b/)) {
    encoder = {
      hasEncoder     : true,
      contentEncoding: { 'content-encoding': 'gzip' },
      createEncoder  : zlib.createGzip
    }
  }

  response.writeHead(200, encoder.contentEncoding)

  if (encoder.hasEncoder) {
    stream = stream.pipe(encoder.createEncoder())
  }

  stream.pipe(response)

}).listen(1337)

# Event Loop

# Blocking Operation Example

let loop = (i, max) => {
  while (i < max) i++
  return i
}

// This operation will block Node.js
// Because, it's CPU-bound
// You should be careful about this kind of code
loop(0, 1e+12)

# Non-Blocking IO Operation Example

let i = 0

const step = max => {
  while (i < max) i++
  console.log('i = %d', i)
}

const tick = max => process.nextTick(step, max)

// this will postpone tick run step's while-loop to event loop cycles
// any other IO-bound operation (like filesystem reading) can take place
// in parallel
tick(1e+6)
tick(1e+7)
console.log('this will output before all of tick operations. i = %d', i)
console.log('because tick operations will be postponed')
tick(1e+8)

event loop diagram (opens new window)

In simpler terms, Event Loop is a single-threaded queue mechanism which executes your CPU-bound code until end of its execution and IO-bound code in a non-blocking fashion.

However, Node.js under the carpet uses multi-threading for some of its operations through libuv (opens new window) Library.

# Performance Considerations

  • Non-blocking operations will not block the queue and will not effect the performance of the loop.
  • However, CPU-bound operations will block the queue, so you should be careful not to do CPU-bound operations in your Node.js code.

Node.js non-blocks IO because it offloads the work to the operating system kernel, and when the IO operation supplies data (as an event), it will notify your code with your supplied callbacks.

# Increase maxSockets

# Basics

require('http').globalAgent.maxSockets = 25

// You can change 25 to Infinity or to a different value by experimenting

Node.js by default is using maxSockets = Infinity at the same time (since v0.12.0 (opens new window)). Until Node v0.12.0, the default was maxSockets = 5 (see v0.11.0 (opens new window)). So, after more than 5 requests they will be queued. If you want concurrency, increase this number.

# Setting your own agent

http API is using a "Global Agent (opens new window)". You can supply your own agent. Like this:

const http = require('http')
const myGloriousAgent = new http.Agent({ keepAlive: true })
myGloriousAgent.maxSockets = Infinity

http.request({ ..., agent: myGloriousAgent }, ...)

# Turning off Socket Pooling entirely

const http = require('http')
const options = {.....}

options.agent = false

const request = http.request(options)

# Pitfalls

  • You should do the same thing for `https` API if you want the same effects
  • Beware that, for example, [AWS](http://docs.aws.amazon.com/sdk-for-javascript/v2/developer-guide/node-configuring-maxsockets.html) will use 50 instead of `Infinity`.