steed

Horsepower for your modules.

Steed is an alternative to async that is ~50-100% faster. It is not currently on-par with async in term of features. Please help us!

Installation
API
Caveats
Why it is so fast?
Acknowledgements
Licence & copyright

Watch Matteo presenting Steed at Node.js Interactive 2015: https://www.youtube.com/watch?v=_0W_822Dijg.

Install

npm i steed --save

steed()

Build an instance of steed, this step is not needed but welcomed for greater performance. Each steed utility likes being used for the same purpose.

steed.parallel([that,] tasks[, done(err, results)])

Executes a series of tasks in parallel.

tasks can either be an array of functions, or an object where each property is a function. done will be called with the results. The that argument will set this for each task and done callback.

Uses fastparallel.

Example:

var steed = require('steed')()
// or
// var steed = require('steed')

steed.parallel([
  function a (cb){
    cb(null, 'a');
  },
  function b (cb){
    cb(null, 'b');
  }
], function(err, results){
  // results is ['a', 'b']
})


// an example using an object instead of an array
steed.parallel({
  a: function a1 (cb){
    cb(null, 1)
  },
  b: function b1 (cb){
    cb(null, 2)
  }
}, function(err, results) {
  // results is  { a: 1, b: 2}
})

// an example using that parameter
// preferred form for max speed
function run (prefix, a, b, cb) {
  steed.parallel(new State(prefix, a, b, cb), [aT, bT], doneT)
}

// can be optimized by V8 using an hidden class
function State (prefix, a, b, cb) {
  this.a = a
  this.b = b
  this.cb = cb
  this.prefix = prefix
}

// because it is not a closure inside run()
// v8 can optimize this function
function aT (cb){
  cb(null, this.a);
}

// because it is not a closure inside run()
// v8 can optimize this function
function bT (cb){
  cb(null, this.b);
}

// because it is not a closure inside run()
// v8 can optimize this function
function doneT (err, results) {
  if (results) {
    results.unshift(this.prefix)
    results = results.join(' ')
  }
  this.cb(err, results)
}

run('my name is', 'matteo', 'collina', console.log)

Benchmark for doing 3 calls setImmediate 1 million times:

non-reusable setImmediate: 1781ms
async.parallel: 3484ms
neoAsync.parallel: 2162ms
insync.parallel: 10252ms
items.parallel: 3725ms
parallelize: 2928ms
fastparallel with results: 2139ms

These benchmarks where taken on node v4.1.0, on a MacBook Pro Retina Mid 2014 (i7, 16GB of RAM).

steed.series([that,] tasks[, done(err, results)])

Executes a series of tasks in series.

Uses fastseries.

Example:

var steed = require('steed')()
// or
// var steed = require('steed')

steed.series([
  function a (cb){
    cb(null, 'a');
  },
  function b (cb){
    cb(null, 'b');
  }
], function(err, results){
  // results is ['a', 'b']
})


// an example using an object instead of an array
steed.series({
  a: function a (cb){
    cb(null, 1)
  },
  b: function b (cb){
    cb(null, 2)
  }
}, function(err, results) {
  // results is  { a: 1, b: 2}
})

// an example using that parameter
// preferred form for max speed
function run (prefix, a, b, cb) {
  steed.series(new State(prefix, a, b, cb), [aT, bT], doneT)
}

// can be optimized by V8 using an hidden class
function State (prefix, a, b, cb) {
  this.a = a
  this.b = b
  this.cb = cb
  this.prefix = prefix
}

// because it is not a closure inside run()
// v8 can optimize this function
function aT (cb){
  cb(null, this.a);
}

// because it is not a closure inside run()
// v8 can optimize this function
function bT (cb){
  cb(null, this.b);
}

// because it is not a closure inside run()
// v8 can optimize this function
function doneT (err, results) {
  if (results) {
    results.unshift(this.prefix)
    results = results.join(' ')
  }
  this.cb(err, results)
}

run('my name is', 'matteo', 'collina', console.log)

Benchmark for doing 3 calls setImmediate 1 million times:

non-reusable setImmediate: 3887ms
async.series: 5981ms
neoAsync.series: 4338ms
fastseries with results: 4096ms

These benchmarks where taken on node v4.2.2, on a MacBook Pro Retina Mid 2014 (i7, 16GB of RAM).

steed.waterfall(tasks[, done(err, ...)])

Runs the functions in tasks in series, each passing their result to the next task in the array. Quits early if any of the tasks errors.

Uses fastfall.

Example:

var steed = require('steed')()
// or
// var steed = require('steed')

steed.waterfall([
  function a (cb) {
    console.log('called a')
    cb(null, 'a')
  },
  function b (a, cb) {
    console.log('called b with:', a)
    cb(null, 'a', 'b')
  },
  function c (a, b, cb) {
    console.log('called c with:', a, b)
    cb(null, 'a', 'b', 'c')
  }], function result (err, a, b, c) {
    console.log('result arguments', arguments)
  })

// preferred version for maximum speed
function run (word, cb) {
  steed.waterfall(new State(cb), [
    aT, bT, cT,
  ], cb)
}

// can be optimized by V8 using an hidden class
function State (value) {
  this.value = value
}

// because it is not a closure inside run()
// v8 can optimize this function
function aT (cb) {
  console.log(this.value)
  console.log('called a')
  cb(null, 'a')
}

// because it is not a closure inside run()
// v8 can optimize this function
function bT (a, cb) {
  console.log('called b with:', a)
  cb(null, 'a', 'b')
}

// because it is not a closure inside run()
// v8 can optimize this function
function cT (a, b, cb) {
  console.log('called c with:', a, b)
  cb(null, 'a', 'b', 'c')
}

Benchmark for doing 3 calls setImmediate 100 thousands times:

non-reusable setImmediate: 418ms
async.waterfall: 1174ms
run-waterfall: 1432ms
insync.wasterfall: 1174ms
neo-async.wasterfall: 469ms
waterfallize: 749ms
fastfall: 452ms

These benchmarks where taken on node v4.2.2, on a MacBook Pro Retina Mid 2014 (i7, 16GB of RAM).

steed.each([that,] array, iterator(item, cb), [, done()])

Iterate over all elements of the given array asynchronosly and in parallel. Calls iterator with an item and a callback. Calls done when all have been processed.

The that argument will set this for each task and done callback.

each does not handle errors, if you need errors, use map.

Uses fastparallel.

Example:

var steed = require('steed')()
// or
// var steed = require('steed')

var input = [1, 2, 3]
var factor = 2

steed.each(input, function (num, cb) {
  console.log(num * factor)
  setImmediate(cb)
}, function () {
  console.log('done')
})

// preferred version for max speed
function run (factor, args, cb) {
  steed.each(new State(factor), work, cb)
}

// can be optimizied by V8 using an hidden class
function State (factor) {
  this.factor = factor
}

// because it is not a closure inside run()
// v8 can optimize this function
function work (num, cb) {
  console.log(num * this.factor)
  cb()
}

run(factor, input, console.log)

Benchmark for doing 3 calls setImmediate 1 million times:

non-reusable setImmediate: 1781ms
async.each: 2621ms
neoAsync.each: 2156ms
insync.parallel: 10252ms
insync.each: 2397ms
fastparallel each: 1941ms

These benchmarks where taken on node v4.2.2, on a MacBook Pro Retina Mid 2014 (i7, 16GB of RAM).

steed.eachSeries([that,] array, iterator(item, cb), [, done(err)])

Iterate over all elements of the given array asynchronously and in series. Calls iterator with an item and a callback. Calls done when all have been processed.

The that argument will set this for each task and done callback.

eachSeries does not handle errors, if you need errors, use mapSeries.

Uses fastseries.

Example:

var steed = require('steed')()
// or
// var steed = require('steed')

var input = [1, 2, 3]
var factor = 2

steed.eachSeries(input, function (num, cb) {
  console.log(num * factor)
  setImmediate(cb)
}, function (err) {
  console.log(err)
})

// preferred version for max speed
function run (factor, args, cb) {
  steed.eachSeries(new State(factor), work, cb)
}

// can be optimizied by V8 using an hidden class
function State (factor) {
  this.factor = factor
}

// because it is not a closure inside run()
// v8 can optimize this function
function work (num, cb) {
  console.log(num * this.factor)
  cb()
}

run(factor, input, console.log)

Benchmark for doing 3 calls setImmediate 1 million times:

non-reusable setImmediate: 3887ms
async.mapSeries: 5540ms
neoAsync.eachSeries: 4195ms
fastseries each: 4168ms

These benchmarks where taken on node v4.2.2, on a MacBook Pro Retina Mid 2014 (i7, 16GB of RAM).

steed.map([that,] array, iterator(item, cb), [, done(err, results)])

Performs a map operation over all elements of the given array asynchronously and in parallel. The result is an a array where all items have been replaced by the result of iterator.

The that argument will set this for each task and done callback.

Calls iterator with an item and a callback. Calls done when all have been processed.

Uses fastparallel.

Example:

var steed = require('steed')()
// or
// var steed = require('steed')

var input = [1, 2, 3]
var factor = 2

steed.map(input, function (num, cb) {
  setImmediate(cb, null, num * factor)
}, function (err, results) {
  if (err) { throw err }

  console.log(results.reduce(sum))
})

function sum (acc, num) {
  return acc + num
}

// preferred version for max speed
function run (factor, args, cb) {
  steed.map(new State(factor, cb), args, work, done)
}

// can be optimizied by V8 using an hidden class
function State (factor, cb) {
  this.factor = factor
  this.cb = cb
}

// because it is not a closure inside run()
// v8 can optimize this function
function work (num, cb) {
  setImmediate(cb, null, num * this.factor)
}

function done (err, results) {
  results = results || []
  this.cb(err, results.reduce(sum))
}

run(2, [1, 2, 3], console.log)

Benchmark for doing 3 calls setImmediate 1 million times:

non-reusable setImmediate: 1781ms
async.map: 3054ms
neoAsync.map: 2080ms
insync.map: 9700ms
fastparallel map: 2102ms

These benchmarks where taken on node v4.2.2, on a MacBook Pro Retina Mid 2014 (i7, 16GB of RAM).

steed.mapSeries([that,] array, iterator(item, cb), [, done(err, results)])

Performs a map operation over all elements of the given array asynchronosly and in series. The result is an a array where all items have been replaced by the result of iterator.

Calls iterator with an item and a callback. Calls done when all have been processed.

The that argument will set this for each task and done callback.

Uses fastseries.

Example:

var steed = require('steed')()
// or
// var steed = require('steed')

var input = [1, 2, 3]
var factor = 2

steed.mapSeries(input, function (num, cb) {
  setImmediate(cb, null, num * factor)
}, function (err, results) {
  if (err) { throw err }

  console.log(results.reduce(sum))
})

function sum (acc, num) {
  return acc + num
}

// preferred version for max speed
function run (factor, args, cb) {
  steed.mapSeries(new State(factor, cb), args, work, done)
}

// can be optimizied by V8 using an hidden class
function State (factor, cb) {
  this.factor = factor
  this.cb = cb
}

// because it is not a closure inside run()
// v8 can optimize this function
function work (num, cb) {
  setImmediate(cb, null, num * this.factor)
}

function done (err, results) {
  results = results || []
  this.cb(err, results.reduce(sum))
}

run(2, [1, 2, 3], console.log)

Benchmark for doing 3 calls setImmediate 1 million times:

non-reusable setImmediate: 3887ms
async.mapSeries: 5540ms
neoAsync.mapSeries: 4237ms
fastseries map: 4032ms

These benchmarks where taken on node v4.2.2, on a MacBook Pro Retina Mid 2014 (i7, 16GB of RAM).

steed.queue(worker, concurrency)

Creates a new queue. See fastq for full API.

Arguments:

worker, worker function, it would be called with that as this, if that is specified.
concurrency, number of concurrent tasks that could be executed in parallel.

Example:

var steed = require('steed')()
// or
// var steed = require('steed')

var queue = steed.queue(worker, 1)

queue.push(42, function (err, result) {
  if (err) { throw err }
  console.log('the result is', result)
})

function worker (arg, cb) {
  cb(null, arg * 2)
}

Benchmarks (1 million tasks):

setImmedidate: 1313ms
fastq: 1462ms
async.queue: 3989ms

Obtained on node 4.2.2, on a MacBook Pro 2014 (i7, 16GB of RAM).

Caveats

This library works by caching the latest used function, so that running a new parallel does not cause any memory allocations.

The done function will be called only once, even if more than one error happen.

Steed has no safety checks: you should be responsible to avoid sync functions and so on. Also arguments type checks are not included, so be careful in what you pass.

Why it is so fast?

This library is caching functions a lot. We invented a technique to do so, and packaged it in a module: reusify.
V8 optimizations: thanks to caching, the functions can be optimized by V8 (if they are optimizable, and we took great care of making them so).
Don't use arrays if you just need a queue. A linked list implemented via objects is much faster if you do not need to access elements in between.

Acknowledgements

Steed is sponsored by nearForm.

The steed logo was created, with thanks, by Dean McDonnell

License

MIT

mcollina/steed

mcollina

Reviews

Repository Details

steed

Install

API

steed()

steed.parallel([that,] tasks[, done(err, results)])

steed.series([that,] tasks[, done(err, results)])

steed.waterfall(tasks[, done(err, ...)])

steed.each([that,] array, iterator(item, cb), [, done()])

steed.eachSeries([that,] array, iterator(item, cb), [, done(err)])

steed.map([that,] array, iterator(item, cb), [, done(err, results)])

steed.mapSeries([that,] array, iterator(item, cb), [, done(err, results)])

steed.queue(worker, concurrency)

Caveats

Why it is so fast?

Acknowledgements

License

More Repositories