In a nutshell:

HappyPack makes webpack builds faster by allowing you to transform multiple files in parallel.

See "How it works" below for more details.

• webpack initial build times are horrifying in large codebases (3k+ modules)
• something that works against both a one-time build (e.g. for a CI) and continuous builds (i.e. --watch during development)

npm install --save-dev happypack

In your webpack.config.js, you need to use the plugin and tell it of the loaders it should use to transform the sources. Note that you must specify the absolute paths for these loaders as we do not use webpack's loader resolver at this point.

var HappyPack = require('happypack');

exports.plugins = [
  new HappyPack({
    // loaders is the only required parameter:
    loaders: [ 'babel?presets[]=es2015' ],

    // customize as needed, see Configuration below
  })
];

Now you replace your current loaders with HappyPack's loader (possibly use an env variable to enable HappyPack):

exports.module = {
  loaders: {
    test: /.js$/,
    loaders: [ 'happypack/loader' ],
    include: [
      // ...
    ],
  }
};

That's it. Now sources that match .js$ will be handed off to happypack which will transform them in parallel using the loaders you specified.

These are the parameters you can pass to the plugin when you instantiate it.

Each loader entry consists of the name or path of loader that would transform the files and an optional query string to pass to it. This looks similar to what you'd pass to webpack's loader config.

NOTE

HappyPack doesn't work with all webpack loaders as some loader API are not supported.

See this wiki page for more details on current Loader API support.

It is possible to omit this value and have HappyPack automatically infer the loaders it should use, see "Inferring loaders" below for more information

Defaults to: null

A unique id for this happy plugin. This is used to generate the default cache name and is used by the loader to know which plugin it's supposed to talk to.

Normally, you would not need to specify this unless you have more than one HappyPack plugin defined, in which case you'll need distinct IDs to tell them apart. See "Using multiple instances" below for more information on that.

Defaults to: "1"

Whether the plugin should be activated. This is for convenience when you want to conditionally disable HappyPack based on, for example, an environment variable.

Defaults to true

Path to a folder where happy's cache and junk files will be kept. It's safe to remove this folder after a run but not during it!

Defaults to: .happypack/

Whether HappyPack should re-use the compiled version of source files if their contents have not changed since the last compilation time (assuming they have been compiled, of course.)

Recommended!

Defaults to: true

Path to a file where the JSON cache will be saved to disk and read from on successive webpack runs.

Defaults to .happypack/cache--[id].json

An object that is used to invalidate the cache between runs based on whatever variables that might affect the transformation of your sources, like NODE_ENV for example.

You should provide this if you perform different builds based on some external parameters. THIS OBJECT MUST BE JSON-SERIALIZABLE.

Defaults to: {}

A function that computes a signature for a file. This signature is used by the cache to figure out whether the file contents have changed since the last time it was compiled.

The function signature is:

(filePath: String) -> String

Defaults to: a function that yields the last-modified-at (mtime) timestamp of the given file.

This number indicates how many Node VMs HappyPack will spawn for compiling the source files. After a lot of tinkering, I found 4 to yield the best results. There's certainly a diminishing return on this value and increasing beyond 8 actually slowed things down for me.

Keep in mind that this is only relevant when performing the initial build as HappyPack will switch into a synchronous mode afterwards (i.e. in watch mode.) Also, if we're using the cache and the compiled versions are indeed cached, the threads will be idle.

A custom thread-pool to use for retrieving worker threads. Normally, this is managed internally by each HappyPlugin instance, but you may override this behavior for better results.

See "Shared thread pools" below for more information about this.

Defaults to: null

Enable this to log messages from internal HappyPack modules to STDOUT; useful for troubleshooting sometimes.

Defaults to: false

HappyPack sits between webpack and your primary source files (like JS sources) where the bulk of loader transformations happen. Every time webpack resolves a module, HappyPack will take it and all its dependencies, find out if they need to be compiled[1], and if they do, it distributes those files to multiple worker "threads".

Those threads are actually simple node processes that invoke your transformer. When the compiled version is retrieved, HappyPack serves it to its loader and eventually your chunk.

[1] When HappyPack successfully compiles a source file, it keeps track of its mtime so that it can re-use it on successive builds if the contents have not changed. This is a fast and somewhat reliable approach, and definitely much faster than re-applying the transformers on every build.

For configuration convenience, we can instruct a HappyPack plugin to use the loaders specified in webpack's original module.loaders config by defining a happy: ... option on that specific loader.

For example, the following config will cause HappyPack to work with the babel loader against all .js source files:

// @file: webpack.config.js
module.exports = {
  module: {
    loaders: [{
      test: /\.js$/,
      loaders: [ 'babel' ],
      happy: { id: 'js' } // <-- see this
    }]
  },

  plugins: [
    new HappyPack({ id: 'js' }) // no need to specify loaders manually, yay!
  ]
}

Disclaimer

Using this method to configure loaders will cause HappyPack to overwrite webpack's loader options objects to replace the source loaders with happy's loader at run-time (ie, the config will be irreversibly mutated!).

I did not find any way to work around this, so be warned!

It's possible to define multiple HappyPack plugins for different types of sources/transformations. Just pass in a unique id for each plugin and make sure you pass it their loaders. For example:

// @file webpack.config.js
exports.plugins = [
  new HappyPack({
    id: 'jsx',
    threads: 4,
    loaders: [ 'babel-loader' ]
  }),

  new HappyPack({
    id: 'coffeescripts',
    threads: 2,
    loaders: [ 'coffee-loader' ]
  })
];

exports.module.loaders = [
  {
    test: /\.js$/,
    loaders: [ 'happypack/loader?id=jsx' ]
  },

  {
    test: /\.coffee$/,
    loaders: [ 'happypack/loader?id=coffeescripts' ]
  },
]

Now .js files will be handled by the first Happy plugin which will use babel-loader to transform them, while .coffee files will be handled by the second one using the coffee-loader as a transformer.

Note that each plugin will properly use different cache files as the default cache file names include the plugin IDs, so you don't need to override them manually. Yay!

Normally, each HappyPlugin instance you create internally manages its own threads which are used to compile sources. However, if you have a number of plugins, it can be more optimal to create a thread pool yourself and then configure the instances to share that pool, minimizing the idle time of threads within it.

Here's an example of using a custom pool of 5 threads that will be shared between loaders for both JS and SCSS/LESS/whatever sources:

// @file: webpack.config.js
var HappyPack = require('happypack');
var happyThreadPool = HappyPack.ThreadPool({ size: 5 });

module.exports = {
  // ...
  plugins: [
    new HappyPack({
      id: 'js',
      threadPool: happyThreadPool
    }),

    new HappyPack({
      id: 'styles',
      threadPool: happyThreadPool
    })
  ]
};

For the main repository I tested on, which had around 3067 modules, the build time went down from 39 seconds to a whopping ~10 seconds when there was yet no cache. Successive builds now take between 6 and 7 seconds.

Here's a rundown of the various states the build was performed in:

The builds above were run on Linux over a machine with 12 cores.

TODO: test against other projects

2.1.2

• Process argv will no longer be passed to the child processes spawned by HappyPack (#47)
• Support for the target loader context variable. Thanks to @Akkuma (#46)

2.1.1

• Fix an issue with loading invalid/corrupt cache or source-map files. Big thanks to @benhughes for providing the patch in #42

2.1.0

• SourceMap support

2.0.6

• a new option cacheSignatureGenerator to handle use cases such as [#35]

2.0.5

• now using mkdirp for creating the temp directory to support nested ones

2.0.4

• fix an issue with cache not being utilized on node v0.10 (fs.statSync doesn't exist with that name there - thanks to [@XVincentX])

2.0.2

• Fixed an issue that was causing loaders running in foreground to not receive the compiler options

2.0.1

• Package in NPM is now compact

2.0.0

• Pitching loader support
• More complete loader API support
• More convenient configuration interface

1.1.4

• Fixed an issue where the cache was being improperly invalidated due to cacheContext not being stored properly (#17, thanks to @blowery)

1.1.3

• Fixed an issue where the initial cache was not being saved properly

1.1.2

• Fixed an issue on old node versions (0.10) with the EventEmitter API (#10)
• Fixed an issue that was breaking the compiler if an invalid threads option was passed (evaluating to NaN)

1.1.1

• Unrecognized and invalid config parameters will cause the process to abort.
• The active version is logged on launch.

1.1.0

• now supporting basic webpack loaders
• dropped the transformer parameter as it's no longer needed
• cache now defaults to true
• now using a forking model utilizing node.js's process.fork() for cleaner threading code

1.0.2

• the loader will now accept IDs that aren't just numbers