Performance

RelativeCI - In-depth bundle stats analysis and monitoring - Interview with Viorel Cojocaru

Webpack's performance out of the box is often enough for small projects. That said, it begins to hit limits as your project grows in scale, and it's a frequent topic in webpack's issue tracker.

There are a couple of ground rules when it comes to optimization:

  1. Know what to optimize.
  2. Perform fast to implement tweaks first.
  3. Perform more involved tweaks after.
  4. Measure the impact as you go.

Sometimes optimizations come with a cost. You could, for example, trade memory for performance or end up making your configuration more complicated.

If you hit memory limits with webpack, you can give it more memory with node --max-old-space-size=4096 node_modules/.bin/wp --mode development kind of invocation. Size is given in megabytes, and in the example you would give 4 gigabytes of memory to the process.

Measuring impact#

As discussed in the previous chapter, generating stats can be used to measure build time. webpack.debug.ProfilingPlugin and cpuprofile-webpack-plugin are able to emit the timings of plugin execution as a file you can pass to Chrome Inspector. The latter generates a flame graph as well.

High-level optimizations#

Webpack uses only a single instance by default, meaning you aren't able to benefit from a multi-core processor without extra effort. This is where solutions like thread-loader come in. webpack-plugin-ramdisk writes the build output to a RAM disk and it can help during development and in case you have to perform many successive builds.

Low-level optimizations#

Specific lower-level optimizations can be nice to know. The key is to allow webpack to perform less work. Consider the examples below:

  • Use faster source map variants during development or skip them. Skipping is possible if you don't process the code in any way.
  • Use @babel/preset-env to transpile fewer features for modern browsers and make the code more readable and more comfortable to debug while dropping source maps.
  • Skip polyfills during development. Attaching a package, such as core-js, to the development version of an application adds processing overhead.
  • Polyfill less of Node and provide nothing instead. For example, a package could be using Node process which in turn will bloat your bundle if polyfilled. See webpack documentation for the default values.
  • Starting from version 5, there's a file system level cache that can be enabled by setting cache.type = "filesystem". To invalidate it on configuration change, you should set cache.buildDependencies.config = [__filename]. Webpack handles anything watched by the build automatically including plugins, loaders, and project files.

Loader specific optimizations#

Loaders have their optimizations as well:

  • Perform less processing by skipping loaders during development. Especially if you are using a modern browser, you can skip using babel-loader or equivalent altogether.
  • Use either include or exclude with JavaScript specific loaders. Webpack traverses node_modules by default, and executes babel-loader over the files unless it has been configured correctly.
  • Parallelize the execution of expensive loaders using thread-loader. Given workers come with an overhead in Node, the loader is worth it only if the parallelized operation is heavy.

Optimizing rebundling speed during development#

Rebundling times during development can be improved by pointing the development setup to a minified version of a library, such as React. In React's case, you lose propType-based validation but if speed is paramount, this technique is worth it.

module.noParse accepts a RegExp or an array of RegExps. In addition to telling webpack not to parse the minified file you want to use, you have to point react to it by using resolve.alias. The idea is discussed in detail in the Consuming Packages chapter.

You can encapsulate the idea within a function:

exports.dontParse = ({ name, path }) => ({
  module: { noParse: [new RegExp(path)] },
  resolve: { alias: { [name]: path } },
});

To use the function, you call it as follows:

dontParse({
  name: "react",
  path: path.resolve(
    __dirname, "node_modules/react/cjs/react.production.min.js",
  ),
}),

After this change, the application should be faster to rebuild, depending on the underlying implementation. The technique can also be applied to production.

Given module.noParse accepts a regular expression if you wanted to ignore all *.min.js files, you could set it to /\.min\.js/.

Not all modules support module.noParse. They should not have a reference to require, define, or similar, as that leads to an Uncaught ReferenceError: require is not defined error.

Webpack 4 performance tricks#

There are various webpack 4 specific tricks to improve performance:

  • If output.futureEmitAssets is set, webpack 5 related logic is enabled. Based on Shawn Wang, it reduces memory usage and improves situation.
  • Sometimes there are version related performance regressions which can be fixed in the user space Kenneth Chau has compiled a great list of them for webpack 4. The main ideas are related to simplifying stats.toJson using ts-loader with experimentalWatchApi and setting output.pathinfo to false.
  • Jared Palmer mentions that setting optimization property and its splitChunks, removeAvailableModules, and removeEmptyChunks properties to false can improve performance in the development mode.

Conclusion#

You can optimize webpack's performance in multiple ways. Often it's a good idea to start with more accessible techniques before moving to more involved ones. The exact methods you have to use depend on the project.

To recap:

  • Start with high-level techniques that are fast to implement first.
  • Lower level techniques are more involved but come with their wins.
  • Since webpack runs using a single instance by default, parallelizing is worthwhile.
  • Especially during development, skipping work can be acceptable thanks to modern browsers.
The official build performance guide and Web Fundamentals by Google have more tips.
Previous chapter
Build Analysis
Next chapter
Output

This book is available through Leanpub (digital), Amazon (paperback), and Kindle (digital). By purchasing the book you support the development of further content. A part of profit (~30%) goes to Tobias Koppers, the author of webpack.

Need help?