How did I cut in half my project's webpack build time ?
Who never complained about the infinite duration of a webpack build on a project ?
I'm currently working on a big web application coded in React/Redux with server side rendering.
The application exists since 2015 and it has evolved a lot since then (M6web tech blog).
TLDR;
Never, ever, ever, ever work on performance improvements or optimization without monitoring!
If you want to optimize the duration of a job, you have to monitor precisely the duration of it and all its sub-steps.
By doing that, you can really focus on the most expensive task.
This will save you from wasting time on optimizations that will have little impact on the system as a whole.
Use existing monitoring tools! Create them if they don't exist!
What was the problem with webpack ?
For several weeks/months my colleagues had been complaining about the duration of our yarn build
command.
The purpose of this command is to build the distributable package of our application in a production target with webpack.
I even heard:
- "This command, I don't run it locally anymore, it takes too much time."
- "My computer starts ventilating heavily every time I run this command. There's nothing else I can do!"
Depending on the machine on which the build was launched, it took between 5 and 12 minutes.
It is not possible to have a build that takes so long.
webpack
is not a slow bundler.
It is our use of webpack
that makes it slow.
Focus error, a morning lost
Since this command launches a webpack build in production
mode, I figured out that the culprit was webpack config itself.
Given that I've dug deep into webpack, I thought it would be interesting to focus on this performance concern.
I have indeed open sourced a set of workshop to learn how to use webpack from scratch (https://webpack-workshop.netlify.com).
So at the end of January I took one day to improve the situation.
I had my own idea of the task that would take the most. So I tried to improve it, spending my entire morning on it.
I just managed to gain 17 seconds.
I'm not going to lie, I was very disappointed with what I achieved.
The concern in my strategy was however obvious.
I started off with a preconceived idea "This is definitely the stage that takes the longest."
Nothing was objective in my analysis.
To improve the performance of an application it is necessary to focus on objective facts.
Successful afternoon
When I came back from my lunch break, I was motivated to win more than those poor 17 seconds.
Then I remembered the Pareto principle.
The Pareto principle (also known as the 80/20 rule, the law of the vital few, or the principle of factor sparsity) states that, for many events, roughly 80% of the effects come from 20% of the causes.
Wikipedia
There is probably one step that takes up most of the webpack build time.
Pareto principle adapted to webpack could mean "80% of the build time is caused by 20% of the config"
Let's find the culprit ! 🎉
I had to determine the build time of each loader, of each plugin.
I was very lucky, the webpack community has already proposed a plugin that allows to measure everything.
And it is very easy to install. ♥️
stephencookdev / speed-measure-webpack-plugin
⏱ See how fast (or not) your plugins and loaders are, so you can optimise your builds
The first step to optimising your webpack build speed, is to know where to focus your attention.
This plugin measures your webpack build speed, giving an output like this:
Install
npm install --save-dev speed-measure-webpack-plugin
or
yarn add -D speed-measure-webpack-plugin
Requirements
SMP requires at least Node v6. But otherwise, accepts all webpack versions (1, 2, 3, and 4).
Usage
Change your webpack config from
const webpackConfig = {
plugins: [new MyPlugin(), new MyOtherPlugin()],
};
to
const SpeedMeasurePlugin = require("speed-measure-webpack-plugin");
const smp = new SpeedMeasurePlugin();
const webpackConfig = smp.wrap({
plugins: [new MyPlugin(), new MyOtherPlugin()],
});
and you're done! SMP will now be printing timing output to the console by default.
Check out the examples…
Here are the results I got:
SMP ⏱
General output time took 4 mins, 5.68 secs
SMP ⏱ Plugins
IgnorePlugin took 57.73 secs
TerserPlugin took 39.022 secs
ExtractCssChunksPlugin took 3.13 secs
OptimizeCssAssetsWebpackPlugin took 1.6 secs
ManifestPlugin took 1.55 secs
WebpackPwaManifest took 0.326 secs
ContextReplacementPlugin took 0.129 secs
HashedModuleIdsPlugin took 0.127 secs
GenerateSW took 0.059 secs
DefinePlugin took 0.047 secs
EnvironmentPlugin took 0.04 secs
LoadablePlugin took 0.033 secs
Object took 0.024 secs
SMP ⏱ Loaders
babel-loader, and
rev-replace-loader took 2 mins, 11.99 secs
module count = 2222
modules with no loaders took 1 min, 57.86 secs
module count = 2071
extract-css-chunks-webpack-plugin, and
css-loader, and
postcss-loader, and
sass-loader took 1 min, 43.74 secs
module count = 95
css-loader, and
postcss-loader, and
sass-loader took 1 min, 43.61 secs
module count = 95
file-loader, and
rev-replace-loader took 4.86 secs
module count = 43
file-loader took 2.67 secs
module count = 32
raw-loader took 0.446 secs
module count = 1
@bedrock/package-json-loader took 0.005 secs
module count = 1
script-loader took 0.003 secs
module count = 1
As expected, it's not great!
But at least I'm starting to get who the culprits are.
We can see that for 2222 Javascript modules takes up 2mins but for only 95 Sass files 1min43 🤣.
Damn node-sass
Once the migration from node-sass
to sass
(new Sass re-implementation) and the update of sass-loader
, I was shocked!
It took me about 10 minutes because there were few breaking changes and I gained more than 1min30 on the build time.
sass-loader
made big improvements on performances, you should definitely make sure you use the last version.
I lost a morning on gaining 17 seconds and I spent 10 minutes to win 1min30.🤣
IgnorePlugin, TerserPlugin
TerserPlugin
is used to uglify the javascript code in order to reduce its size and readability. It's a relatively long process, but 39 seconds is too much.
Just by updating the version of TerserPlugin to use the one integrated in Webpack, I managed to reduce by 20 seconds the build time.IgnorePlugin
is a core plugin that was used a lot in our application to avoid loading certain scripts in order to reduce the weight of the site.
It was necessary, but today with Webpack we can use much better than that. Dynamic Import, ContextReplacement, there are plenty of solutions. As a general rule, we should avoid compiling files and then not using them.
Recommendations from the community
To improve the build perfs webpack provides a web page listing the actions to take to hunt what takes time.
I strongly advise to have a look at it.
https://webpack.js.org/guides/build-performance/
Final Result
SMP ⏱
General output time took 2 mins, 18.27 secs
Based on precise and concrete measures, I was able to drastically improve the webpack build of my application.
No more computers suffering just to compile a bit of JS and SASS.
I could have lost whole days on futile modifications if I had not measured precisely what penalized the build.
ℹ️
- Use
Speed Measure Plugin
to debug webpack build time - Track your build time evolution to detect big evolution before merge
- Follow webpack performances recommandations
- Look at webpack 5 new caching strategies
- Keep your webpack config up to date
Top comments (1)
It as really interesting read.
I went ahead to search for sass instead of node-sass. I found following note over there
Now I am wondering that did I land on correct sass implementation?
The README says it is slow but you gained on build time. Can you please elaborate more?