Ask HN: How to performance test React webapps for rendering speed regressions?
As React webapps get large and start to contain a lot of complex components, they often do not perform well and manifest low FPS in browser. As a result, engineers add strategies such as memoization (React.memo, shouldComponentUpdate) to avoid unnecessary computation, however they logic is fragile to code changes.
Is there a way to automatically test performance (maybe in terms of FPS) of React applications regularly in CI pipelines? Any tools for this out there?
For frontend performance: record the frame rate (via requestAnimationFrame, look at simple frame rate counters), memory, count of DOM nodes on the page during each of your integration tests as a good start. You can get a lot of this via window.performance and memory info you need to pass an extra flag into chrome headless.
It’s also useful to instrument this in your production environment (called real user metrics) for some % of user sessions and combine it with more synthetic tests.
The backend is of course really easy, just measure request times (min/avg/max) and time to first byte.
Edit: Don’t know why I’m being downvoted for this answer? Lol
Thanks a lot, this is something I was specifically looking for!
Don't mind the downvotes. Thank you for taking the time to post such an insightful post.
+1 on instrumenting production. Best way to find out regressions, and future optimizations.
I don't do front end, but I asked my team and they linked me to:
https://medium.com/@paularmstrong/twitter-lite-and-high-perf...
https://github.com/markerikson/react-redux-links/blob/master...
Might be worth going through. Hopefully they're useful.
Hey, that's my links list! :) Hope that's helpful!
While it's probably not 100% relevant, you might also want to glance at some of the things I've done trying to benchmark performance for the React-Redux library itself, including running benchmarks in Puppeteer and capturing metrics like render times and FPS [0].
Now, my experience has been that I have to deliberately crank up the number of components on screen and dispatch absurd numbers of Redux actions a second just to get the FPS to drop below 60FPS, so I'm not sure how much that would be applicable to real-world apps, but at least you can see how I've tried to capture the values.
As a general observation, I'd highly encourage folks to use the Profiler that's built into the React DevTools to view render behavior and find components that may be re-rendering when it wasn't necessary [1] [2].
Also, Brian Vaughn has been working on a major rewrite of the React DevTools to both improve internal speed and add new features, and you can play around with it now [3].
Finally, if you _really_ want to get fine-grained, there's a little-known profiling/tracing API that can be used to capture render times in parts of your app [4] [5] [6].
[0] https://github.com/reduxjs/react-redux-benchmarks
[1] https://reactjs.org/blog/2018/09/10/introducing-the-react-pr...
[2] https://www.netlify.com/blog/2018/08/29/using-the-react-devt...
[3] https://github.com/bvaughn/react-devtools-experimental
[4] https://github.com/bvaughn/rfcs/blob/profiler/text/0000-prof...
[5] https://gist.github.com/bvaughn/8de925562903afd2e7a12554adcd...
[6] https://gist.github.com/bvaughn/25e6233aeb1b4f0cdb8d8366e54a...
Amazing list! Thanks! Just pinged you on Twitter too. I'll add a note here for anyone interested that I've been looking at performance testing as part of our functional/regression testing strategy. Haven't used it yet, but this Jest library with puppeteer looks promising: https://github.com/jj4th/jest-puppeteer-performance
Even better these days, just use the profiler in Chrome. It’ll show you which components are mounting and rendering. I find that I can isolate troublesome components at a glance now.
Before you spend time profiling a performance issue, make sure you're using the production build of React. The development build is significantly slower. It sounds obvious, but I've made the mistake of trying to optimize performance issues that simply didn't exist in production.
At Asana, we actually test performance in production with a A/B test split. We use Interana (but any data aggregator would work) to analyze the results. Trying to measure the regressions in a sandbox did not match our Real User Metrics (RUM). In order for this to work though, you need to develop a strong mental model of React performance so you can optimize it further.
Do you have a blog post or something like that about React performance or the mental model you got? It would be an interesting read.
This is probably the best one we have in public https://blog.asana.com/2017/08/performance-asana-app-rewrite.... If you have specific questions, we would be happy to write a follow up one! A lot of the work we've done has been measuring time from request to finished render.
A few ideas:
- https://firebase.google.com/docs/perf-mon/get-started-web (this is a new feature released by Google. Luckily you don't need Firebase to use this. https://www.youtube.com/watch?v=KYYjdWSrRjI&t is a good tutorial on wiring your app up)
- https://reactjs.org/docs/perf.html
- lighthouse by Google
You could just run lighthouse on your app periodically for an in depth audit on your site. That way you wouldn't need to slow down your builds with a test (not sure what's best practice here. Just trying to throw out some basic ideas)
> As React webapps get large and start to contain a lot of complex components, they often do not perform well and manifest low FPS in browser.
Is this something you have actually experienced? Can you cite an example of running into this issue? I've built fairly complex UIs with React (tables with nested dropdowns, financial statements with a lot of data points, dynamic charts / graphs, elements with animations, etc. and haven't run into this issue.
The only time I've seen people have trouble w/ performance is when they use something like Redux or Apollo on top of React. The problem in those cases of course is Redux or Apollo, not React itself.
I'm not OP but at work we have a actually not-so-complex website that takes a full second to render a new page when switching, on my MacBook. On production and without loading anything via the network.
I blame it on create-react-app, which feels very very bloated to me. Whenever I write React from scratch it feels super performant, even with Redux and much more complex layouts than the page I mentioned.
Hi,
as a Create React App maintainer I'd love if you could share some scenarios where CRA itself would cause performance issues. CRA itself is mostly just developer tooling and every other feature is opt-in. The team and many contributors (including people on WebPack team) have spent quite a lot of time on optimizing the production builds CRA produces; building an React only app won't have much else than React and React-DOM itself, while we also automatically do more advanced things like bundle spitting too.
Create React app includes almost no extra client facing code other than react. There is an issue with how you made the application.
That's like saying VSCode makes your site slower.
Many of the answers on this post are confusing web performance for rendering speed.
Looking for the same. Ideally something that integrates well with Storybook or Testcafe.
WebdriverIO recently got support for Lighthouse, but also the devTools plungin has support for page load.
If I were to set this up I'd use devtools to throttle the network down to 3g, and I would actually render the browser in the test.
https://github.com/webdriverio/webdriverio/tree/master/packa...
A few resources I've found useful:
- lighthouse (https://github.com/GoogleChrome/lighthouse). You can implement this with their Node CLI to automate in your pipeline and output the results somewhere.
- calibre (https://calibreapp.com/). Gives you nice graphs and snapshots over time.