19-Dec

How to force-refresh your app every once in a while

3 min read

By Kristofer Selbekk

December 19, 2020

Today's users tend to keep your web site open for ages. No, not because they love using it all the time, but because they opened it in a tab - most likely on their phone - and now it's there forever.

I log all errors in my web apps to Sentry, and every time I deploy a new version, I see a ton of these buggers cluttering my logs:

ChunkLoadError: Loading chunk 14 failed.
(missing: https://www.some-web.app/static/js/14.2df2391b.chunk.js)
  at None (/some-view:1:3732)
  at None (/bundles/0f61fb778be65ca983e6e18120fad3b0bfd49226.js:1:533023)
  at N (/bundles/0f61fb778be65ca983e6e18120fad3b0bfd49226.js:1:290176)
  at Pu (/bundles/0f61fb778be65ca983e6e18120fad3b0bfd49226.js:1:404668)
  at Cs (/bundles/0f61fb778be65ca983e6e18120fad3b0bfd49226.js:1:396050)
...
(9 additional frame(s) were not displayed)

The what and why

First - let me explain what a chunk is, and why this happens.

A chunk is - well - a chunk of JavaScript. They are created whenever you code-split your application. I typically end up splitting my app up per route, so I get a single chunk per page in my app. Whenever the user navigates to that particular page (or before, if I preload it), the relevant chunk is requested from the server.

I deploy most of my web apps to Netlify (like this one), and whenever I deploy a new version, they wipe the old files and upload the new ones (well, they create a new deployment without any of the stuff from the previous deploy). So whenever the user continues using my site after a deployment (often days later), and goes to visit a new route, it can't find the requested chunk!

The fix

Now, there are several ways to mitigate this problem. You could deploy to an Amazon S3 bucket instead, and keep all old built files forever), you could pre-load all of the resources whenever a user visits the landing page, or you could use a CDN for hosting your stuff, which will cache your files.

I didn't want to do either of those though, because I'm lazy, and felt inspired. So I found a great workaround - already built into React Router! 🤯

The BrowserRouter component accepts a forceRefresh prop, which will turn all of those client-side links into regular ol' <a /> tags. This will make the app receive a new version of your web app from the server, alleviating all potential issues with stale JavaScript and missing chunks. In addition, the user gets your newly deployed version sooner than they would have otherwise.

So let's say you want to force your app to be updated at least once every 6 hours. How would you implement something like that? Turns out - it's pretty simple!

First, we create a new wrapper component for the BrowserRouter component - we'll call it AutoRefreshingRouter:

import { BrowserRouter } from "react-router-dom";

export const AutoRefreshingRouter = ({ children }) => {
  return <BrowserRouter>{children}</BrowserRouter>;
};

Next, we create a state variable shouldRefresh and pass it to the BrowserRouter's forceRefresh prop.

import React from "react";
import { BrowserRouter } from "react-router-dom";

export const AutoRefreshingRouter = ({ children }) => {
  const [shouldRefresh, setShouldRefresh] = React.useState(false);
  return <BrowserRouter forceRefresh={shouldRefresh}>{children}</BrowserRouter>;
};

Finally, let's change that flag after an hour with a call to React's useEffect and a timeout!

import React from "react";
import { BrowserRouter } from "react-router-dom";

export const AutoRefreshingRouter = ({ children }) => {
  const [shouldRefresh, setShouldRefresh] = React.useState(false);
  React.useEffect(() => {
    const id = setTimeout(
      () => setShouldRefresh(true),
      1000 * 60 * 60 * 6 // 6 hours in milliseconds
    );
    return () => clearTimeout(id);
  }, []);

  return <BrowserRouter forceRefresh={shouldRefresh}>{children}</BrowserRouter>;
};

Here, we ask the browser to wait 6 hours, and then set the shouldRefresh flag to true. We make sure to clear the timeout in the clean-up return value function if we ever re-mount the component.

And that's it! Since I deployed this fix, I haven't seen a single bug in production. My users are happier, and I don't get a bad feeling whenever I see Sentry telling me I screwed up.

Hope this helps you in your project as well. Thanks for reading!