Careful with Caching

August 7th, 2022
tech
A friend recently shared this graphic on Facebook:

(I've added the "wrong" overlay.)

This is clearly incorrect: there's no way Massachusetts has more prisons than colleges. (MA actually has the largest ratio of colleges to prisons in the US.) After putting a link to the original source in the Facebook discussion, however, we found something pretty weird: people on mobile were seeing the incorrect map, but people on desktop were seeing a corrected one:

It turns out that Facebook was appending a tracking parameter, ?fbclid=... on desktop, but not on mobile. Normally this wouldn't do anything, because the site would ignore that in determining what page to return, but this site is apparently configured with a cache.

Many sites use caches to make it easier to serve pages. If you ask them for a page they'll give it to you (which might require a lot of work to generate) and then save a copy. Then when someone else asks for the same page, they can return the saved copy instead of putting in all that work to regenerate it. The site has a cached copy of /usa-prison-v-college in its cache with outdated content, but since ?fbclid=... is always followed by a new token those requests will never be found in the cache, and they return the current, corrected, page.

If you have a cache, what do you do when you change the page, like they did here? There are two main approaches:

  • Have a way to tell the cache the page has changed and it should forget its copy.

  • Always cache for a short time. Even just one minute can take a lot of load off a server that is getting thousands of requests for a hot page.

In this case they apparently didn't do either: we were running into this yesterday, and as of right now the site is still returning incorrect data.

(On the original question, comparing the number of colleges to the number of prisons is pretty silly: if one state runs large prisons and small colleges is that any better than a state that does the reverse? Comparing the number of people in prison vs college would make much more sense.)

Comment via: facebook, lesswrong

Recent posts on blogs I like:

Development RCTs Are Good Actually

In defense of trying things out

via Thing of Things March 25, 2024

Clarendon Postmortem

I posted a postmortem of a community I worked to help build, Clarendon, in Cambridge MA, over at Supernuclear.

via Home March 19, 2024

How web bloat impacts users with slow devices

In 2017, we looked at how web bloat affects users with slow connections. Even in the U.S., many users didn't have broadband speeds, making much of the web difficult to use. It's still the case that many users don't have broadband speeds, both …

via Posts on March 16, 2024

more     (via openring)