2024-12-01
Aaron and I fixed an issue with WPD today, I made the PR and he gave it a sanity check.

Issue 22 - Concurrent requests caused the cache (aiohttp-client-cache), which was using sqlite, to lock and fail.

This is a pretty big bottleneck, and was reported as early as June - Exception thrown when executing multiple parallel requests. As more and more users started using the main instance, database errors became a lot more frequent, second to ratelimits.

The solution? Switching to Redis. aiohttp-client-cache can cache to Redis instead of /tmp or an sqlite database. I setup an env file, with type validation - that took a bit of time. The cache setup itself was easy, a few lines and an if-else to decide between file or redis-based caching.

The associated pull request also adds some other stuff, namely:

• Eliot logging (I'm a fan of this now lol, just wish it didn't add a level of indentation)
• Type validation for API Responses
• Cache efficiency - images aren't cached anymore
• Cancel requests when the client disconnects
• Save requests where possible - this adds metadata retrieval to a part_id -> story_id request, it's comforting to have type validation when making embedded requests like this.
• The frontend allows IDs to be typed in again, which is something people missed
• Readability and standardization changes

late night programming again, I made a few silly mistakes that Aaron caught and fixed, love how seriously he takes testing lol - something to learn from him.