They used to have a really good blog post about their caching infra which for some reason was deleted. Archive link: https://web.archive.org/web/20210205121832/https://redditblo...
> On March 14th, Apollo made nearly 1 billion requests against our API in a single day, triggered in part by our system outage. After the outage, Apollo started making 53% fewer calls per day. If the app can operate with half the daily request volume, can it operate with fewer?
That's a backend server making a request that doesn't count against its rate limit as soon as it can again.
The push notification backend is described as making a request every 6 seconds for each user. If it isn't backing off correctly and delaying a requeueing of the next job, that can drastically increase the request rate.
If a multitude of bots are written immediately retry upon receiving an error without backing off, and are now stuck in loops of flooding the servers with requests, could the number of requests become a greater factor than the simplicity of the responses?