A lot of things in a typical social media service can be segmented trivially. The hardest part is the feed because it requires querying many database servers at once, but then it can be cached and served quicker.
Isn’t cache invalidation the hardest part of the feed? Once a user with millions of followers tweets, doesn’t that essentially invalidate millions of caches? That doesn’t sound like a trivial problem.
Because you can't edit tweets, the database is monotonically increasing (append-only) so scaling it isn't that hard. Cache invalidation basically just adds a new item. Deleting a tweet actually would be the harder/more expensive operation, but it's also less common.
Editing a tweet is also easy because the feeds would only store its id. You'd only need to invalidate one cache where the tweet object itself is stored.
