My favourite is having to send the activities sequentially, meaning you can very easily block the queue when a request fails.

And that’s just the network architecture. Database architecture is another kind of hell. Like a simple delete operation taking multiple minutes because there’s a multitude of triggers, some of which take very long. That in itself is not bad, but the fact that the api waits for all the operations to succeed or fail (or the more usual case, timeout) is bonkers. Either fix the db or do it in the background.

I was excited for Lemmy a year and a half ago, which quickly passed. Thinking of migrating my server to some alternative. If Sublinks launches eventually, I’m migrating in an instance, currently thinking of writing an api compatibility layer between Lemmy and Piefed to migrate without anyone noticing.