A couple of weeks ago my microblog.pub instance stopped updating the feed. And I've been kind of procrastinating on fixing it until yesterday because… IDK, had too many unread books.

Anyway, I figured out the problem and fixed it (details are not really important), but I wanted to appreciate a design choice that I totally wouldn't have thought about. Upon receiving an event, the engine doesn't try to process it; instead it sticks it into the input queue, which is processed in the background. My feed wasn't updating because the processing of one of the queue elements got stuck, but! When I unwedged it, it started chugging through all the backlog of events it received in two weeks. It took about a way to process it, but not a single post was lost. Which is important for a federated system, because your peers won't resend them again.