Oh no, I missed this one, thanks for linking! This is a really good write-up and there are a lot of similarities with what I'm designing for strfry proxy. Like some of the comments, I'm not sure if partitioning on ID will be optimal in all cases. I can also imagine variations on pubkey and created_at.

Some variant of consistent hashing will also be necessary I think, for failure recovery, rebalancing, changing the number of backend relays, etc.