I’m also considering an asynchronous approach, which might break some of the OP’s constraints.
My application already generates UUIDs for persisted documents, but we now want to additionally generate sequential IDs in the form of monotonically incrementing integers. Occasional gaps are permissible but collisions are not. I’m not using postgres as a storage backend and would prefer not to introduce it for just this purpose.
So I’m considering 2 different approaches:
- Use Swarm to manage
Counterprocesses, dumping state to e.g. S3 on shutdown - Enqueue an event to asynchronously generate the integer ID after document creation
The tradeoff to option 1 is Counters scale linearly with document types, and S3 failures (though unlikely) would be borderline unrecoverable. Option 2’s big downfall, IMO, is that we’d have to accept eventually consistent IDs.
:ets interests me as a possibility, but I don’t have much experience with it. I’m wary of any gotchas in a distributed setting
