Post #1823384

2026-04-25 05:02 UTC

@flippac @jer_gib Yes, I am in fact conceiving of the "state" recorded by the snapshot algorithm as being an event log.

Replies (2)

@lindsey@recurse.social 2026-04-25 05:17

@flippac @jer_gib Also, the Chandy-Lamport snapshot algorithm (which is the one I was trying to write down safety and liveness specifications for!) isn't particularly fault-tolerant (it won't terminate if marker messages are lost, for instance), but, hey, at least it won't record a bad snapshot.

Open ##1823385
@flippac@types.pl 2026-04-25 13:51

@lindsey @jer_gib Yeah, I'd been thinking in terms of local node state when I said that because the idea a global snapshot can be derived without one in general strikes me as... well, not entirely compatible with a message passing model to begin with (if you've got a fixed node count you might be able to spot when something's become common knowledge in the way the n-player generalisation of rollback netcode for games works?)

Open ##1823386