search

willemt / raft

C implementation of the Raft Consensus protocol, BSD licensed
Other
1.13k stars 271 forks source link

Allow entries to apply during snapshot #86

Closed yossigo closed 6 years ago

yossigo commented 6 years ago

Currently raft_apply_entry() and raft_apply_all() will avoid applying entries if a snapshot is in progress. Assuming the FSM is able to provide the proper isolation, I don't see any real reason to prevent this in the library but @willemt perhaps you do?

This requires care in maintaining the different indexes, but we anyway must assume the commit index can advance during snapshots (see #84 for example).

willemt commented 6 years ago

That makes a lot of sense. For example, you could isolate a key value store FSM when it's being snapshotted by iterating over the keys and being aware of log entries setting values.

However, I like that the default behaviour is so strict. It makes it easier to not break things. Therefore I think a good way to allow this is by adding flags to raft_begin_snapshot. This is what I'm thinking about doing:

#define  RAFT_SNAPSHOT_NONBLOCKING_APPLY 1

int raft_begin_snapshot(raft_server_t *me_, int flags);
willemt commented 6 years ago

Good optimisation for FSMs that can use this