Open MathieuBordere opened 1 year ago
I tried to address this previously in canonical/raft#321 (see also canonical/raft#320), but wouldn't be surprised if I missed something there.
Allright, it was related to behaviour observed by the reported of https://github.com/canonical/dqlite/issues/588 maybe the raft implementation was not recent enough.
Indeed, that merge commit is only present in Raft v0.16.0 and up, and microk8s v1.26 builds with Raft v0.14.0.
We should gracefully handle a leader node running out of disk. Currently the leader will not step down when running out of disk, resulting in an unavailable cluster. A first step could be for the leader to just step down if it detects an out of disk state.