Mathieu Borderé
Mathieu Borderé
SQLite doesn't support the "DROP database" SQL statement and says to delete the database file instead. In dqlite's case it lives in memory. SQLite's xDelete VFS method to delete the...
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...
@ktsakalozos Has reported that running `set sysctl -w fs.aio-max-nr=0` on a running node will quickly result in the node running out of disk due to creation of a lot of...
I will complete this issue with the results of experiments and measurements, it's more of a placeholder right now.
I think we should do more checks within the code to assert that raft index of entries being appended to the log corresponds to the index that the uv backend...
There have been some discussions and issues regarding how raft determines the start index of an open segment. The issue first popped up in canonical/raft#264 where the concept of a...
Detected in jepsen run: https://github.com/canonical/jepsen.dqlite/actions/runs/6188624145/job/16801028094 artifact: [jepsen-data-append-partition,member-failure.zip](https://github.com/canonical/raft/files/12616404/jepsen-data-append-partition.member-failure.zip)
In addition to the versions installed by the package managers, add runs to the Github actions workflows that test against the latest released SQLite. Would have caught #564 earlier.
It's possible for dqlite to request a `raft_barrier` through `leader_barrier` while a `raft_barrier` is already in progress. It happens in one case where a new raft leader tries to commit...
Make sure all outstanding requests on the gateway are properly replied to when we close the gateway or when the node of which the gateway is part of loses leadership,...