Jepsen is useful for consistency checking, but also just as a way to drive the operation of a dqlite cluster while simulating various workloads and problematic events. Once we get a decent collection of bpftrace scripts for tracing dqlite/raft, we can combine them with the Jepsen code we've already written (plus some new workloads if that seems appropriate) to see how the database performs under a variety of circumstances and track the impact of our implementation work. I hope that this will be more flexible than writing ad-hoc drivers and workloads with go-dqlite, which is what we're doing currently.
Jepsen is useful for consistency checking, but also just as a way to drive the operation of a dqlite cluster while simulating various workloads and problematic events. Once we get a decent collection of bpftrace scripts for tracing dqlite/raft, we can combine them with the Jepsen code we've already written (plus some new workloads if that seems appropriate) to see how the database performs under a variety of circumstances and track the impact of our implementation work. I hope that this will be more flexible than writing ad-hoc drivers and workloads with go-dqlite, which is what we're doing currently.