Raft doesn’t magically make a database multi master. It’s a consensus algorithm that has a leader election process. There’s still a leader, and therefore a single master to which writes must be directed. The problem it solves is ambiguity about who the active master is at the moment.
Right. The idea of active-active OrioleDB multimaster is to apply changes locally and in parallel send it to the leader. Then sync on commit and ensure there is no conflicts.
Jepsen does much more than basic end-to-end tests, including intentionally partitioning the cluster. Tests written for a non-distributed system are downright friendly compared to what Jepsen does to distributed systems.
Those are very expensive and booked months in advance. Aphyr sometimes comments on HN and is of course a better source on this but that's what I recall reading.
