On Reliable Persistence
A short investigation on reliable persistence: Crash consistency is hard but possible at a huge performance cost Data safty must rely on distributed solution Raft Perhaps we should just give up manual fsync and rely on OS background flush corruption detection & correction at startup replication