The version we're working on now uses the go Raft implementation to handle cluster configuration. Then each server periodically (< 30s) performs read repairs to

The version we're working on now uses the go Raft implementation to handle cluster configuration. Then each server periodically (< 30s) performs read repairs to ensure that it has the latest writes for its spots in the hash ring. Writes can either be done via either write once or quorum, with write once being the most often used thing. Since most of the time this data is in aggregate, we don't care about losing a few writes.