7. High-Water Mark

Background

Distributed systems keep several copies of data. This gives fault tolerance and higher availability. One way to get strong consistency is a leader-follower setup. The leader handles every write. The followers replicate that data from it.

Each transaction on the leader is first committed to a write-ahead log (WAL). That is the log the leader replays, to recover from a crash. A write counts as successful as soon as it is written to the WAL. That WAL lives on the leader. Replication then happens asynchronously. The leader can push each mutation to the followers