Read Repair
Summary: A mechanism in leaderless replication where a client that notices stale data while reading from multiple replicas automatically writes the newer value back to the stale replica.
Sources: chapter5
Last updated: 2026-04-15
Read repair is an opportunistic way of keeping replicas in sync. When a client makes a read from several nodes in parallel, it can detect stale responses by comparing the version numbers or timestamps (source: chapter5, p. 178).
If the client notices that one of the replicas is returning an older version of the data, it can send a write request back to that replica with the newer value it found (source: chapter5, p. 178).
Read repair is particularly effective for data that is frequently read.