Read Repair

When a client makes a read from several nodes in parallel, it can detect any stale responses and write newer values back to stale replicas. This works well for frequently-read values.