Hinted handoff temporarily stores a copy of a write when its intended replica is unavailable, then replays that write after the replica recovers. section 8.4.3
The process works as follows:
- A coordinator detects that a replica is offline or fails to respond within the server-side write timeout. section 8.4.3
- The coordinator writes a local hint containing a copy of the write request. section 8.4.3
- When the replica returns, the coordinator replays the saved requests and hands the missed writes to that node. section 8.4.3
- Hint generation stops after the configured max_hint_window_in_ms period; regular repairs and read repairs can then reconcile missed data. section 8.4.3
A replica goes offline during a write
Suppose a write reaches a coordinator while one replica is down. The coordinator saves a hint locally instead of losing the intended update. After the replica recovers, it replays the hint. This behavior also enables consistency level ANY to accept a write through a nonreplica node when all replicas are unavailable. section 8.2.2 section 8.4.3
Figure 8.20 In a hinted handoff, nodes send missed writes to offline nodes once they recover.
Trace the recovery path in which nodes send missed writes to offline replicas. section 8.4.3
