Hinted handoff
When a replica is down, the write is stored at a neighbour node as a "hint" — on reconnection the hints are replayed so the data eventually becomes consistent.
The Dynamo paper (Amazon, 2007); a standard technique in Cassandra, ScyllaDB and Riak. Lets writes succeed even though part of the replica set is down (= better availability under partial failure). Hints expire after some time (default 3 h in Cassandra) — eternal hints fill disks.
Combined with read repair (fix at read time when replicas diverge) and anti-entropy repair (periodic full sync via Merkle trees) as three layers of consistency convergence.