Replication

Replication keeps multiple copies of the same data (usually across different machines, racks, or regions) so a single node failure doesn't lose data or take down reads/writes. The dominant topology is single-leader (primary-replica): one node accepts all writes and streams a change log to one or more followers, which apply it and serve reads. Multi-leader replication lets more than one node accept writes (useful across regions/datacenters) at the cost of needing conflict resolution when two leaders accept concurrent writes to the same record. Leaderless replication (Dynamo-style: Cassandra, Riak, DynamoDB) has any node accept a write and relies on quorum reads/writes (`W + R > N`) plus read-repair/anti-entropy to converge, trading strict ordering for very high availability. Replication can be synchronous (the primary waits for the replica to ack before confirming the write — zero data loss on failover, but higher write latency) or asynchronous (the primary confirms immediately and streams to replicas after — fast writes, but a crashed primary can lose the last few unreplicated writes).