Leaderless replication
A replication model without a designated leader/primary — any replica can accept writes and reads. Consistency is instead achieved via quorums (W + R > N) and repair mechanisms. The Dynamo model, used by Cassandra and Riak.
Mechanic: data is replicated to N nodes; a write is counted as successful when W nodes have acknowledged, a read queries R nodes. If W + R > N, it's guaranteed that read and write overlap in at least one node → the latest value is seen. Since there's no leader, there's no single point of failure and writes can be accepted anywhere (good for high availability and multi-datacenter). Price: weaker consistency and a need for conflict resolution (versions can diverge) plus repair (read repair, anti-entropy). Distinct from single-leader (one write node) and multi-leader (several write nodes with sync). The client often chooses the consistency level per operation. The core of the AP side of CAP. Related to quorum and conflict resolution.