Multi-leader
A replication model with several "leaders" that can all accept writes and then replicate to each other. Useful for geographically distributed datacenters and offline-first apps — but it requires conflict resolution when the same data is changed in two places at once.
Use case: one datacenter per region where each takes local writes (low latency) and replicates asynchronously to the others; or apps that must work offline (calendar, notes) and sync when they come online. Win: writes can happen near the user and the system survives a region/connection failing. The core problem: write conflicts — if the same record is changed in two datacenters, they must be merged deterministically (last-write-wins via timestamp, application logic, or CRDTs). Distinct from single-leader (simple but a bottleneck/SPOF) and leaderless (quorum-based). Complex but sometimes necessary. Related to conflict resolution and CRDT.