Explain the Raft consensus algorithm used in distributed databases.

Raft ensures distributed consensus for replicated state machines. Nodes are leader, follower, or candidate. Leader election: followers timeout waiting for heartbeat, become candidates, request votes. Leader handles all writes, replicates log entries to followers. Safety: committed entries are durable. Split-brain prevented by requiring majority quorum. Simpler than Paxos. Used in etcd, CockroachDB, TiKV. Handles leader failure, network partitions, and ensures linearizability.