Raft

Back to Consensus Algorithms

A consensus algorithm designed for understandability. Separates consensus into leader election, log replication, and safety. A leader is elected to manage the replicated log; followers replicate the leader’s log. Used in production by etcd (Kubernetes), CockroachDB, Consul, and TiKV.

Key Properties

  • Paxos (equivalent capability, harder to understand)
  • Leader Election (Raft includes built-in leader election)

distributed-systems consensus