Distributed Consensus: The Raft Algorithm — Leader Election & Log Replication

🗳️ Raft Cluster

Nodes — click to kill / restore

Network partition

Healthy

Split 2|3

Split 1|4

Current term0

Leaderelecting…

Committed entries0

Alive nodes5 / 5

Partitionnone — healthy

Every Follower counts down its own randomized election timeout. The first to expire becomes a Candidate, bumps its term, and requests votes; a node grants at most one vote per term. Reaching a real majority (3 of 5) makes it Leader, which then replicates log entries and marks them committed once acknowledged by a majority. Kill the leader, split the network, or write a client entry and watch the cluster recover — or, on the minority side of a partition, correctly refuse to.

Gold ring = Leader · amber pulse = Candidate · grey = Follower · dim red = offline. Small squares under each node are its log — green means committed.