Raft Consensus Explained

We build Raft from scratch: why distributed systems need consensus, what linearizability requires, and how Raft achieves it with a leader, a log, and a term clock. Every decision is justified by the failure it prevents.

Topics covered:

• The problem: why "just replicate" doesn't work without total order
• Leader election: why a random timeout prevents livelock
• Log replication: how the leader enforces ordering across followers
• The election restriction: why the leader must have the most up-to-date log
• Commit rules: what "committed" means and why a majority is required
• Safety invariants: leader completeness and election safety
• Network partitions: how Raft stays consistent when the network splits
• Performance: the cost of synchronous replication and how multi-raft scales