Stalled but Safe: Raft and Paxos in Practice

Deep End Era. Engineering with Sankalp · Oct 28, 2026 · 59m

One afternoon, five healthy machines stop accepting writes because they cannot agree which of them is in charge. Episode one hundred and forty three explains why that can happen and why it is the system working as designed: a 1985 proof says agreement cannot be guaranteed when any machine might crash and messages can be delayed, so practical algorithms choose never to be wrong rather than always to answer. We work through Lamport's Paxos, born as a joke about a Greek parliament, and Raft, designed to be understood, then follow them into real outages at Cloudflare and GitHub, the Jepsen tests…