Raft - The Understandable Distributed Protocol
summary
Raft presents concrete rules: leader election plus log replication, randomized 150–300 ms election timeout, and quorum commits that require a majority so split partitions cannot both commit. Implementation evidence includes go-raft and about 28 ports, production use in CoreOS SCD, and the LogCabin/RAMCloud reference. The protocol specifies rolling back uncommitted log entries after partitions, offers snapshotting strategies to bound logs, and treats membership changes as a separate, tricky procedure.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 0:17Distributed Consensus
- 1:08Paxos
- 1:45Roles
- 2:54Raft
- 3:35Implementations
- 4:36What is Raft
- 5:34HighLevel Overview
- 7:31Leader Election
- 9:34Split Vote
- 16:37Log Replication
- 19:23Network Partitions
- 19:55Vector Clocks
related talks
15mRaft: Distributed Consensus for Dummies
42mPractical Raft Consensus
44mStreams: past, present, and future
44mThe Road to Akka Cluster, and Beyond
01h 00mThe Secret Life of Distributed Systems
45m(...)Distributed Systems
57mScaling the CAP Theorem for Modern Distributed Systems
31mRPC Frameworks Overview