← All courses
Distributed Systems · Advanced
Raft and Paxos Explained: Consensus, ZooKeeper and etcd
How a group of machines agrees on a single value when any of them can crash and the network can lose or delay messages: the FLP impossibility result and the partial-synchrony assumptions real systems use to work around it, quorums, Paxos and Multi-Paxos, Raft leader election and log replication, and the coordination services built on top of them including ZooKeeper, etcd, and Consul. The final phase covers what actually goes wrong in production - split brain, quorum loss, botched membership changes, and clock-skew assumptions.
Helpful first: Distributed Systems Explained: CAP, Clocks and Replication.
Step 5 of 6 in the System design and architecture path. Up next: System Design Interview Prep: Framework and Questions.