Consensus & Replication

Replicated logs · Raft · Quorums

1:00

How distributed systems keep multiple server copies in sync using quorum-based consensus — covering the replicated state machine model, Raft's leader election and log replication, quorum math, and fault tolerance.

By the end, you can

Explain why the replicated state machine model reduces consensus to log ordering.
State the Two Generals result and explain why perfect agreement over an unreliable channel is impossible — only arbitrarily probable, never certain.
State the FLP result and explain why a slow node looks identical to a dead one.
Describe the three Raft roles (follower, candidate, leader) and how terms work.
Compute the majority quorum size for any N using ⌊N/2⌋ + 1.
Trace the steps of a Raft leader election from timeout to majority win.
Explain why a committed entry survives a leader crash.
Calculate the minimum cluster size needed to tolerate f crash failures (2f + 1) and the larger size needed to tolerate f Byzantine (malicious) failures (3f + 1).
Describe how quorum overlap prevents split-brain during a network partition.
Distinguish Paxos from Raft and identify when Byzantine fault tolerance is needed.

Up next in Networking, Distributed Systems & Databases