# Common Interview and Exam Questions — Distributed Systems & Consensus (Raft / Paxos)

Source: https://www.skillbyai.com/en/distributed-consensus/dc-interview-questions

> Answer the most frequently asked consensus questions with crisp, structured explanations.

## Questions you should be able to answer aloud

1. *Why does a consensus cluster need a majority, and why use odd sizes?* Majorities intersect; an even size adds cost without tolerating more failures. 2. *Walk through a Raft leader election and a split vote.* Timeout, term increment, self-vote, RequestVote, majority wins, randomised timeouts break ties. 3. *How does Raft guarantee a new leader has all committed entries?* Election restriction plus majority overlap. 4. *Why can a leader not commit an old-term entry by counting replicas?* The Figure 8 scenario; commit a current-term entry instead. 5. *Paxos vs Raft?* Same quorum logic; Raft fixes leader election, log structure and membership details. 6. *What is FLP and how do real systems live with it?* Safety always, liveness under partial synchrony. 7. *2PC vs consensus?* All participants vs a majority; 2PC blocks on coordinator failure. 8. *How would you build a distributed lock?* Lease in etcd or ZooKeeper plus fencing tokens checked by the resource.

## Structure beats memorisation

For any 'design with consensus' question, cover four points in order: what is replicated (the log), who decides order (the leader), what makes it safe (majority overlap plus the voting rule), and what happens on failure (election, catch-up, fencing). Draw the cluster before you talk.

## Explaining Raft like cricket captaincy

The team (cluster) has one captain (leader) per match (term). If the captain is unreachable for too long, a player volunteers (candidate) and needs most of the team's support; players back only someone who has followed the match at least as closely as they have. Decisions recorded in the scorebook by a majority are never erased.

**Quiz:** An interviewer asks why a 4-node Raft cluster is rarely used. What is the best answer?

- [ ] Raft does not support even-sized clusters
- [ ] Four nodes cannot elect a leader
- [x] It tolerates the same single failure as 3 nodes while needing 3 acknowledgements per commit
- [ ] Four nodes always cause split brain

*Answer:* It tolerates the same single failure as 3 nodes while needing 3 acknowledgements per commit. A majority of 4 is 3, so only one failure is tolerated, exactly like 3 nodes, while every write waits for more acknowledgements.