Roadmaps / Distributed Systems / Interview topics
Reference
Interview topics
Use these for backend, infrastructure and distributed systems roles. Each follow-up is the question a good interviewer asks next. The Staff line is what lifts the answer.
Phase 1
| Question | Follow-up | A Staff answer adds |
|---|---|---|
| How do you detect a failed node? | What if it is only slow? | Timeouts as a trade-off, a measured false-suspicion rate and a fix |
| Can you rely on wall clocks? | What do you use instead? | Skew and drift, logical clocks and where a bounded-skew assumption is fine |
| What does exactly-once mean? | How do you get it? | At-least-once delivery plus idempotent handlers, and where it breaks |
| What is eventual consistency? | How do you merge conflicts? | A merge rule, CRDTs where they fit and what the reader may see |
System design prompt (20 minutes): a membership service for 1,000 nodes. Outline: requirements and numbers (3 min), failure detection (5), dissemination (5), false positives (4), observability (3).
Phase 2
| Question | Follow-up | A Staff answer adds |
|---|---|---|
| How does Raft elect a leader? | What prevents two leaders? | Terms, votes, the up-to-date rule and the randomised timeout trade-off |
| When is a write committed? | What happens to uncommitted entries? | The majority rule, log repair and a seeded bug the checker caught |
| What is linearizability? | How do you test it? | A history checker and a failing history you produced |
| What happens in a partition? | Which side stays available? | The CAP trade-off as a concrete behaviour and a measured result |
System design prompt (20 minutes): a lock service. Outline: guarantee and failures (3 min), replication (5), leases and fencing tokens (5), failure cases (4), testing (3).
Phase 3
| Question | Follow-up | A Staff answer adds |
|---|---|---|
| How do you serve linearizable reads? | What does a lease assume? | Log reads, read index and leases with the clock bound written down |
| How do snapshots work? | What if a follower is far behind? | Compaction, installing a snapshot and the cost |
| LSM tree or B-tree? | How do you decide? | Write amplification, read cost and a measurement on your workload |
| How do you shard a store? | What about cross-shard operations? | Routing, rebalancing and the cost of two-phase commit |
System design prompt (20 minutes): a replicated key-value store at 50,000 writes a second. Outline: numbers (3 min), sharding and replication (5), storage engine (4), read path (4), failure and rebalancing (4).
Phase 4
| Question | Follow-up | A Staff answer adds |
|---|---|---|
| How do you test a distributed system? | What does a passing test prove? | Fault injection, a checker, what the test cannot see and a seeded bug |
| How do you upgrade a cluster? | What if a bug appears halfway? | A rolling plan, a rollback rule and a drill that you have run |
| When do you use consensus? | When do you not? | A decision guide with latency and availability costs |
| How do you debug a partial failure? | What do you look at first? | Per-node views, timelines and a runbook built from earlier incidents |
System design prompt (20 minutes): a plan to test a replicated database before launch. Outline: guarantees to check (3 min), faults to inject (5), checkers and models (5), operating drills (4), release criteria (3).