Roadmaps / Data / Interview topics
Reference
Interview topics
Use these for data engineering and analytics engineering 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 |
|---|---|---|
| Why use Parquet? | When is it a bad fit? | Column pruning, encodings and a measured scan size, and point lookups |
| How do you partition a table? | What goes wrong? | The query filters, the file sizes and a compaction plan |
| What does Iceberg or Delta add? | How do they differ? | Atomic commits, snapshots, evolution and a concrete difference you met |
| How do you handle schema changes? | What breaks a reader? | Evolution rules, a contract test and a rollout plan |
System design prompt (20 minutes): a clickstream table for 500 million events a day. Outline: numbers (3 min), format and partitioning (5), compaction and retention (4), query patterns and cost (4), schema changes (4).
Phase 2
| Question | Follow-up | A Staff answer adds |
|---|---|---|
| How do you make a job idempotent? | What about a backfill? | Partition overwrite or merge, a hash check and a copy-then-swap backfill |
| How do you know the numbers are right? | What if there is no source of truth? | Reconciliation, tests and an owner who signs off |
| Event time or processing time? | What about late data? | Which one the report needs, a lateness rule and a late-data table |
| How do you model changing customer data? | What is a slowly changing dimension? | A history table, an as-of query and the cost |
System design prompt (20 minutes): a daily revenue pipeline the finance team trusts. Outline: requirements (3 min), sources and reconciliation (5), job design (5), tests and quarantine (4), alerting and ownership (3).
Phase 3
| Question | Follow-up | A Staff answer adds |
|---|---|---|
| How does change capture work? | What about deletes and schema changes? | Log-based capture, tombstones and a plan for each |
| How do you get exactly-once results? | What breaks it? | At-least-once delivery plus an idempotent sink and a restart test |
| What is a watermark? | How late is too late? | The lateness rule, the late-data table and the cost of waiting |
| How do you size a Kafka topic? | What limits throughput? | Partitions, keys and ordering, with a measurement |
System design prompt (20 minutes): sync a Postgres database to a lake with a 10-second lag. Outline: requirements (3 min), capture and transport (5), sink and idempotency (5), failure drills (4), monitoring (3).
Phase 4
| Question | Follow-up | A Staff answer adds |
|---|---|---|
| What is a data contract? | How do you enforce it? | A test in the producer's pipeline and an owner who is told |
| How do you handle a deletion request? | What about backups and snapshots? | A runbook, verification queries and a stated time limit |
| How do you control cost? | How do you find the problem? | Bytes scanned by query, a fix with a measurement and a budget alert |
| How do you choose a data platform? | When is a warehouse enough? | The workload, the team and a measured comparison |
System design prompt (20 minutes): a data platform for ten teams. Outline: goals (3 min), storage and compute (5), ownership and contracts (5), governance and deletion (4), cost and support (3).