GhostpilotHome · Interview questions · Data Engineer
Data Engineer interview questions
Data engineer interviews centre on strong SQL and building pipelines people can trust. Expect questions about late data, failed runs and what happens when a source changes without warning.
What interviewers look for
- SQL. Window functions, joins and query performance.
- Pipeline design. Idempotent, recoverable, observable jobs.
- Modelling. Tables that analysts can use correctly.
How the interview usually runs
- Recruiter call.
- SQL and coding screen.
- Pipeline or data modelling design.
- Behavioural round with stakeholders in mind.
Stages vary by company, so ask the recruiter what to expect before you start preparing.
10 Data Engineer interview questions, and how to answer them
1. Find the second-highest salary per department.
What they want to see: Window functions.
How to answer: Use DENSE_RANK with PARTITION BY and explain how ties behave.
2. Batch or streaming?
What they want to see: Choosing by need.
How to answer: Ask how fresh the data must be, then weigh the cost and complexity.
3. How do you make a pipeline idempotent?
What they want to see: Safe reruns.
How to answer: Use overwrite by partition or merge on keys, so a rerun gives the same result.
4. How do you handle late-arriving data?
What they want to see: Real-world data.
How to answer: Use watermarks and reprocessing windows, and be clear on what reports show.
5. Star schema or one wide table?
What they want to see: Modelling trade-offs.
How to answer: Compare ease of use, storage and flexibility for the tools in use.
6. A Spark job is slow. What do you check?
What they want to see: Performance debugging.
How to answer: Look for skew, shuffles, partition size and broadcast joins in the Spark UI.
7. How do you ensure data quality?
What they want to see: Trustworthy data.
How to answer: Add tests on keys, nulls and volumes, plus freshness alerts and ownership.
8. How do you handle schema changes?
What they want to see: Resilience.
How to answer: Use contracts, schema registries and additive changes, and alert on drift.
9. ETL or ELT?
What they want to see: Modern stack.
How to answer: ELT loads raw data first, then transforms in the warehouse, which suits tools like dbt.
10. Tell me about a data incident.
What they want to see: Ownership.
How to answer: Say how you found it, told users, fixed it and prevented it.
A sample answer
Tell me about a data incident.
Finance noticed revenue looked 8% low one morning. I traced it to a source team renaming a column, which turned a join into nulls. I told users within the hour, backfilled the partition and added a schema check and a volume alert. We also agreed a contract with the source team, so changes now reach us first.
A made-up example to show the structure. Use your own experience and real numbers.Common mistakes
- Weak SQL for a role built on it.
- Pipelines that can't be rerun safely.
- Ignoring the people who use the data.
- Talking about tools with no scale or numbers.
Practise this out loud
Run a mock interview for a Data Engineer role. Questions change every time, you answer by voice or typing, and each answer gets a score with feedback.
Frequently asked questions
How much SQL do data engineers need for interviews?
A lot. Be comfortable with joins, window functions, CTEs and reading a query plan.
Do I need Spark?
Check the job description. Many teams work in warehouses with dbt, but Spark is still common at larger data volumes.