What does a data engineer do?
A data engineer builds and runs the systems that move data from where it is created to where people can use it. That means pulling data from source systems, cleaning and reshaping it, loading it into a warehouse and keeping those jobs reliable. Interview questions test each step, along with SQL, data modeling and how you handle failures.
We could not find an O*NET OnLine profile under this title. One close O*NET OnLine match is Data Warehousing Specialists, whose tasks include "Develop data warehouse process models, including sourcing, loading, transformation, and extraction" and "Verify the structure, accuracy, or quality of warehouse data." Other listed tasks are mapping data between source systems and warehouses, and troubleshooting.
The BLS Occupational Outlook Handbook page for database administrators and architects does not mention data engineers. It says "Database architects design and build new databases for systems and applications," and gives their median wage as $139,500 in May 2025. It projects employment of database administrators and architects to grow 4 percent from 2025 to 2035.
Which duties turn into which interview questions?
Each duty listed for the work tends to produce a question. The table maps O*NET tasks for data warehousing specialists to the question each one usually leads to and what the interviewer is checking. These questions are drawn from the job's published duties, and they do not come from a survey of real interviews.
| Duty (from O*NET) | Question it tends to produce | What the interviewer is checking |
|---|---|---|
| Develop extraction, transformation and loading processes | Walk me through a pipeline you built. | Design choices and why you made them |
| Verify the structure, accuracy or quality of warehouse data | How do you know the data in your pipeline is correct? | Testing habits and data checks |
| Map data between source systems and warehouses | How would you model this data for reporting? | Data modeling basics |
| Design and implement warehouse database structures | When would you denormalize a table? | Tradeoffs between storage, speed and simplicity |
| Provide troubleshooting support | Tell me about a pipeline failure you fixed. | Calm debugging and prevention |
| Create documentation such as metadata and entity diagrams | How do you document a dataset for other teams? | Whether others can use your work |
Use the table to plan your stories and to guess which technical questions will come up. The posting usually names the tools, so match each one to a project. If the posting names a tool you haven't used, prepare to explain the closest one you know.
What technical questions do data engineers get?
Plan for a technical round. Expect SQL, questions about extract, transform and load processes, data modeling and how pipelines are scheduled and recovered. Interviewers care about your reasoning as much as the answer, so talk through your steps and say what you would check before you commit to a design.
1. What is the difference between ETL and ELT?
AWS's explainer on ETL describes extract, transform and load as "the process of combining data from multiple sources." It describes ELT as an extension that "reverses the order of operations," loading data into the target system before processing it. The interviewer is checking that you know both and can say when you would pick each.
2. What is the difference between an inner join and a left join?
The PostgreSQL documentation says a left outer join keeps every row of the left table at least once, and fills the right table's columns with nulls when there is no match. An inner join keeps only the rows that match. Expect a follow-up where you write a query, and say out loud how you would check for duplicate rows after a join.
3. How do you make a pipeline safe to rerun?
This tests a core habit. Apache Airflow's best practices page says that because Airflow can retry a failed task, "the tasks should produce the same outcome on every re-run." It warns that an INSERT during a rerun can create duplicate rows and suggests an UPSERT instead.
Example answer (made up, for illustration): "I design each task so running it twice gives the same result. For a daily load, I write to a partition for that date and overwrite it on each run, instead of appending rows. Where I have to merge into a table, I use an upsert keyed on the record's ID. I also make sure the task reads a fixed date range passed in by the scheduler, so a rerun next week processes the same day's data and not today's."
4. What is a DAG, and how do you schedule pipelines?
Airflow's documentation on Dags calls a Dag "a model that encapsulates everything needed to execute a workflow," including how many times to retry tasks. Explain how you set dependencies between tasks, how you handle retries and what you do when an upstream source arrives late.
5. How would you model sales data for reporting?
The interviewer is checking basic data modeling. Talk about a fact table of events, such as orders, with dimension tables for things like customers, products and dates. Explain how you choose the grain, meaning what one row represents, and why that choice matters for every report built on top.
6. How do you check data quality?
O*NET lists verifying the accuracy and quality of warehouse data as a core task. Describe the checks you run, such as row counts against the source, null checks on required fields, uniqueness on keys and range checks on numbers. Say what happens when a check fails, and who gets told.
What behavioral questions do data engineers get?
Behavioral questions ask about real situations from your past work. For data engineers they often cover failures, deadlines and working with analysts or software engineers. Use one real example per answer in STAR order: situation, task, action and result. The STAR method guide explains the format with more examples.
7. Tell me about a pipeline failure you fixed.
The interviewer is listening for how you found the cause and what you changed to stop it happening again. A good answer shows calm, a clear order of steps and a fix that lasts.
Example answer (made up, for illustration): "A morning dashboard showed sales at zero for one region. My task was to find out why before the leadership meeting. I traced it back and found the source system had renamed a column overnight, so our load job matched nothing and still reported success. I patched the mapping, reran the day's load and told the analysts the numbers were fixed. Afterward I added a check that fails the job if a key column is missing or a region's row count drops sharply."
8. Tell me about a time a stakeholder needed data faster than you could deliver it.
They want to see how you handle pressure and tradeoffs. Explain what you delivered first, what you left for later and how you made the limits clear. The page on competing priorities covers this kind of answer.
9. Tell me about a time you improved a slow pipeline.
The interviewer is checking whether you measure before you change things. Say how you found the slow step, what you changed and how much faster it ran. Only give a number you can explain.
10. Tell me about working with an analyst or data scientist who used your data.
Data engineers build for other people. The interviewer wants to see that you ask what users need and document your work so they can use it without asking you.
What situational questions test a data engineer?
Situational questions describe a problem and ask what you would do. They test judgment under real constraints, such as a broken source, a missed deadline or a cost limit. Explain what you would check first, who you would tell and how you would keep a short fix from becoming a permanent problem.
11. A source system starts sending duplicate records. What do you do?
The interviewer wants a short-term fix and a lasting one.
Example answer (made up, for illustration): "First I'd stop the bad data spreading, by pausing downstream jobs or flagging the affected tables for users. Then I'd confirm the pattern, such as which records repeat and since when. In the pipeline I'd remove duplicates on a reliable key so reports are correct again. I'd also contact the team that owns the source system, since the real fix belongs there, and add a uniqueness check so we hear about it next time."
12. An analyst says yesterday's numbers look wrong. What do you do?
They are checking your debugging order. Ask which numbers and compared with what, then check whether the pipeline ran and whether row counts match the source. Tell the analyst what you found, even if the data turns out to be correct.
13. You're asked to build a new pipeline in a week with an unclear spec. What do you do?
The interviewer wants to hear you ask the right questions. Find out who will use the data, how fresh it must be and what one row should mean. Build the simplest version that meets that, and agree on what comes later.
What follow-up questions tend to come next?
Technical interviewers push on your answers to see how deep your knowledge goes. Expect questions about scale, failure and the choices you made. Prepare for these on each project you plan to discuss, so your answers stay specific and you don't overstate your part.
- "How much data did that pipeline handle each day?"
- "What happens if this job fails halfway through?"
- "Why did you choose that tool over another one?"
- "How would this design change if the data grew ten times?"
- "What part did you build, and what did others build?"
How should you prepare for a data engineer interview?
Practice SQL by writing queries out loud, since many interviews ask you to explain while you type. Pick two pipelines you built and rehearse how they work, how they fail and what you would change. Read the posting and match each tool and duty to the table above.
The free interview questions tool lets you paste the job and pick the round, then shows likely questions, why each is asked and what the interviewer listens for. Related guides include data analyst interview questions, data scientist interview questions and software engineer interview questions.
To rehearse behavioral answers out loud, the Prept Mock room puts you on a clock and scores each answer out of 10, checked against your resume. It is included in the free plan. For more story practice, read the list of behavioral interview questions.