Data Engineer
Interview Questions.
15 practice questions across 14 skills — each with what the interviewer is really listening for. Prepare with intent, not guesswork.
Free · No signup · Grouped by skill
How to use this page
Don't memorise answers. For each question, read what the interviewer is looking for, then practise answering out loud in your own words with a real example. The notes describe strong answers and common red flags — they are for your prep, not a script.
ETL
2 questionsWalk me through how you would design a pipeline to load daily sales data into a warehouse.
IntermediateWhat the interviewer is looking for: Strong answers cover extraction, validation, transformation, idempotent loads, scheduling, and handling late or duplicate data. A pipeline with no idempotency or data-quality checks is a red flag.
What is the difference between ETL and ELT, and when does ELT make more sense?
IntermediateWhat the interviewer is looking for: Look for transforming after load in a powerful warehouse for flexibility and scale, versus transforming before load. Not knowing the distinction or its trade-offs is a red flag.
Apache Spark
1 questionHow does Apache Spark achieve parallelism, and what causes a slow Spark job?
AdvancedWhat the interviewer is looking for: Strong candidates explain partitioning and lazy evaluation, and diagnose skew, shuffles, and spills. Blaming 'not enough memory' without understanding shuffles or skew is a red flag.
SQL
1 questionWrite a query to deduplicate records keeping the latest version per key. Explain your approach.
IntermediateWhat the interviewer is looking for: Look for a window function with ROW_NUMBER partitioned by key ordered by timestamp, and handling ties. A GROUP BY that loses non-aggregated columns is a red flag.
Python
1 questionHow would you use Python to make a data pipeline resilient to bad input rows?
IntermediateWhat the interviewer is looking for: Strong answers validate schemas, quarantine bad records, log and alert, and avoid failing the whole batch on one bad row. Letting one malformed row crash the pipeline is a red flag.
Snowflake
1 questionHow do you model tables in Snowflake for analytics that stay fast as data grows?
AdvancedWhat the interviewer is looking for: Look for dimensional modelling, sensible clustering, and awareness of compute/storage separation and cost. Dumping raw wide tables with no modelling is a red flag.
Data Modeling
1 questionHow do you decide on a data model — star schema versus a wide denormalised table?
AdvancedWhat the interviewer is looking for: Strong candidates weigh query patterns, maintainability, and storage, and know when denormalisation helps performance. Applying one pattern dogmatically regardless of use case is a red flag.
Data Quality
1 questionHow do you ensure and monitor data quality across your pipelines?
IntermediateWhat the interviewer is looking for: Look for automated checks (freshness, volume, nulls, uniqueness), alerting, and clear ownership. Finding out about data issues only when a stakeholder complains is a red flag.
Apache Airflow
1 questionHow would you orchestrate dependent jobs with Airflow so a failure does not corrupt downstream data?
AdvancedWhat the interviewer is looking for: Strong answers use DAG dependencies, retries, idempotent tasks, and stopping downstream on upstream failure. Fire-and-forget schedules with no dependency handling are a red flag.
Apache Kafka
1 questionHow do you handle a streaming source like Kafka differently from a batch load?
AdvancedWhat the interviewer is looking for: Look for understanding of partitions, consumer offsets, exactly/at-least-once semantics, and out-of-order events. Treating a stream exactly like a batch file is a red flag.
AWS
1 questionHow do you keep cloud data costs under control on AWS as pipelines scale?
IntermediateWhat the interviewer is looking for: Strong candidates partition and compress data, avoid scanning everything, right-size compute, and monitor spend. Full-table scans and always-on large clusters are red flags.
Problem-Solving
1 questionA downstream dashboard shows wrong numbers. How do you trace it back through the pipeline?
AdvancedWhat the interviewer is looking for: Look for working backward layer by layer, checking transformations and source data, and using lineage. Guessing at the cause without tracing lineage is a red flag.
Collaboration
1 questionHow do you collaborate with analysts and scientists who consume your data?
BasicWhat the interviewer is looking for: Strong answers treat consumers as customers, document schemas and SLAs, and communicate changes. Breaking downstream consumers with silent schema changes is a red flag.
ELT
1 questionHow do you decide between a batch pipeline and a real-time streaming approach for a use case?
AdvancedWhat the interviewer is looking for: Strong candidates weigh freshness needs against complexity and cost, defaulting to batch unless latency truly requires streaming. Building streaming because it is trendy is a red flag.
Docker
1 questionHow do you use Docker and Git to make pipelines reproducible across environments?
IntermediateWhat the interviewer is looking for: Look for containerised, version-controlled, pinned environments so runs are consistent. Environment drift and "works on my laptop" pipelines are red flags.