Data Engineering · for Experienced

ETL Interview Questions for Experienced (2026 Prep Guide)

10 min read6 easy · 8 medium · 7 hardLast updated: 22 Apr 2026

Expect rigour on schema evolution, data quality, and warehousing patterns alongside classic algorithms. At the mid-career bar, Explaining query plans and join strategies aloud separates strong candidates.

Part of the hub:ETL Interview Guide →

Start a personalised drill ETL Questions Hub

Data-engineering interviews test pipeline reasoning, SQL depth, and system-design intuition in equal measure. In the for experienced track specifically, interviewers weight ETL as a proxy for both depth and judgement — the combination that separates an offer from a "close but not this cycle" decision. Ownership of data quality, SLAs, and observability earns senior-level signal.

The fastest way to internalise ETL is deliberate practice against progressively harder scenarios. Begin with the fundamentals so you can discuss definitions, invariants, and trade-offs without fumbling vocabulary. Then move into scenario drills drawn from cases like E-commerce order funnels with late-arriving events. The goal isn't recall — it's the habit of restating a problem, surfacing assumptions, and narrating your decision process out loud.

Interviewers also listen for boundary awareness. When ETL appears in a panel, strong candidates acknowledge where their approach breaks: cost envelope, latency under load, consistency trade-offs, or organisational constraints. Interviewers weight partitioning, idempotency, and schema evolution heavily. Your answers should explicitly name the two or three dimensions on which the solution could flip, and which one you'd optimise given the user's priorities.

Finally, calibrate your preparation against actual panel dynamics. Rehearse each ETL answer out loud, time-box it to three minutes, and iterate based on recorded playback. Pair written study with two to three full mock interviews before the target loop. Clear reasoning about batch-vs-stream trade-offs is a strong differentiator. Showing up with clear structure, measurable examples, and one honest boundary beats a longer monologue on any rubric that actually exists.

Preparation roadmap

Step 1
Days 1–2 · Fundamentals
Re-read the ETL basics end to end. If you can't explain it in 90 seconds to a smart non-expert, you're not ready for the panel follow-ups.
Step 2
Days 3–4 · Scenario drills
Run six timed drills anchored in real cases — e.g. Media clickstream rollups feeding ML training sets. Verbalise your thinking; recorded audio beats silent practice.
Step 3
Days 5–6 · Panel simulation
Two full-loop mock interviews with a peer or adaptive coach. Score yourself against a rubric: restatement, trade-offs, execution, communication.
Step 4
Day 7 · Weakness blitz
Target your worst rubric cell from the mocks. Do three focused 20-minute drills specifically on that gap — not new content.
Step 5
Day 8+ · Cadence
Hold a 30-minute daily drill plus one weekly mock until the target interview. Consistency compounds faster than marathon weekends.

Top interview questions

Q1.Walk me through a common pitfall when using ETL under load.
medium
Explaining query plans and join strategies aloud separates strong candidates. With ETL, the classic pitfall is optimising the common path while ignoring tail behaviour.
Example
Real pipeline: Kafka → bronze (Delta) → silver (schema-validated) → gold (aggregated). Idempotency at each layer.
Common mistakes
- Ignoring skew — one hot key balloons executors while the rest idle.
- Benchmarking on cold cache — production hits warm cache and the numbers invert.
Follow-up: What breaks first if the job runs on half the cluster?
Q2.How would you design a test plan for ETL?
medium
Write the happy-path tests first; then add boundary, concurrency, and rollback tests around ETL so regressions are caught cheaply.
Example
dbt example: `{{ incremental() }}` with `unique_key=[user_id, event_id]` reliably dedupes replayed CDC events.
Common mistakes
- Benchmarking on cold cache — production hits warm cache and the numbers invert.
- Ignoring skew — one hot key balloons executors while the rest idle.
Follow-up: How do you detect and recover from duplicate writes in production?
Q3.Design a scalable system that centres on ETL. What are the top 3 trade-offs?
hard
At scale, ETL forces choices between strong consistency, cost envelope, and blast-radius containment. I'd surface all three up front.
Example
Imagine a 2 TB Spark job: setting `spark.sql.shuffle.partitions=400` and broadcasting a 10 MB dim table cut runtime from 45m to 6m.
Common mistakes
- Ignoring skew — one hot key balloons executors while the rest idle.
- Benchmarking on cold cache — production hits warm cache and the numbers invert.
Follow-up: Walk me through the observability you would add before shipping this.
Q4.Describe a real-world failure mode of ETL and how you'd detect it before customers notice.
hard
The classic failure is silent skew on ETL. Interviewers weight partitioning, idempotency, and schema evolution heavily. Detect it with a small canary that double-writes and compares counts.
Example
Real pipeline: Kafka → bronze (Delta) → silver (schema-validated) → gold (aggregated). Idempotency at each layer.
Common mistakes
- Benchmarking on cold cache — production hits warm cache and the numbers invert.
- Ignoring skew — one hot key balloons executors while the rest idle.
Follow-up: Where does your solution fail if data arrives out of order?
Q5.How do you prioritise improvements to ETL when time and budget are limited?
medium
Map work to an impact × effort grid; pick the top-right quadrant first and schedule the rest visibly so ETL stakeholders see the plan.
Example
dbt example: `{{ incremental() }}` with `unique_key=[user_id, event_id]` reliably dedupes replayed CDC events.
Common mistakes
- Ignoring skew — one hot key balloons executors while the rest idle.
- Benchmarking on cold cache — production hits warm cache and the numbers invert.
Follow-up: If latency had to drop 10x, what would you change first?
Q6.What metrics would you track to know ETL is working well?
medium
Define input quality, throughput, and error-rate metrics up front — post-hoc metric design on ETL always misses the real regressions.
Example
Imagine a 2 TB Spark job: setting `spark.sql.shuffle.partitions=400` and broadcasting a 10 MB dim table cut runtime from 45m to 6m.
Common mistakes
- Benchmarking on cold cache — production hits warm cache and the numbers invert.
- Ignoring skew — one hot key balloons executors while the rest idle.
Follow-up: How would the answer change if the table was 100x larger?
Q7.How would you explain a trade-off in ETL to a skeptical senior stakeholder?
hard
Lead with the outcome change, then show the trade-off as a small, concrete number. Clear reasoning about batch-vs-stream trade-offs is a strong differentiator.
Example
Real pipeline: Kafka → bronze (Delta) → silver (schema-validated) → gold (aggregated). Idempotency at each layer.
Common mistakes
- Ignoring skew — one hot key balloons executors while the rest idle.
- Benchmarking on cold cache — production hits warm cache and the numbers invert.
Follow-up: What breaks first if the job runs on half the cluster?
Q8.What's the smallest proof-of-concept that demonstrates ETL clearly?
easy
Prefer a runnable Jupyter / REPL snippet with inputs and outputs over prose; interviewers can re-run it and probe immediately.
Example
dbt example: `{{ incremental() }}` with `unique_key=[user_id, event_id]` reliably dedupes replayed CDC events.
Common mistakes
- Benchmarking on cold cache — production hits warm cache and the numbers invert.
- Ignoring skew — one hot key balloons executors while the rest idle.
Follow-up: How do you detect and recover from duplicate writes in production?
Q9.How would you debug a slow ETL implementation?
medium
Always bisect against a known-good baseline; that tells you whether ETL regressed or the environment did.
Example
Imagine a 2 TB Spark job: setting `spark.sql.shuffle.partitions=400` and broadcasting a 10 MB dim table cut runtime from 45m to 6m.
Common mistakes
- Ignoring skew — one hot key balloons executors while the rest idle.
- Benchmarking on cold cache — production hits warm cache and the numbers invert.
Follow-up: Walk me through the observability you would add before shipping this.
Q10.Walk me through a scenario where ETL was the wrong tool for the job.
hard
Small data with hard latency bounds are a classic mismatch — ETL shines where throughput dominates, not cold-start speed.
Example
Real pipeline: Kafka → bronze (Delta) → silver (schema-validated) → gold (aggregated). Idempotency at each layer.
Common mistakes
- Benchmarking on cold cache — production hits warm cache and the numbers invert.
- Ignoring skew — one hot key balloons executors while the rest idle.
Follow-up: Where does your solution fail if data arrives out of order?
Q11.How do you document ETL so a new teammate can ramp up quickly?
medium
Capture the decision log, not just the current state — the "why not" around ETL is what a newcomer actually needs.
Example
dbt example: `{{ incremental() }}` with `unique_key=[user_id, event_id]` reliably dedupes replayed CDC events.
Common mistakes
- Ignoring skew — one hot key balloons executors while the rest idle.
- Benchmarking on cold cache — production hits warm cache and the numbers invert.
Follow-up: If latency had to drop 10x, what would you change first?
Q12.What's one question you'd ask the interviewer about ETL?
easy
Ask what they'd change if they were rebuilding ETL from scratch — it almost always surfaces the team's real pain points.
Example
Imagine a 2 TB Spark job: setting `spark.sql.shuffle.partitions=400` and broadcasting a 10 MB dim table cut runtime from 45m to 6m.
Common mistakes
- Benchmarking on cold cache — production hits warm cache and the numbers invert.
- Ignoring skew — one hot key balloons executors while the rest idle.
Follow-up: How would the answer change if the table was 100x larger?
Q13.Describe an end-to-end example that uses ETL.
medium
Consider a real-world example: E-commerce order funnels with late-arriving events. That scenario exercises ETL end-to-end under realistic load.
Example
Real pipeline: Kafka → bronze (Delta) → silver (schema-validated) → gold (aggregated). Idempotency at each layer.
Common mistakes
- Ignoring skew — one hot key balloons executors while the rest idle.
- Benchmarking on cold cache — production hits warm cache and the numbers invert.
Follow-up: What breaks first if the job runs on half the cluster?
Q14.What are the top 3 interviewer follow-ups after a strong ETL answer?
hard
Senior panels probe on blast radius, cost envelope, and operational load — rehearse those three before the loop.
Example
dbt example: `{{ incremental() }}` with `unique_key=[user_id, event_id]` reliably dedupes replayed CDC events.
Common mistakes
- Benchmarking on cold cache — production hits warm cache and the numbers invert.
- Ignoring skew — one hot key balloons executors while the rest idle.
Follow-up: How do you detect and recover from duplicate writes in production?
Q15.How would you onboard a junior engineer to work on ETL?
medium
Give them a reading list, a 30-day scoped project, and a mentor check-in cadence. The scope is the lever for ETL.
Example
Imagine a 2 TB Spark job: setting `spark.sql.shuffle.partitions=400` and broadcasting a 10 MB dim table cut runtime from 45m to 6m.
Common mistakes
- Ignoring skew — one hot key balloons executors while the rest idle.
- Benchmarking on cold cache — production hits warm cache and the numbers invert.
Follow-up: Walk me through the observability you would add before shipping this.
Q16.What's a non-obvious trade-off that only shows up in production with ETL?
hard
Tail latency and cold-start behaviour: both invisible in staging, both punishing when a real workload hits ETL.
Example
Real pipeline: Kafka → bronze (Delta) → silver (schema-validated) → gold (aggregated). Idempotency at each layer.
Common mistakes
- Benchmarking on cold cache — production hits warm cache and the numbers invert.
- Ignoring skew — one hot key balloons executors while the rest idle.
Follow-up: Where does your solution fail if data arrives out of order?
Q17.How would you split preparation time between theory and practice for ETL?
easy
Front-load theory, back-load mocks. The last 5 days before an interview are for simulated loops, not new content.
Example
dbt example: `{{ incremental() }}` with `unique_key=[user_id, event_id]` reliably dedupes replayed CDC events.
Common mistakes
- Ignoring skew — one hot key balloons executors while the rest idle.
- Benchmarking on cold cache — production hits warm cache and the numbers invert.
Follow-up: If latency had to drop 10x, what would you change first?
Q18.What resources accelerate ETL prep in the last 48 hours before an interview?
easy
Do 2 timed drills with a peer reviewer, then sleep. The marginal return on content in hour 47 is negative.
Example
Imagine a 2 TB Spark job: setting `spark.sql.shuffle.partitions=400` and broadcasting a 10 MB dim table cut runtime from 45m to 6m.
Common mistakes
- Benchmarking on cold cache — production hits warm cache and the numbers invert.
- Ignoring skew — one hot key balloons executors while the rest idle.
Follow-up: How would the answer change if the table was 100x larger?
Q19.What's the difference between junior and senior expectations on ETL?
hard
Junior: execute correctly under supervision. Senior: define the problem, choose the tool, own the outcome for ETL.
Example
Real pipeline: Kafka → bronze (Delta) → silver (schema-validated) → gold (aggregated). Idempotency at each layer.
Common mistakes
- Ignoring skew — one hot key balloons executors while the rest idle.
- Benchmarking on cold cache — production hits warm cache and the numbers invert.
Follow-up: What breaks first if the job runs on half the cluster?
Q20.What is ETL and why is it relevant to this interview round?
easy
Panels use ETL as a fast litmus test — it's hard to fake fluency, so being concise and precise pays off. Clear reasoning about batch-vs-stream trade-offs is a strong differentiator.
Example
dbt example: `{{ incremental() }}` with `unique_key=[user_id, event_id]` reliably dedupes replayed CDC events.
Common mistakes
- Benchmarking on cold cache — production hits warm cache and the numbers invert.
- Ignoring skew — one hot key balloons executors while the rest idle.
Follow-up: How do you detect and recover from duplicate writes in production?
Q21.How would you explain ETL to a non-technical stakeholder?
easy
Lead with "what changes for the user / business", then a 2-sentence mechanism, then one trade-off the stakeholder cares about.
Example
Imagine a 2 TB Spark job: setting `spark.sql.shuffle.partitions=400` and broadcasting a 10 MB dim table cut runtime from 45m to 6m.
Common mistakes
- Ignoring skew — one hot key balloons executors while the rest idle.
- Benchmarking on cold cache — production hits warm cache and the numbers invert.
Follow-up: Walk me through the observability you would add before shipping this.

Interactive

Practice it live

Practising out loud beats passive reading. Pick the path that matches where you are in the loop.

Try this question liveRun the question through the adaptive AI coach — it grades structure, trade-offs, and communication.Start a mock interviewFull-loop mock with rubric scoring and a recorded replay — ready in under a minute.Get a personalised roadmapTell the coach your target, and get a day-by-day prep plan tuned to your gaps.

Explore by domain

Related roles

Related skills

Practice with an adaptive AI coach

Personalised plan, live mock rounds, and outcome tracking — free to start.

Start a personalised drill Run a mock round

Difficulty mix

This guide is weighted 6 easy · 8 medium · 7 hard — use it as a structured study sheet.

Crisp framing for ETL questions interviewers actually ask
A difficulty-balanced set: 6 easy · 8 medium · 7 hard
Real-world scenarios like Fintech transaction streams with exactly-once semantics — grounded in day-one operational reality