Interview preparation guide
ETL Interview Questions With Pipeline Exercises
Practice ETL interview questions by explaining the model, working through a small problem and checking failure cases. These original exercises test reasoning rather than recall of a claimed employer question list.
Scope and assumptions
The example is an in-memory Python transformation with fictional rows. A production extract-transform-load pipeline also needs source contracts, persistence, access controls and recovery behavior.
Confirm the version, deployment model and permissions in the interview. The examples use fictional data and are deliberately small. A snippet demonstrates one idea; it is not a complete production system.
How do ETL and ELT differ?
The order of transformation and loading differs. Choose based on the platform, governance and workload. The label does not by itself establish data quality or a reliable recovery process.
What makes an incremental load safe?
Define the watermark or change signal, handle late and repeated events, and commit progress consistently with the intended output. A timestamp greater-than filter can miss late arrivals or duplicate work if the boundary is unclear.
How do you handle schema changes?
Classify compatible and incompatible changes, validate inputs and keep a visible failure or quarantine route. Silently dropping an unexpected field can hide a broken source contract.
What should be monitored?
Track freshness, counts, rejected rows, duplicates and reconciliation against source expectations. A process exit code alone cannot show that the right data arrived.
Worked exercise
Keep the latest event for each order ID using the event sequence number. The sample intentionally includes a repeated order and an input order that is not chronological.
rows = [
{"id": "A", "sequence": 2, "amount": 12},
{"id": "B", "sequence": 1, "amount": 8},
{"id": "A", "sequence": 1, "amount": 10},
]
latest = {}
for row in rows:
old = latest.get(row["id"])
if old is None or row["sequence"] > old["sequence"]:
latest[row["id"]] = row
print(sum(row["amount"] for row in latest.values()))Expected result and explanation
The output is 20: order A contributes its sequence-2 amount of 12, and B contributes 8. The code is independent of input ordering for these distinct sequence values. Equal sequences with conflicting content need an explicit policy rather than an accidental first-row-wins result.
Before changing the example, predict the result. Then vary one input or constraint and explain why the behavior should change. Keep the distinction between a measured result and an assumption clear.
Debugging and follow-up
If a rerun changes totals, inspect deduplication keys, source ordering and checkpoint commits. Test a repeated batch and a late event in a sandbox before claiming the load is idempotent.
- What evidence would distinguish your first explanation from an alternative?
- Which boundary, permission or failure case should be checked?
- How would you reproduce the issue without affecting live users?
Practice checklist
Explain one design decision aloud and support it with the worked example. Use the official documentation below to check version-specific behavior.
- I stated the relevant version and environment.
- I can explain the expected result and an edge case.
- I can separate correctness, performance and security concerns.
- I know which parts require a real runtime or service to verify.
Questions, answered
Are these questions used by a particular employer?
No. They are original practice questions. Employers can assess different tools, versions and levels of depth.
How should I verify the exercise?
Use an isolated local or sandbox environment with the stated prerequisites. Compare the expected result, then test an edge case. Do not run service commands against a shared or production system.
Sources and examples
ResumizeAI team. Examples are fictional teaching material, not customer results. Replace sample details with your own facts.
Your experience. A complete resume.
Build in Resumize with ATS-friendly templates, guided editing and a live preview. Start free, then upgrade when you need more.
Free account required. Free-plan PDF exports include a watermark.