ETL Developer Interview Preparation Guide
An ETL developer interview can test whether you notice a semantic change that still passes a type check. Practice tracing source meaning through transformation, validation, and downstream consumption, with a clear plan for rejecting or resolving incompatible inputs.
Detect a change beyond the schema
Practice case: “A supplier keeps the same numeric delivery-duration field but changes its meaning from hours to minutes, and a load completes successfully with implausible results; how would you investigate?” Ask when the change occurred, which source versions exist, and whether the unit is documented anywhere outside the field type. This original exercise concerns semantic compatibility. A column remaining numeric does not mean the transformation still interprets the values correctly.
Trace a small labeled record from the supplier payload through staging, conversion, and the final output. Compare the original source value with its documented unit and expected business meaning. Avoid immediately dividing every historical value by a constant; the dataset may contain mixed versions or earlier corrections. Explain what evidence would distinguish a source-contract change from a transformation defect or a genuine shift in the underlying delivery process.
Make conversion assumptions inspectable
Document the input unit, target unit, accepted range, missing-value policy, and transformation version. Decide where a unit indicator or explicit contract version should be carried if the source can provide it. A conversion step should be testable with a small example whose expected interpretation is unambiguous. The objective is not merely to make values look plausible, but to preserve meaning in a way downstream consumers can inspect and maintain.
Separate checks that reveal an anomaly from checks that establish the correct interpretation. A range warning may detect unusual values but cannot prove their unit. Explain how you would retain the original payload and communicate uncertainty while investigating under the organization's actual retention rules. Airflow's repeatability guidance supports explicit input boundaries; it does not tell you whether a supplier value means hours, minutes, or an entirely different measure.
Handle a mixed-version delivery
Follow-up: “What would you do if one file contains records produced before and after the supplier's unit change without a version marker?” State that a global conversion would be unsafe without further evidence. Ask whether timestamps, source identifiers, or a corrected extract can distinguish the populations. Compare holding the load, isolating ambiguous records, and requesting a replacement. The right action depends on the downstream consequence and who can accept partial or delayed data.
Keep any inferred classification visibly uncertain rather than labeling a guessed unit as verified. If stakeholders choose a temporary estimate, record the rule and its limitations separately from confirmed data. Do not use the exercise to invent a universal anomaly threshold or imply that statistical separation reliably recovers missing semantics. A candidate can demonstrate sound judgment by identifying when the available evidence cannot support an automatic transformation.
Reprocess without duplicating downstream facts
Follow-up: “How would you correct previously loaded records after the supplier provides an authoritative mapping of affected rows?” Define the correction scope by source identity and transformation version. Ask how downstream tables reference those rows and which reports require a revision. Reprocessing is not simply rerunning the latest job if its inputs have changed or if writes append new rows alongside the incorrect facts already present in the destination.
Propose a controlled correction rehearsal using a bounded sample and compare before-and-after values, row counts, and identifiers. Check that unaffected rows remain unchanged and that repeated application does not duplicate the correction. Communicate which outputs were republished and which remain historical snapshots. Keep real database commands out of the hypothetical answer until the storage semantics, permissions, recovery plan, and accountable owner are established for the actual environment.
Present a maintainable transformation story
Practice explaining how the team detected the semantic break, established source meaning, chose a holding policy, corrected history, and improved the contract. Include who supplied authoritative information and who approved consumer-facing revisions. A transformation script alone is not the entire outcome. If your experience was limited to implementation, describe that contribution accurately rather than claiming ownership of supplier governance or every affected business report.
These original prompts are not employer question-bank material or a validated ETL answer key. Use the worksheet to rehearse source meaning and correction boundaries, then review the actual employer's platforms and responsibilities. Hypothetical values and supplier behavior remain fictional. Successful replay of a small example does not establish correctness of an entire historical dataset, and an anomaly detector does not replace an authoritative definition of the source field.
ETL Developer Interview Preparation Guide — decision rehearsal
| Case checkpoint | Hypothetical evidence or constraint | My reasoning and next check |
|---|---|---|
| Meaning | Numeric duration silently changes units | Trace source contract before converting |
| Contract | Schema check passes while semantics fail | Record units and transformation version |
| Ambiguity | One unmarked file mixes old and new units | Hold or isolate unsupported interpretations |
| Correction | Authoritative affected-row mapping arrives | Bound replay and preserve unaffected records |
| Maintenance | Supplier meaning crosses several owners | Explain ownership and revision communication |
Useful information is only the beginning.
Choose relevant opportunities, show your real strengths, and prepare for the roles that matter.
See how Princive helps →