How should you handle schema drift in a lakehouse pipeline?

• A. Ignore drift and continue with old schemas.
• B. Implement schema drift handling with schema evolution features, use late-bound schemas, implement tolerant reads, and employ data quality checks to catch and adapt to changes.
• C. Lock schemas to prevent changes.
• D. Convert all data to a single fixed schema upfront.

Answer: (B)

Handling schema drift in a lakehouse pipeline means designing for changing data structures so ingestion remains robust as sources evolve. The best approach uses schema evolution features to allow new columns or type changes to be accommodated without rewriting history, applies late-bound schemas so the schema is determined at read time rather than write time, and employs tolerant reads so queries can handle both older and newer data gracefully. Pairing this with data quality checks helps detect drift early and trigger adaptive changes to the pipeline, such as updating the schema or adjusting transformations.

This approach works better than ignoring drift, which can cause failures when data doesn’t match the old schema, or locking schemas, which blocks evolution and breaks new data. Converting everything to a fixed upfront schema is inflexible and often impractical as data sources continuously evolve.