How do you handle data versioning when schemas change?

• A. Delete older versions to free space.
• B. Maintain multiple versions, use data contracts, and provide backward compatibility or migration scripts.
• C. Never change schemas.
• D. Change schema without any compatibility plan.

Answer: (B)

When schemas evolve, you handle data versioning by treating schema changes as versioned contracts and planning for compatibility and migration. Maintain multiple versions of the data, formalize each version with data contracts that specify structure, types, and constraints, and provide backward-compatible changes or explicit migration scripts to move data from older versions to newer ones. This approach preserves historical data for reproducibility, lets downstream processes know exactly what to expect for each version, and ensures pipelines can transition smoothly.

Deleting older versions loses valuable history and auditability. Never changing schemas is impractical as needs evolve. Changing a schema without a compatibility plan will break existing consumers and pipelines.