A team computes updated customer aggregates every night and wants to refresh only the corresponding rows in an existing Unity Catalog feature table without overwriting untouched entities. Which Databricks Feature Engineering pattern is appropriate?
Answer & explanation
Correct answer: C
WHY C: The documented pattern for partial updates is FeatureEngineeringClient.write_table with mode='merge', which updates matching primary keys and leaves other rows unchanged. WHY NOT A: Full rewrites are not required for partial updates. WHY NOT B: Local export-import is unnecessary and brittle. WHY NOT D: create_training_set is for model training data assembly, not feature table writes. WHY NOT E: MLflow logging does not write feature table data.