Amazon SageMaker Feature Store introduces UpdateRecord for feature-level writes
aws ·
Amazon SageMaker Feature Store has introduced the UpdateRecord API, enabling feature-level writes that allow users to update specific attributes within a record without reading or rewriting the entire dataset. This new capability addresses previous limitations where updating a single value required a full read-modify-write cycle using PutRecord, which often led to increased latency, higher costs from extra Read Capacity Units, and race conditions known as lost-update problems. The UpdateRecord function supports both Standard (DynamoDB-backed) and In-Memory (ElastiCache-backed) online store tiers, though Standard tier users must opt into the new Standard_V2 storage format to utilize this feature. For existing Standard tier customers, AWS offers two migration strategies: a bulk migration using the Feature Processor SDK for reversible changes, or an in-place switchover via the UpdateFeatureGroup API for zero-downtime transitions. The API ensures data integrity by enforcing EventTime-based ordering, rejecting stale updates that conflict with newer timestamps while preserving unchanged features atomically. Key use cases include streaming feature hydration, backfilling new schema fields, correcting data discrepancies at scale, and managing high-velocity updates in multi-producer environments. Additionally, the service integrates with IAM policies through new condition keys, allowing fine-grained access control over which users can perform partial updates and which specific features they are permitted to modify.
aws