Amazon SageMaker Feature Store introduces UpdateRecord for feature-level writes

Amazon SageMaker Feature Store now supports update records, allowing users to update specific feature values within a feature group without having to read, modify, and rewrite the entire record. This capability is available for both the Standard (DynamoDB-backed) and In-Memory (ElastiCache-backed) online store tiers. Traditionally, updating a single feature value required a full read-modify-write cycle using PutRecord, which added latency, consumed unnecessary read capacity, and introduced race conditions.

The new UpdateRecord API call streamlines this process by allowing users to provide only the features they wish to change, with the updates applied atomically to the existing record. Features not included in the request remain unchanged. The UpdateRecord API data flow involves the client application calling the API with the changed features, which are then validated, ordered by event time to prevent stale writes, and merged atomically.

A full record snapshot is automatically replicated to the offline store for accurate training datasets. Multiple pipelines can write their own features independently to the same record without coordination. To use feature-level writes with the Standard Tier, a new storage format named Standard_V2 must be opted into during feature group creation.

Existing Standard storage type feature groups can be migrated to Standard_V2 using the Feature Processor SDK, either through a bulk migration with downtime or a rolling migration with minimal impact.