Amazon SageMaker Feature Store Introduces New APIs for Record Management

Amazon SageMaker Feature Store, a fully managed repository for machine learning features, has announced the addition of two new APIs: BatchWriteRecord and ListRecords. These APIs address key operational challenges faced by ML teams, such as high-throughput data ingestion and record discovery.

The BatchWriteRecord API allows users to write up to 25 records across multiple feature groups in a single request, significantly reducing connection overhead and improving throughput. It supports partial success, per-record TTL control, and preserves EventTime-based ordering guarantees, ensuring data consistency.

The ListRecords API enables enumeration of record identifiers within a feature group, facilitating record lifecycle management and cleanup, especially for In-Memory storage tiers where records can become orphaned. It works with both DynamoDB-backed and Redis-backed storage tiers, supporting pagination for efficient browsing.

These enhancements enable more scalable and operationally efficient workflows, such as bulk data ingestion pipelines, compliance data deletion, and real-time feature group browsing. The APIs are available with the necessary IAM permissions and require the latest SDK versions.

Overall, these updates improve the robustness and operational flexibility of Amazon SageMaker Feature Store, supporting advanced ML workflows and data governance practices.