Amazon SageMaker Feature Store now supports batch ingestion and direct record discovery through two new APIs. BatchWriteRecord accepts as many as 25 entries across multiple feature groups in one request, while ListRecords enumerates active record identifiers in either the Standard or In-Memory storage tier.
Batch writes replace loops that previously required one PutRecord call for every record in every feature group. Each entry succeeds or fails independently, preserves event-time ordering and can set its own time-to-live value. Failed or unprocessed entries are returned for selective retries rather than rolling back successful writes. The API is therefore not transactional, and clients must avoid resubmitting records that already succeeded.
ListRecords addresses a different operational gap. Existing get and delete calls require the exact identifier, leaving In-Memory customers unable to find orphaned records if an index was lost. The new paginated call lists current identifiers while excluding deleted and expired entries, allowing teams to inspect or remove data without an offline store. Standard-tier queries may still use an online-store scan, so large inventories require attention to pagination and cost. Both additions are available through the SageMaker runtime APIs; Python users need the latest Boto3 release or SageMaker Python SDK 3.8.0 or later.