SageMaker Feature Store Gets Batch Writes and Record Listing — Way Easier to Use!

Hey everyone, it's Shiichan! Today I found a quietly useful update on the machine learning feature side.

AWS What's New aws.amazon.com

What was announced?

According to AWS's What's New, Amazon SageMaker Feature Store now supports batch feature writes and record listing. It's a small-sounding update, but I think it'll really help anyone working with large volumes of data.

The story so far

Until now, writing records to Feature Store basically meant calling the API one record at a time. Ingesting a lot of data meant a lot of requests, and finding a record meant already knowing its identifier ahead of time.

What changes

With the new BatchWriteRecord, you can write multiple records across multiple feature groups in a single request — fewer API calls and lower latency than writing one at a time. It targets the online store, the offline store, or both, and returns individual record failures without failing the whole request, so partial ingestion issues are easier to handle. You can also set time-to-live at the record, request, or feature group level.

The other addition, ListRecords, lets you page through a feature group's contents without knowing record identifiers in advance. That's handy for browsing and auditing what's in a feature group, recovering identifiers you've lost track of, and managing record lifecycles.

Dive Deep

There's also a change to offline store configuration: you can now create Glue and Iceberg tables with custom names. Nice if your team likes to keep its data catalog tidy.

It's available in every AWS Region where Amazon SageMaker Feature Store is offered.

Wrap-up

• SageMaker Feature Store adds BatchWriteRecord (batch writes) and ListRecords (record listing)
• BatchWriteRecord writes multiple records across multiple feature groups in one request, and reports per-record failures individually
• TTL can be set at the record, request, or feature group level
• ListRecords pages through records without needing identifiers up front, useful for audits and recovery
• Offline store Glue and Iceberg tables can now have custom names

If you're a data scientist or MLOps engineer ingesting and managing lots of feature data, this quiet update will make a real difference in your day-to-day!