aws / aws/sagemaker-python-sdk

FeatureStore create_dataset() does not support cross-account Feature Store (shared Glue database name mismatch)

Open
#5,412 0 comments 0 reactions 0 assignees View on GitHub
component: feature store type: feature request
Dominant language
Python
Stars
2.3k
Forks
1.3k
Avg merge
1d 22h
Merged PRs (30d)
35

Description

**Description**

When using SageMaker Feature Store in a cross-account setup, create_dataset() fails because it assumes the Glue database name returned from the Feature Group description is directly usable in the consumer account.

In cross-account architectures, this assumption breaks.

**Details**

- Feature Groups are owned in a source account
- Offline store is accessed from a consumer account via a shared Glue database
- The Feature Group description returns the database name from the source account
- The shared Glue database name in the consumer account is different
- create_dataset() retrieves the database name directly from the Feature Group description without any way to override or remap it:

As a result, create_dataset() queries a database that does not exist in the consumer account.

**Expected behavior**

The SDK should support cross-account Feature Store usage by allowing:

- A database name override, or
- A mapping mechanism for shared Glue databases

Contributor guide

Open the contributing guide

Research direction

Start at FeatureStore create_dataset() and trace how it obtains the Glue database name from the Feature Group description. Compare that value with the shared database name used in a consumer account, then determine how the requested override or mapping should be represented. Done means cross-account dataset creation can target the consumer database and the behavior is covered by relevant tests.

Written by the indexing model from the issue text.

Assessment

Tech stack
aws, python
Domain
cloud, databases, machine-learning
Issue type
Feature
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Mostly clear
Newbie friendliness
35/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.