Enable Lake Formation with Feature Groups

When you enable AWS Lake Formation on a feature group in Amazon SageMaker Feature Store, you can enforce column-level, row-level, and cell-level security for the feature data in your offline store. Instead of managing access through individual IAM policies on Amazon S3 and AWS Glue resources, you use the Lake Formation grant and revoke permission model to control which users and roles can access specific features and records. For more information about Lake Formation, see the AWS Lake Formation Developer Guide.

Prerequisites

Before you enable Lake Formation, verify that you have the following:

Key concepts

The following concepts are important for understanding how Lake Formation works with Feature Store.

Lake Formation permission model compared to IAM-only model

By default, access to AWS Glue Data Catalog tables, including those created by Feature Store, is controlled through IAM policies alone. When you enable Lake Formation, access requires both IAM permissions and Lake Formation permissions. Lake Formation uses a grant and revoke model where you explicitly grant permissions, such as SELECT or DESCRIBE, on specific databases, tables, or columns to IAM principals.

Hybrid access mode

When you enable Lake Formation, you choose whether to use hybrid access mode or Lake Formation-only mode:

S3 deny policy

Even after you enable Lake Formation on the AWS Glue Data Catalog table, users with direct Amazon S3 access (through IAM policies) can bypass Lake Formation by reading the underlying Amazon S3 objects directly. To close this gap, apply an Amazon S3 bucket policy that denies direct access to the offline store prefix for all principals except the Lake Formation service role and the Feature Store execution role.

Configuration reference

Use the LakeFormationConfig class to configure Lake Formation. You pass this configuration to FeatureGroupManager.create() when creating a new feature group, or use the individual parameters directly with enable_lake_formation() for existing feature groups.

Understanding acknowledge_risk

The acknowledge_risk parameter is a safety gate. By setting it to True, you acknowledge the following:

How it works

When you enable Lake Formation, the SDK performs a three-phase setup:

After all phases complete, the SDK logs a recommended S3 bucket deny policy. Review and apply this policy to prevent direct Amazon S3 access that bypasses Lake Formation access control.

Usage examples

Create a new feature group with Lake Formation access control

The following example creates a new feature group with an offline store and enables Lake Formation access control in a single operation.

from sagemaker.core.helper.session_helper import Session
from sagemaker.core.shapes import (
    FeatureDefinition,
    OfflineStoreConfig,
    OnlineStoreConfig,
    S3StorageConfig,
)
from sagemaker.mlops.feature_store import FeatureGroupManager, LakeFormationConfig

session = Session()
region = session.boto_region_name

role_arn = "arn:aws:iam::<account-id>:role/<execution-role-name>"
registration_role_arn = "arn:aws:iam::<account-id>:role/<registration-role-name>"
fg_name = "my-feature-group"
bucket = session.default_bucket()
offline_s3_uri = f"s3://{bucket}/feature-store/{fg_name}"

feature_definitions = [
    FeatureDefinition(feature_name="customer_id", feature_type="String"),
    FeatureDefinition(feature_name="event_time", feature_type="String"),
    FeatureDefinition(feature_name="age", feature_type="Integral"),
    FeatureDefinition(feature_name="total_purchases", feature_type="Integral"),
    FeatureDefinition(feature_name="avg_order_value", feature_type="Fractional"),
]

fg = FeatureGroupManager.create(
    feature_group_name=fg_name,
    record_identifier_feature_name="customer_id",
    event_time_feature_name="event_time",
    feature_definitions=feature_definitions,
    online_store_config=OnlineStoreConfig(enable_online_store=False),
    offline_store_config=OfflineStoreConfig(
        s3_storage_config=S3StorageConfig(s3_uri=offline_s3_uri),
        table_format="Iceberg",
    ),
    role_arn=role_arn,
    description="A feature group with Lake Formation-managed offline store",
    # Example configuration — adjust parameters for your use case.
    lake_formation_config=LakeFormationConfig(
        enabled=True,
        use_service_linked_role=False,
        registration_role_arn=registration_role_arn,
        # Set to True to keep existing IAM-based access (hybrid mode).
        hybrid_access_mode_enabled=False,
        acknowledge_risk=True,
    ),
    region=region,
)

Enable Lake Formation on an existing feature group

The following example enables Lake Formation on a feature group that already exists.

from sagemaker.mlops.feature_store import FeatureGroupManager

fg = FeatureGroupManager.get(feature_group_name="my-existing-feature-group")
result = fg.enable_lake_formation(
    use_service_linked_role=False,
    registration_role_arn="arn:aws:iam::<account-id>:role/<registration-role-name>",
    hybrid_access_mode_enabled=False,
    acknowledge_risk=True,
)

Use Lake Formation-enabled feature groups with SageMaker AI jobs

You can access Lake Formation-enabled feature groups from SageMaker AI training and processing jobs. To read Lake Formation-enabled feature data, you can either query the data through Athena or use the Lake Formation GetTemporaryGlueTableCredentials API to vend temporary Amazon S3 credentials scoped to the AWS Glue table. For more information about configuring Lake Formation permissions for compute roles, see Lake Formation permissions reference in the Lake Formation documentation.

Cross-account access

Lake Formation supports sharing feature group data across AWS accounts. When the producer account enables Lake Formation on a feature group, it can grant cross-account access to consumer accounts using either the named resource method or Lake Formation tag-based access control (LF-TBAC).

Cross-account sharing requires the following setup:

For prerequisites, step-by-step instructions, and best practices for cross-account sharing, see the following topics in the AWS Lake Formation Developer Guide:

Grant fine-grained access after setup

After you enable Lake Formation, the Lake Formation administrator can grant fine-grained permissions to other IAM principals using the Lake Formation console, API, or CLI. Lake Formation supports column-level, row-level, and cell-level access control.

For instructions on granting and managing Lake Formation permissions, see the following topics in the AWS Lake Formation Developer Guide: