Amazon SageMaker Feature Store now supports SageMaker Python SDK V3
Fine-grained Lake Formation access controls and Iceberg storage tuning now available directly in the SageMaker Feature Store SDK—no separate tools needed.
View original announcement →Visual Summary
graph TD
A{{SageMaker Feature Store SDK v3}}:::announced
B((Data Scientists)):::external
C([Lake Formation Access Controls]):::feature
D([Iceberg Table Properties]):::feature
E(AWS Lake Formation):::compute
F(Amazon S3):::storage
G(Apache Iceberg Tables):::storage
H([Column & Row Level Security]):::feature
I(Amazon Athena):::compute
B ==>|"uses SDK v3"| A
A -->|"enforces"| C
A -->|"configures"| D
C -->|"registers with"| E
E -->|"governs"| H
D -->|"compaction & snapshots"| G
G -->|"stored in"| F
I -.->|"queries"| G
classDef announced fill:#ff9900,stroke:#ec7211,color:#fff,font-weight:bold
classDef compute fill:#e3f2fd,stroke:#1565c0,color:#1565c0
classDef storage fill:#e8f5e9,stroke:#2e7d32,color:#2e7d32
classDef feature fill:#fff3e0,stroke:#e65100,color:#e65100
classDef external fill:#f5f5f5,stroke:#616161,color:#616161
What's New
Amazon SageMaker Feature Store now supports the SageMaker Python SDK v3, bringing a modernized, modular interface for managing feature groups alongside two significant new capabilities. Data scientists can now enforce column-level and row-level access controls on offline store data via AWS Lake Formation integration, and configure Apache Iceberg table properties such as compaction and snapshot expiration directly through the SDK. These additions allow teams to govern feature data access and optimize offline store performance without leaving the SDK or managing separate tooling.
How It Works
- SDK v3 Interface: Install SageMaker Python SDK v3.8.0 or later to access the new modular Feature Store APIs, which reduce boilerplate code and streamline feature group lifecycle management compared to the v2 SDK.
- Lake Formation Access Controls: At feature group creation time, data scientists can opt in to Lake Formation integration, which registers the offline store's Iceberg table with AWS Lake Formation and enables fine-grained column-level and row-level permissions to be applied through standard Lake Formation policies.
- Iceberg Table Properties Configuration: The SDK v3 exposes Iceberg table property settings—such as compaction strategies and snapshot expiration intervals—that can be set directly when creating or updating a feature group, allowing the offline store to be tuned for query performance and storage efficiency.
- Offline Store Backend: The offline store continues to use Apache Iceberg as the underlying table format in Amazon S3, and the new properties configuration maps directly to native Iceberg table properties supported by the format.
- Single SDK Workflow: All governance and optimization operations are surfaced through the same SDK, eliminating the need to separately invoke AWS Glue, Lake Formation APIs, or Iceberg tooling outside of the SageMaker context.
Why It's Important
- Stronger Data Governance: Column-level and row-level access controls allow organizations to enforce least-privilege access to sensitive feature data (e.g., PII columns), which is critical for regulated industries and multi-team environments where not all consumers should see all features.
- Reduced Operational Complexity: Previously, teams had to manage Lake Formation permissions and Iceberg table maintenance (compaction, snapshot cleanup) through separate AWS consoles or APIs; consolidating these into the SDK reduces the operational surface area and potential for misconfiguration.
- Improved Offline Store Performance: Configuring compaction and snapshot expiration directly addresses common pain points with Iceberg tables at scale—unchecked snapshot accumulation degrades query planning performance, and lack of compaction leads to small-file proliferation that slows Athena or EMR queries.
- Modern SDK Adoption Path: The v3 SDK's modular design makes it easier to integrate Feature Store operations into CI/CD pipelines and MLOps workflows, lowering the barrier for teams migrating from v2.
- Unified Governance Model: Aligning Feature Store permissions with Lake Formation means security teams can manage ML feature access using the same governance tooling they already use for the broader data lake, simplifying auditing and compliance.
How It's Different
- vs. SDK v2 Feature Store: The v2 SDK required more boilerplate to manage feature groups and had no native support for Lake Formation integration or Iceberg property configuration, forcing teams to use separate AWS service calls for governance and maintenance tasks.
- vs. Manual Lake Formation Setup: Previously, enabling Lake Formation controls on Feature Store offline data required manual registration and permission grants outside the SDK; the new opt-in flag at feature group creation automates this integration end-to-end.
- vs. External Iceberg Management Tools: Teams using Apache Spark or direct Iceberg APIs to run compaction and expire snapshots on Feature Store tables can now configure these behaviors declaratively through the SDK, removing the need for separate scheduled jobs or external orchestration.
- vs. Table-Level Access Only: Prior access control options were coarser-grained; the Lake Formation integration introduces column-level and row-level granularity, which is a meaningful step up for multi-tenant or privacy-sensitive feature sharing scenarios.
When to Prefer It
- Multi-Team Feature Sharing with Sensitive Data: When multiple teams or business units share a feature group but should only access specific columns (e.g., one team sees demographic features but not financial features), use Lake Formation column-level controls to enforce this boundary at the storage layer.
- Regulated Industries: When operating in healthcare, finance, or other regulated domains where audit trails and least-privilege data access are compliance requirements, the Lake Formation integration provides the necessary governance controls on feature data.
- Large-Scale Offline Stores with Query Performance Issues: When feature groups accumulate many small files or large numbers of Iceberg snapshots over time, configure compaction and snapshot expiration properties at creation to proactively manage storage and maintain fast query performance.
- MLOps Pipelines Requiring Programmatic Governance: When building automated pipelines that provision feature groups as part of infrastructure-as-code or CI/CD workflows, the SDK v3's reduced boilerplate and declarative property configuration simplify scripting and reduce error-prone manual steps.
- Teams Migrating from SDK v2: When modernizing existing Feature Store workflows, the v3 SDK provides a natural migration path that also unlocks the new governance and optimization capabilities not available in v2.
Availability
- GA Status: Generally available as of May 12, 2026; this is not a preview release.
- Supported Regions: Available in all AWS Regions where Amazon SageMaker Feature Store is currently supported.
- SDK Requirement: Requires SageMaker Python SDK v3.8.0 or later; the new capabilities are not available in SDK v2.
- Lake Formation Opt-In: Lake Formation access control integration is an opt-in setting configured at feature group creation time; it cannot be retroactively enabled on existing feature groups.
- Pricing: No additional charge is announced for the SDK v3 support itself; standard SageMaker Feature Store, Amazon S3, AWS Lake Formation, and AWS Glue pricing applies for underlying resources.
- Limitations: Iceberg table property support and Lake Formation integration depend on the offline store using the Iceberg table format; feature groups using the Glue table format may not support all new capabilities.
