Paolo Di Francesco

Paolo Di Francesco

MLflow contributor · MLOps & agentic AI at AWS

Accelerating ML experimentation with enhanced security AWS PrivateLink support for Amazon SageMaker with MLflow

1 minute read

Regulated environments don’t get to choose between experiment tracking and network isolation — they need both. With AWS PrivateLink support for Amazon SageMaker with MLflow, data sent to MLflow Tracking Servers stays inside the AWS network, never touching the public internet. In this post, we go all the way: a SageMaker environment in a private VPC with no internet access, running full ML experimentation with MLflow.

Full text here, and GitHub repository here

Co-authored with Xiaoyu Xing and Tomer Shenhar, this post walks through a fully deployable AWS CDK stack: a SageMaker Studio domain in a private VPC without internet access, AWS CodeArtifact as a private PyPI repository so packages can still be installed, and VPC endpoints connecting SageMaker to S3, ECR, CloudWatch, and the managed MLflow tracking server — all over PrivateLink. We then run data preprocessing and XGBoost training with the SageMaker Python SDK’s @remote decorator, with every job tracked in MLflow, and show how to lock the SageMaker execution role down with an aws:SourceVpc condition so MLflow is only reachable from inside the VPC.

Why I wrote this. My 2023 post on securing self-hosted MLflow solved authentication — who can talk to the tracking server. This one is the other half of enterprise security: network isolation — which paths the data is allowed to travel at all. With PrivateLink support in managed MLflow, the no-internet-access setups that regulated customers actually run finally work without workarounds.

You May Also Enjoy

Deploying MLflow models to Amazon SageMaker AI Hosting

Your best model sits in the MLflow Model Registry with a version number and a signature — and then someone asks: “Can we get that behind an endpoint?” With managed MLflow on Amazon SageMaker AI, the distance between “registered in MLflow” and “serving traffic” is much shorter than it used to be. ...

End-to-end lineage with DVC and Amazon SageMaker AI MLflow apps

Production ML teams often struggle to trace the full lineage of a model back to the exact data and code that trained it. In this post, we close that gap by combining DVC for data versioning, Amazon SageMaker AI for scalable processing and training, and Amazon SageMaker AI MLflow Apps for experime...