Amazon SageMaker HyperPod now supports Amazon EBS CSI driver for persistent storage

Amazon SageMaker HyperPod now supports the Amazon Elastic Block Store (EBS) Container Storage Interface (CSI) driver, enabling customers to dynamically provision and manage persistent storage for machine learning workloads on SageMaker HyperPod EKS clusters. This capability allows customers to create, attach, and manage EBS volumes through Kubernetes persistent volume claims, providing storage that persists across pod restarts and node replacements. Customers deploying training and inference workloads need flexible storage allocation while maintaining high performance, but previously required manual EBS volume management outside Kubernetes workflows.

EBS CSI driver support enables customers to dynamically provision volumes based on model requirements, resize volumes without service disruption, and create snapshots for backup and recovery. For training workloads, this provides persistent storage for datasets, model checkpoints, and shared artifacts. For inference workloads, customers can provision model storage, create caching volumes, and maintain event logging. The integration supports both static and dynamic provisioning through Kubernetes storage classes, optimizing storage costs and performance.

To get started, install the Amazon EBS CSI driver as an EKS add-on on your HyperPod EKS cluster, then provision EBS volumes using standard Kubernetes persistent volume claims and storage classes. The EBS CSI driver manages the complete lifecycle of EBS volumes, including creation, attachment, mounting, and cleanup. Volume encryption with customer-managed KMS keys is supported, and volumes can be resized and snapshotted through standard Kubernetes operations.

This feature is available in all AWS Regions where SageMaker HyperPod EKS clusters are supported. To learn more about EBS CSI driver support, see the Amazon SageMaker HyperPod User Guide.