# AWS Adds Managed Cluster Extensions for Observability and User Management in SageMaker HyperPod Slurm

> HyperPod now lets teams declare observability and user-management software through validated APIs or the console, applying it consistently across Slurm nodes.

- Publication: Thoai Dispatch
- Canonical URL: https://vokythoai.dev/article/aws-sagemaker-hyperpod-managed-cluster-extensions
- Section: News
- Author: AWS Machine Learning Blog
- Published: 2026-11-21
- Tags: aws, sagemaker-hyperpod, slurm, observability, cluster-management, user-management, cloudwatch, ml-infrastructure

## What happened

Amazon SageMaker HyperPod launched cluster extensions in November 2026, according to the [AWS Machine Learning Blog](https://aws.amazon.com/blogs/machine-learning/introducing-managed-cluster-extensions-for-observability-and-user-management-in-amazon-sagemaker-hyperpod-slurm/). The feature brings managed software packages for observability and user management, including directory integration, to Slurm clusters. Administrators install, configure, update, and remove these extensions through APIs or the Amazon SageMaker AI console.

The launch extends a year of declarative management work on HyperPod. Earlier in 2026, AWS introduced API-driven Slurm configuration, AMI-based configuration, continuous provisioning, automatic topology management, and deep health checks. Those features covered five layers of cluster configuration: definition, node setup, capacity, network topology, and hardware health. Cluster extensions address a sixth layer: the software running on nodes, which changes throughout the life of a cluster.

## Why it matters

Lifecycle scripts have been the primary mechanism for managing node software. The AWS post identifies several drawbacks: a script runs only when a node is provisioned, so each change requires editing the script and replacing or rerunning nodes; mistakes surface only when the script executes on a node; and nothing reports what is installed, allowing nodes to drift apart.

Cluster extensions aim to make software declarative. HyperPod validates an extension's settings against its schema when submitted, so a misspelled setting is rejected before reaching a node. The service applies the extension to every node, including nodes added by scale-up or replacement, and keeps nodes consistent as the cluster scales and heals. Each extension reports status and per-node health.

AWS states that users retain full access to every node, including the controller node, and can still SSH in, configure nodes directly, and run custom lifecycle scripts. The APIs reduce maintenance, while direct access preserves control.

## Technical details

The first managed extensions are the observability extension (`sagemaker-observability`) and a user-management extension. The observability extension collects node, GPU, Elastic Fabric Adapter (EFA), Slurm job, and Slurm cluster metrics and sends them to Amazon CloudWatch. The source material was truncated at that point, so the full set of destinations and the user-management extension's configuration details are not available here.

HyperPod's earlier declarative features provide context. API-driven Slurm configuration lets administrators define node roles, partitions, and Amazon FSx for Lustre and Amazon FSx for OpenZFS mounts in `CreateCluster` and `UpdateCluster` requests, and connect an external accounting database. AMI-based configuration replaces routine setup scripts with images preconfigured with Slurm, Docker, Enroot, Pyxis, and home directories. Continuous provisioning retries capacity requests in the background, and minimum capacity requirements use `MinInstanceCount`. Automatic topology management maintains network layout as nodes scale or are replaced; on Slurm 25.11 or later, partition-level topology supports tree or block topology per partition. Deep health checks stress GPUs and test NVIDIA Collective Communications Library (NCCL) connectivity, keeping nodes in Slurm maintenance reservations until checks finish.

## What to watch

Teams running shared Slurm clusters for foundation model development should evaluate whether managed extensions replace parts of their lifecycle scripts. AWS recommends using cluster extensions when HyperPod offers a managed capability, and custom AMIs and lifecycle scripts for organization-specific software or initialization not covered by extensions. The validation step and per-node health reporting are the main operational differences from scripts. Because the source material is truncated, readers should consult the original post for the complete list of supported extensions and configuration options.

## Sources

- [AWS Machine Learning Blog](https://aws.amazon.com/blogs/machine-learning/introducing-managed-cluster-extensions-for-observability-and-user-management-in-amazon-sagemaker-hyperpod-slurm/) (2026-11-20)
