AWS adds managed user administration to HyperPod Slurm clusters
Administrators can set group login and privilege rules by node type in SageMaker and connect eligible clusters to Microsoft Active Directory.
AWS announced a managed extension for user administration on November 20 for Amazon SageMaker HyperPod clusters that use Slurm to schedule training jobs. AWS says administrators can manage users, groups and access across eligible nodes from the SageMaker AI console or API, and the extension is available in every region that offers HyperPod.
In the console’s Users tab, administrators can add, edit or remove local users and groups. AWS says administrators can set access by node type, choosing where group members can log in remotely over SSH or run commands with elevated privileges through sudo.
Clusters must use Continuous node provisioning and run a default SageMaker HyperPod Amazon Machine Image released on or after November 2, 2026, in every instance group. Custom images are ineligible.
AWS says HyperPod applies the same local user and group IDs to new and replacement nodes. The company warns that those nodes can accept Slurm jobs one to two minutes before the accounts are ready. A job can fail in that window if it needs an account that has yet to be created.
Administrators can also connect to Microsoft Active Directory over an encrypted connection and use directory identities alongside local accounts. AWS says each cluster can use one directory domain, and directory users and group memberships stay under Active Directory’s control. The extension’s 10,240-character configuration limit fits about 40 to 60 local users with groups, according to AWS.
AWS announced enhanced observability for GPU training on November 19. On Slurm, AWS provides that capability through a separate observability extension.
AWS charges no separate fee for the user management extension. Customers still pay for the AWS resources they use with the extension.
More news

SGLang and Miles add day-zero support for Inkling 2 Mini

VAST Data announces per-job storage controls for VAST AI OS 5.6

AWS introduces cross-Region inference routing for SageMaker HyperPod
