Skip to main content

July 11, 2025 - SUNK v6.6.0 release

SUNK v6.6.0 released with SCIM provisioning, dashboard links, improved node reconciliation, new compute definitions, metrics fixes, segment-calc improvements, and base image upgrades

Update SUNK SUNK v6.6.0 has been released with SCIM provisioning support, enhanced monitoring capabilities, improved node management, new compute definitions, and various infrastructure improvements.

Overview​

SUNK v6.6.0 introduces SCIM provisioning via nsscache, enhanced monitoring with dashboard links, improved node reconciliation, new GPU compute definitions, metrics fixes, segment-calc script improvements, and base image upgrades.

Key changes​

SCIM provisioning via nsscache​

SCIM provisioning for SUNK is now available via nsscache, enabling automated user and group management from your Identity Provider (IdP) to CoreWeave clusters.

Configuration requirements​

The SCIM suggested settings can be found in the slurm chart values-cw.yaml at nsscache.nsscacheConfig.

Required changes from current defaults in the values.yaml files:

  1. nsscache does not use SSSD and it is best set to false:

    • In the Slurm chart: sssdContainer.enabled: false
    • In the Slurm-login chart: Change to directoryCache.source: nsscache

  2. Requires nsscache.existingSecret to be present with auth token in it:

    • nsscache-scim-auth-token key inside the secret with the token

Added ability for dashboard links to be added to SLURM output, providing direct access to monitoring dashboards for jobs and nodes.

Improved node reconciliation​

Enhanced reconcile logic for nodes that are in a reservation, improving cluster stability and resource management.

New compute definitions​

Added new GPU compute definitions:

  • rtxp6000-8x: NVIDIA RTX Pro 6000 Blackwell Server Edition
  • gb300-4x: NVIDIA GB300

Metrics improvements​

  • Fixed adding slurm_cluster label on VMPodScrape Metrics: Resolved label consistency issues
  • Added scraping of MySQL metrics via VMPodScrape: Enhanced monitoring coverage for database components

Segment-calc script improvements​

Improved segment-calc script for visualizing block scheduling, providing better insights into cluster resource allocation and utilization patterns.

Base image upgrades​

Upgraded nccl-test base images for improved performance and compatibility with the latest NVIDIA software stack.

Configuration changes​

SCIM setup​

To enable SCIM provisioning:

  1. Set sssdContainer.enabled: false in the slurm chart
  2. Change directoryCache.source: nsscache in the slurm-login chart
  3. Create a secret with the nsscache-scim-auth-token key containing your SCIM authentication token
  4. Configure nsscache.existingSecret to reference your secret

Compute definitions​

The new compute definitions are automatically available and can be used in job submissions and node configurations.

Migration notes​

Existing SUNK deployments will continue to work, but you may want to:

  1. Review your SSSD configuration if you plan to use SCIM provisioning
  2. Test the new compute definitions in a non-production environment
  3. Update any custom monitoring configurations to take advantage of the new metrics
  4. Verify your SCIM configuration with the new nsscache integration

Additional resources​

For detailed information about configuring and using SUNK v6.6.0, see:

Last updated on