Edit

Azure CycleCloud version 8.9.1

This release introduces several new features and resolves existing issues.

New features

  • The Azure CycleCloud HealthAgent is upgraded with greatly improved configurability of health check conditions and thresholds.

  • The Azure CycleCloud Workspace for Slurm project provides a script to assist in deploying an Entra Application Registration for CycleCloud authentication. See the documentation for instructions.

  • Node properties now include the physical hostname, Nvidia ClusterUID, and platform interconnect group ID to allow persistent tracking of hardware health issues.

  • CycleCloud now supports provisioned performance for PremiumV2 SSD in Managed Disk.

  • The Azure CycleCloud monitoring project now installs new linked node and gpu health dashboards in the configured Azure Monitor Workspace.

  • Single VM clusters now converge without Chef by default. A configuration parameter is added to enable Chef explicitly for testing.

  • Azure CycleCloud Slurm cluster changes:

    • Support for Slurm 25.11.5.
    • The azslurm-exporter now exports sdiag metrics to the configured Azure Monitor Workspace.
    • When monitoring is enabled, the Slurm project provides a Slurm metrics dashboard for the configured Azure Monitor Workspace.

  • Azure CycleCloud UI changes:

    • The Azure CycleCloud “Node Summary” dialog is updated to provide additional information about the node including physical hostname and storage locker details.
    • The Azure CycleCloud “Node Details” dialog is enriched with additional node property data including physical host properties.

Resolved problems

  • Terminate notification didn't work for regular non-scale-set VMs.
  • Slurm clusters using Alma9 HPC Image didn't converge because of mismatched sqlite package versions.
  • The azslurm command line returned an empty exception during allocation when available node count is 0 for a slurm partition with idle~ nodes.
  • Azure CycleCloud nodes running SLES Linux virtual machine images failed during the "Awaiting boot-up" stage.
  • Slurm Clusters showed a temporary failure to converge when waiting for sacctmgr during start-services.sh before retrying.
  • Changing the resource group when editing an account didn't save correctly.
  • Cluster-init scripts unnecessarily re-ran when nodes were redeployed upon deallocation or planned maintenance.
  • The jetpack –h argument didn't display the jetpack help message.
  • Cluster startup sometimes failed with an IndexOutOfBounds exception if the requested virtual machine image couldn't be found.
  • Resolved CVEs: CVE-2026-43512, CVE-2026-42498, CVE-2026-41284, CVE-2026-43513, CVE-2026-41293, CVE-2026-57969
  • Last updated on