Edit

Share via

CycleCloud version 8.7.1

  • Article

New Features

  • The Slurm cluster template now configures the /sched volume to be persistent by default
  • The Slurm cluster template now allows users to add resource tags to execute nodearrays under the Advanced Settings menu
  • The Slurm cluster template now offers users the option to disable creating the dynamic partition with a checkbox in the Advanced Settings menu
  • The Slurm cluster template now uses MaxCount in place of MaxCoreCount; users set a maximum number of nodes rather than cores for each partition
  • The Slurm cluster template features a GPU partition
  • The Slurm cluster template provides a separate option for OS images and cluster-init specs for login nodes
  • The Slurm cluster template sets the /sched directory to be persistent
  • The CycleCloud CLI Entra authentication preview now offers a --use-device-code option to skip browser-interactive authentication on hosts with no browser
  • The CycleCloud CLI Entra authentication preview now provides support for initialization with a Service Principal or Managed ID
  • The max header size is now configurable, and the default was increased to 16k

Resolved Issues

  • The Slurm cluster type failed if the cluster name contained a dash (“-”)
  • The munge service for Slurm clusters was not set to restart upon reboot
  • Nodes in the Slurm dynamic nodes partition were registered incorrectly due to a missing gres flag during slurmd initialization
  • Slurm no longer reports an incorrect number of GPUs for the NDv5 VMs
  • Enabling builtin NFS for /sched with parameter NFSAddress defined caused converge failures on execute nodes
  • On-boot Node Health Checks failed on the most recent HPC VM image versions
  • The unused Thunderball Chef cookbook and references have been removed
  • The Active Directory and LDAP authentication forms were empty even after being configured
  • The DataStore Export UI widget failed to work when the button was clicked
  • The Edit Cluster dialog was slower than expected for clusters with many node arrays
  • Initializing CLI with as a user with the basic User role sometimes failed
  • The CycleCloud CLI had insufficient logging for Entra authentication failures

Known Issues

  • HPC Pack Broker nodes fail to converge breaking SOA workloads. This will be resolved in the next release.
  • The Altair GridEngine demo release is no longer supported. Altair GridEngine users must upload their licensed GridEngine installer binary to their CycleCloud Storage Locker.
  • On some Ubuntu systems, the “cycle_server_webserver” service may be flagged by the needrestart tool and trigger a CycleCloud restart on apt package install or upgrade. This may be worked around by running the following snippet (as root on the CycleCloud VM): 
    mkdir -p /etc/needrestart/conf.d
echo 'push(@{$nrconf{blacklist_mappings}}, qr(^/opt/cycle_server/system/tomcat/temp/));' > /etc/needrestart/conf.d/cycle_server.conf