Generally available: Azure HPC – CycleCloud 8.2.0

CycleCloud 8.2.0 is now available. This release contains the integration with Microsoft HPC Pack and a considerable list of resolved issues. Please review the list in detail below.

New Features:

• New PBS Autoscale project provides more flexibility and insight into decisions via powerful azpbs cli
• CycleCloud now supports Microsoft HPC Pack.
• CycleCloud now includes a new cluster-cost API
• VMs are now tagged with the nodearray as well as the cluster, for nodearray-based cost tracking
• CycleCloud now supports Disk Encryption with Customer Managed Keys for OS (Boot Volume) and Data Disks.
• SGE/UGE now supports Ubuntu 18
• Univa GridEngine is now Altair GridEngine

Resolved Issues:

• An elevation-of-privilege vulnerability existed for a user with access to CycleCloud (CVE-2021-33762)
• Some files in JETPACK_HOME had incorrect permissions allowing non-group readers
• Mitigated an XXE vulnerability in the REST API (CVE-2021-36943)
• Every CycleCloud head node would get the same computer name by default
• Logging messages from nodes were grouped together incorrectly on the Events page
• The Event Grid connection feature did not work with more than one credential installed
• CycleCloud would fail to start if EventGrid was enabled while credentials lacked EventGrid permissions
• Jetpack installations failed on Windows 2019.
• Overprovisioned VMs would show jetpackd errors while awaiting registration in CycleCloud HPC Pack cluster did not offer ReturnProxy
• CycleCloud would show a spurious NullPointerException error when using SSD-based OS disks
• cyclecloud project upload would sometimes fail when checking for the prior existence of a file in Azure Storage
• Jetpack would not return an error when projects referenced in the cluster did not exist in storage
• Removed invalid LICENSE file from LSF project.
• Removed code that triggered an SELinux error with SGE on EL8 platforms.
• In some cases, nodes would be stuck waiting for spot eviction to complete
• Editing node properties in the UI would fail when a cloud-init script was defined
• LSF was leaving orphaned nodes.
• Jetpack installation could sometimes be halted by a rate-limited metadata server
• The custom settings for a graph on the cluster page were reverted when the page auto-refreshed
• Network metrics collected were incorrectly treating rates as bytes transferred
• Spot evictions detected in advance by CycleCloud were not labeled as spot-eviction as the reason in Event Grid
• Nodes that encountered certain errors while starting would get stuck in Acquiring state
• Keyboard navigation was broken.
• Quota consumption for limited-core HBv3 variants was treated as the limited count, not 120 cores
• Core-limited SKUs (eg, HB120-32v3) did not show prices
• SSD disks used the wrong price for sizes exactly between two levels
• Upgraded default Slurm install to 20.11.7
• Updated embedded azcopy to v10.11.0

Read the release notes.