Maintained with ☕️ by
IcePanel logo

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]([.]( * Azure CycleCloud * Features * [ Azure CycleCloud](