Slurm is updated to 20.11.4 and supports CentOS 8 and Ubuntu 18.
Infiniband-based SKUs have accelerated networking disabled by default, to avoid device-ordering issues
Resolved Issues:
MPI jobs in Slurm failed due to missing slurm-libmpi
Setting webServerContextPath in cycle_server.properties did not work properly
Pre-installing Jetpack on an image would incorrectly report an installation failure.
Default NFS mounting options were incompatible with NFS version 3
Jetpack PATH was incorrectly ordered on Windows.
Nodes whose images had embedded data disks could not be restarted once terminated
CycleCloud 7.x clusters would not run cluster-init if CycleCloud was upgraded to version 8
When scaling up large clusters, CycleCloud's self-throttling was too strict, keeping it from making progress
Execute nodes would sometimes get "resource not found" errors that would prevent scaling up
Healthcheck did not support Python3
Evicted spot nodes for MPI jobs in Slurm could not be restarted
In some cases, nodes would not start with the following error: "This node does not match existing scaleset attributes: Configuration.cyclecloud.mounts.additional_nfs.export_path, Configuration.cyclecloud.mounts.additional_nfs.mountpoint"
In certain cases, execute nodes could not be restarted after editing the cluster through the UI
Nodes with disks that referenced snapshots could not be started
Jetpack would incorrectly handle a list of IP addresses for network shares
The restore.sh script failed because it tried to start the (now-removed) broker service
Waiting for CloudInit would timeout after 5 minutes (increased to 60 minutes)
Page rendering was broken in some versions of Safari
The image picker displayed the incorrect image ID for CentOS 7
CycleCloud CLI incorrectly used the deprecated tostring() method on arrays
CycleCloud encountered a "Conversion: 2" error when processing insufficient-quota error messages for VMs