aws/aws-parallelcluster v3.2.0

AWS ParallelCluster v3.2.0

on GitHub

We're excited to announce the release of AWS ParallelCluster 3.2.0

Upgrade

How to upgrade?

sudo pip install --upgrade aws-parallelcluster

ENHANCEMENTS

• Add support for memory-based job scheduling in Slurm
  • Configure compute nodes real memory in the Slurm cluster configuration.
  • Add new configuration parameter Scheduling/SlurmSettings/EnableMemoryBasedScheduling to enable memory-based scheduling in Slurm.
  • Add new configuration parameter Scheduling/SlurmQueues/ComputeResources/SchedulableMemory to override default value of the memory seen by the scheduler on compute nodes.
• Improve flexibility on cluster configuration updates to avoid the stop and start of the entire cluster whenever possible.
  • Add new configuration parameter Scheduling/SlurmSettings/QueueUpdateStrategy to set the preferred strategy to adopt for compute nodes needing a configuration update and replacement.
• Improve failover mechanism over available compute resources when hitting insufficient capacity issues with EC2 instances. Disable compute nodes by a configurable amount of time (default 10 min) when a node launch fails due to insufficient capacity.
• Add support to mount existing FSx for ONTAP and FSx for OpenZFS file systems.
• Add support to mount multiple instances of existing EFS, FSx for Lustre / for ONTAP/ for OpenZFS file systems.
• Add support for FSx for Lustre Persistent_2 deployment type when creating a new file system.
• Prompt user to enable EFA for supported instance types when using pcluster configure wizard.
• Add support for rebooting compute nodes via Slurm.
• Improved handling of Slurm power states to also account for manual powering down of nodes.
• Add NVIDIA GDRCopy 2.3 into the product AMIs to enable low-latency GPU memory copy.

CHANGES

• Upgrade EFA installer to version 1.17.2
  • EFA driver: efa-1.16.0-1
  • EFA configuration: efa-config-1.10-1
  • EFA profile: efa-profile-1.5-1
  • Libfabric: libfabric-aws-1.16.0~amzn2.0-1
  • RDMA core: rdma-core-41.0-2
  • Open MPI: openmpi40-aws-4.1.4-2
• Upgrade NICE DCV to version 2022.0-12760.
• Upgrade NVIDIA driver to version 470.129.06.
• Upgrade NVIDIA Fabric Manager to version 470.129.06.
• Change default EBS volume types from gp2 to gp3 for both the root and additional volumes.
• Changes to FSx for Lustre file systems created by ParallelCluster:
  • Change the default deployment type to Scratch_2.
  • Change the Lustre server version to 2.12.
• Do not require PlacementGroup/Enabled to be set to true when passing an existing PlacementGroup/Id.
• Add parallelcluster:cluster-name tag to all the resources created by ParallelCluster.
• Do not allow setting PlacementGroup/Id when PlacementGroup/Enabled is explicitly set to false.
• Add lambda:ListTags and lambda:UntagResource to ParallelClusterUserRole used by ParallelCluster API stack for cluster update.
• Restrict IPv6 access to IMDS to root and cluster admin users only, when configuration parameter HeadNode/Imds/Secured is true as by default.
• With a custom AMI, use the AMI root volume size instead of the ParallelCluster default of 35 GiB. The value can be changed in cluster configuration file.
• Automatic disabling of the compute fleet when the configuration parameter Scheduling/SlurmQueues/ComputeResources/SpotPrice
  is lower than the minimum required Spot request fulfillment price.
• Show requested_value and current_value values in the change set when adding or removing a section during an update.
• Disable aws-ubuntu-eni-helper service in DLAMI to avoid conflicts with configure_nw_interface.sh when configuring instances with multiple network cards.
• Remove support for Python 3.6.
• Set MTU to 9001 for all the network interfaces when configuring instances with multiple network cards.
• Remove the trailing dot when configuring the compute node FQDN.

BUG FIXES

• Fix the default behavior to skip the ParallelCluster validation and test steps when building a custom AMI.
• Fix file handle leak in computemgtd.
• Fix race condition that was sporadically causing launched instances to be immediately terminated because not available yet in EC2 DescribeInstances response
• Fix support for DisableSimultaneousMultithreading parameter on instance types with Arm processors.
• Fix ParallelCluster API stack update failure when upgrading from a previus version. Add resource pattern used for the ListImagePipelineImages action in the EcrImageDeletionLambdaRole.
• Fix ParallelCluster API adding missing permissions needed to import/export from S3 when creating an FSx for Lustre storage.