Document history for the AWS PCS User Guide

AWS PCS is now available in Asia Pacific (Malaysia) (ap-southeast-5).

CloudFormation templates are available to get started in the Asia Pacific (Malaysia) AWS Region. For more information, see Use CloudFormation to create a sample AWS PCS cluster and CloudFormation templates to create a sample AWS PCS cluster.

AWS PCS updated the AWSPCSServiceRolePolicy to support multiple network interfaces for compute instances. For more information, see AWS managed policies for AWS Parallel Computing Service.

AWS PCS supports Slurm 26.05. Clusters running 25.11, 25.05, and 24.11 can update directly to 26.05. A single sample AMI now supports every Slurm version that is not EOL, and the AMI name no longer contains a Slurm version. For more information, see Slurm versions in AWS PCS and Release notes for Slurm versions in AWS PCS.

The MIG topic now lists what a MIG node needs on every boot, and recommends a node lifecycle action that applies it and fails the node if MIG mode can't be enabled. For more information, see Configuring Multi-Instance GPU (MIG) in AWS PCS.

You can configure Slurm Generic Resource (GRES) records for a compute node group with the gresCustomSettings property, and declare the schedulable GRES with the Gres custom Slurm setting. AWS PCS also enables GRES autodetection by default on compute node groups that have GPUs. For more information, see Configuring custom GRES settings in AWS PCS.

The Slurm REST API reads the username from the name field within the id claim of a JWT. The top-level sun, username, and custom userclaimfield claims are deprecated on Slurm 26.05 and later. For more information, see Authenticating with Slurm REST API in AWS PCS.

Added a topic that describes how Slurm models CPU topology and GPU affinity, what AWS PCS configures on each Slurm version, and how to configure the socket layout of a compute node group. A troubleshooting topic for INVALID_REG node registrations is also available. For more information, see Configuring hardware topology in AWS PCS.

Added a topic that describes how to partition NVIDIA GPUs into Multi-Instance GPU (MIG) instances and schedule them as GRES types on a compute node group, including the known limitations for powered-down nodes. For more information, see Configuring Multi-Instance GPU (MIG) in AWS PCS.

AWS PCS is now available in Asia Pacific (Melbourne) (ap-southeast-4).

CloudFormation templates are available to get started in the Asia Pacific (Melbourne) AWS Region. For more information, see Use CloudFormation to create a sample AWS PCS cluster and CloudFormation templates to create a sample AWS PCS cluster.

Updated the login node configuration script topic to describe supplying cluster details on the command line to skip the AWS PCS API call. Also added guidance for connecting one login node to multiple AWS PCS clusters, including clusters in different AWS Regions.

AWS PCS is now available in Asia Pacific (New Zealand) (ap-southeast-6).

CloudFormation templates are available to get started in the Asia Pacific (New Zealand) AWS Region. For more information, see Use CloudFormation to create a sample AWS PCS cluster and CloudFormation templates to create a sample AWS PCS cluster.

Corrected the example script for retrieving the JWT signing key to use --version-id instead of --version-stage, and updated endpoint URLs in the Slurm REST API topics to use http://, because the slurmrestd endpoint is HTTP only.

Added guidance that AWS PCS does not support rebooting the instance during a lifecycle action script.

Added AWS PCS agent version 1.5.1-1 in the user guide.

Added a limitation to the cluster version update page: clusters that use Slurm CLI Filter Plugins don't support rolling updates.

Added a troubleshooting topic that describes why jobs can be missing from Slurm accounting reports after a period of high job submission rate, how to confirm the cause in the Slurm controller log, and how to set the CommitDelay slurmdbd parameter for a cluster. For more information, see Troubleshooting missing job records in Slurm accounting.

Updated the getting started tutorial so that compute node groups mount shared storage and forward node logs with node lifecycle actions instead of launch template user data.

Added a task-oriented guide for configuring a managed Prometheus collector to scrape Slurm OpenMetrics from a AWS PCS cluster controller.

Updated the user guide for Slurm installers. Added version 25.11.7-1.

Updated the description of Amazon EC2 Capacity Blocks for ML on the Capacity Reservations page.

Removed an unsupported recommendation from the node lifecycle actions page. Added a note to the configure-cloudwatch-logs vetted script clarifying that the script preserves existing Amazon CloudWatch agent configuration.

Added constraint on QOS existence in accounting database for QOS queue settings.

Added AWS PCS agent version 1.5.0-1 in the user guide. This version adds support for node lifecycle actions.

Added a topic that describes how AWS PCS manages the Slurm configuration of a cluster with Slurm configless mode. In this mode, compute nodes and login nodes get their configuration from the cluster controller. For more information, see Slurm configuration in AWS PCS.

AWS PCS is now available in Asia Pacific (Seoul) (ap-northeast-2).

CloudFormation templates are available to get started in the Asia Pacific (Seoul) AWS Region. For more information, see Use CloudFormation to create a sample AWS PCS cluster and CloudFormation templates to create a sample AWS PCS cluster.

Updated the user guide for Slurm installers. Added versions 23.11.10-5, 24.05.8-3, 24.11.7-2, 25.05.8-1, 25.05.8-2, 25.11.6-1, and 25.11.6-2. AWS PCS marked the 24.05 versions as deprecated. For more information, see Software installers to build custom AMIs for AWS PCS.

Updated the Option 1 (rolling update) procedure to reflect that single-version AMIs are now supported. For more information, see Update the scheduler version of an AWS PCS cluster.

You can now update the scheduler version on an existing AWS PCS cluster without rebuilding your infrastructure. For more information, see Updating the scheduler version of a cluster in AWS PCS.

For the corresponding CloudFormation resource, see AWS::PCS::Cluster in the AWS CloudFormation User Guide.

AWS PCS is now available in Africa (Cape Town) (af-south-1) and Asia Pacific (Jakarta) (ap-southeast-3).

CloudFormation templates are available to get started in the Africa (Cape Town) and Asia Pacific (Jakarta) AWS Region. For more information, see Use CloudFormation to create a sample AWS PCS cluster and CloudFormation templates to create a sample AWS PCS cluster.

Added documentation for the AWS PCS-ready DLAMI Base GPU AMI, an AWS-maintained AMI for running AI/ML and HPC workloads on AWS PCS. For more information, see Using PCS-ready DLAMI with AWS PCS.

Individual compute node groups can now override the cluster-level scale-down idle time with their own scaleDownIdleTimeInSeconds setting. This feature requires Slurm version 25.11 or later.

AWS PCS is now available in Asia Pacific (Osaka) (ap-northeast-3) and South America (São Paulo) (sa-east-1).

CloudFormation templates are available to get started in the Asia Pacific (Osaka) and South America (São Paulo) AWS Regions. For more information, see Use CloudFormation to create a sample AWS PCS cluster and CloudFormation templates to create a sample AWS PCS cluster.

You can now configure compute node groups with mixed scaling, where minimum instances are greater than zero and less than maximum instances. This maintains a baseline of always-on capacity with dynamic scale-out. This feature requires Slurm 24.05 or later. For more information, see Creating a compute node group in AWS PCS.

Updated the AMI topic for AWS PCS agent 1.4.0-1. The AWS PCS agent now supports configuring and starting the slurmd daemon for any Slurm version that is compatible with the Slurm controller running on the cluster. For more information, see Software installers to build custom AMIs for AWS PCS and AWS PCS agent versions.

Updated the AMI topic for Slurm installer 25.05.7-1. For more information, see Software installers to build custom AMIs for AWS PCS and AWS PCS agent versions.

AWS PCS is now available in Europe (Spain) (eu-south-2).

CloudFormation templates are available to get started in the Europe (Spain) AWS Region. For more information, see Use CloudFormation to create a sample AWS PCS cluster and CloudFormation templates to create a sample AWS PCS cluster.

Updated the user guide for Slurm 25.11.2-1 support. Slurm 25.11 is now the default version. Sample AMIs for Slurm 25.11 are now based on Amazon Linux 2023. For more information, see the following:

AWS PCS now delivers scheduler audit logs separately through the PCS_SCHEDULER_AUDIT_LOGS log type for clusters running Slurm 25.11 or later. For more information, see Scheduler audit logs in AWS PCS.

AWS PCS now delivers slurmdbd logs (Slurm 24.11+) and slurmrestd logs (Slurm 25.05+) through the existing PCS_SCHEDULER_LOGS log type. For more information, see Scheduler logs in AWS PCS.

AWS PCS is now available in Europe (Milan) (eu-south-1).

CloudFormation templates are available to get started in the Europe (Milan) AWS Region. For more information, see Use CloudFormation to create a sample AWS PCS cluster and CloudFormation templates to create a sample AWS PCS cluster.

Updated the AMI topic for AWS PCS agent 1.3.2-1. Fixed an issue affecting RHEL 8.10 and Rocky Linux 8.10 compute node bootstrap. For more information, see Software installers to build custom AMIs for AWS PCS and AWS PCS agent versions.

AWS PCS is now available in Asia Pacific (Mumbai) (ap-south-1) and Europe (Paris) (eu-west-3).

CloudFormation templates are available to get started in the Asia Pacific (Mumbai) AWS Region and Europe (Paris) AWS Region. For more information, see Use CloudFormation to create a sample AWS PCS cluster and CloudFormation templates to create a sample AWS PCS cluster.

Slurm REST API is now supported for Slurm 25.05 or later. For more information, see Slurm REST API in AWS PCS.

Updated the AMI topic for AWS PCS Slurm installers 24.11.7-1, and 25.05.5-1. For more information, see Software installers to build custom AMIs for AWS PCS and AWS PCS agent versions.

AWS PCS now supports Slurm CLI filter plugins to run custom Lua scripts that validate and modify job submission parameters before they reach the Slurm controller. Use CLI filters to enforce custom policies, set default parameters, and provide user guidance during job submission. This feature requires Slurm version 25.05 or later. For more information, see Use Slurm CLI Filter Plugins to customize job submission in AWS PCS.

Updated the AMI topic for AWS PCS agent 1.3.1-1. For more information, see Software installers to build custom AMIs for AWS PCS and AWS PCS agent versions.

Updated the AMI topic for AWS PCS agent 1.3.0-1 and Slurm installers 24.11.6-2, 24.05.8-2, and 23.11.10-4. Updated list of supported operating systems. For more information, see Software installers to build custom AMIs for AWS PCS and AWS PCS agent versions.

Fixed some errors in the multi-cluster login node configuration script. For more information, see AWS PCS multi-cluster login node configuration script code.

AWS PCS now supports cluster secret rotation to enhance security. For more information, see Rotating cluster secrets in AWS PCS.

Updated minimum administrator permissions to support cluster secret rotation. For more information, see Minimum permissions for AWS PCS.

Added a new topic that provides a script to configure a standalone login node to connect to multiple AWS PCS clusters. The script automates the configuration of multiple Slurm sackd daemons and creates activation scripts for cluster interaction.

For more information, see Connecting a standalone login node to multiple clusters in AWS PCS.

Updated the user guide for Slurm 25.05 support. Slurm 25.05 is now the default version. For more information, see the following:

Updated the AMI topic for AWS PCS agent 1.2.2-1. For more information, see Software installers to build custom AMIs for AWS PCS and AWS PCS agent versions.

AWS PCS adds support for multiple new features:

Added troubleshooting guidance for diagnosing and resolving compute node bootstrap issues. For more information, see Troubleshoot compute node bootstrap and registration problems in AWS PCS.

AWS PCS now supports Amazon EC2 Capacity Blocks for ML, which enable you to reserve GPU-based accelerated computing instances for your clusters. For more information, see Using Amazon EC2 Capacity Blocks for ML with AWS PCS.

Minimum permissions to support Capacity Blocks are now part of the minimum permissions for a service administrator. For more information, see Minimum permissions for AWS PCS.

AWS PCS updated the AWSPCSServiceRolePolicy to support Capacity Blocks. For more information, see AWS managed policies for AWS Parallel Computing Service.

Enhanced the instance profile documentation with comprehensive CLI instructions for creating IAM roles and instance profiles. Added step-by-step procedures for setting up instance profiles using the AWS CLI and improved guidance for finding instance profiles used with AWS PCS.

For more information, see IAM instance profiles for AWS Parallel Computing Service.

Added documentation for SPANK (Slurm Plug-in Architecture for Node and job Kontrol) plugins that you can use to extend and modify Slurm's behavior during job launch and execution on AWS PCS clusters.

For more information, see Extend Slurm functionality on AWS PCS with SPANK plugins .

Added support for IPv6 networking when creating AWS PCS clusters. You can now choose IPv6 as the network type for your cluster, with corresponding updates to VPC requirements, subnet configuration, security group settings, and cluster creation procedures.

For more information, see AWS PCS VPC and subnet requirements and considerations and Creating a cluster in AWS PCS.

AWS PCS is now available in Europe (London) (eu-west-2).

CloudFormation templates are available to get started in the Europe (London) AWS Region. For more information, see Use CloudFormation to create a sample AWS PCS cluster and CloudFormation templates to create a sample AWS PCS cluster.

You can now have AWS PCS create a basic instance profile and security group for you when you create a cluster and compute node group in the console. For more information, see:

Added a new managed policy that grants permission to AWS PCS compute nodes to connect to AWS PCS clusters. For more information, see AWS managed policy: AWSPCSComputeNodePolicy .

Use job completion logs to record details about jobs when they complete, at no additional cost. For more information, see Job completion logs in AWS PCS.

AWS PCS is now available in AWS GovCloud (US-East) (us-gov-east-1) and AWS GovCloud (US-West) (us-gov-west-1).

CloudFormation templates are available to get started in the AWS GovCloud (US) Regions. For more information, see Use CloudFormation to create a sample AWS PCS cluster and CloudFormation templates to create a sample AWS PCS cluster.

For more information about the AWS PCS service endpoints in AWS GovCloud (US) Regions, see Endpoints and service quotas for AWS PCS.

For more information about differences in AWS GovCloud (US) Regions, see AWS PCS in AWS GovCloud (US) in the AWS GovCloud (US) User Guide .

Updated the AMI topic for AWS PCS agent 1.2.1-1. For more information, see Software installers to build custom AMIs for AWS PCS.

Slurm accounting is now supported for Slurm 24.11 or later. For more information, see Slurm accounting in AWS PCS.

AWS SDK: 2025-05-15

Updated the user guide for Slurm 24.11.5 support. For more information, see the following:

Updated the Slurm versions frequently asked questions (FAQ) about Slurm versions nearing or beyond end of life (EOL). For more information, see Frequently asked questions about Slurm versions in AWS PCS.

Learn how to get details for an AWS PCS compute node group, such as its ID, ARN, and AMI ID. For more information, see Get compute node group details in AWS PCS.

Updated the AMI topic for Slurm installer 24.05.7-1. For more information, see Software installers to build custom AMIs for AWS PCS.

Added internal non-adjustable quotas for the maximum number of compute node groups per cluster and the maximum number of queues per cluster. For more information, see Internal quotas.

Id is now TemplateId for the CustomLaunchTemplate property in the CloudFormation template. For more information, see Resources in Parts of a CloudFormation template for AWS PCS.

Added a new topic that describes the changes for each version of the AWS PCS agent. For more information, see AWS PCS agent versions.

Added more information to the Slurm versions topic that describes important support dates and detailed release notes for AWS PCS support for Slurm. For more information, see Slurm versions in AWS PCS.

Updated the AMI topic for AWS PCS agent 1.2.0-1. For more information, see Software installers to build custom AMIs for AWS PCS.

Added a topic to the user guide that provides an example of how to use CloudFormation with AWS PCS. The topic provides a procedure to use a sample CloudFormation template to create the sample AWS PCS cluster, and briefly describes the sections of that template. For more information, see Get started with CloudFormation and AWS PCS.

Updated the user guide for Slurm 24.05 support. For more information, see Software installers to build custom AMIs for AWS PCS and Release notes for AWS PCS sample AMIs.

Updated NVIDIA driver and CUDA versions in the Slurm 23.11 sample AMIs. For more information, see Release notes for AWS PCS sample AMIs.

Updated the AMI topic for Slurm installer 23.11.10-3. For more information, see Software installers to build custom AMIs for AWS PCS.

Updated the AMI topic for AWS PCS agent 1.1.1-1. For more information, see Software installers to build custom AMIs for AWS PCS.

Updated the AMI topic for AWS PCS agent 1.1.0-1 and Slurm installer 23.11.10-2. For more information, see Software installers to build custom AMIs for AWS PCS.

For more information, see Supported operating systems in AWS PCS.

We reorganized the user guide to bring topics to the top level, moved some topics to their own pages, and grouped similar topics together.

Updated the AMI topic for Slurm 23.11.10 and libjwt 17.0. For more information, see Software installers to build custom AMIs for AWS PCS and Step 3 – Install Slurm.

Simplified and corrected the release notes for AMIs. For more information, see Release notes for AWS PCS sample AMIs.

Added a topic that describes the KMS key policy required for encrypted EBS volumes in AWS PCS. For more information, see Required KMS key policy for use with encrypted EBS volumes in AWS PCS.

Updated AMI-related documentation to refer to AWS PCS agent version 1.0.1-1. For more information, see Software installers to build custom AMIs for AWS PCS and Step 2 – Install the AWS PCS agent.

Added a troubleshooting chapter with a topic about EC2 instances being automatically replaced after a reboot. For more information, see Troubleshooting problems in AWS Parallel Computing Service.

The ec2:DescribeInstanceTypeOfferings permission is now required for the CreateComputeNodeGroup and UpdateComputeNodeGroup API actions. For more information, see Minimum permissions for AWS PCS.

For more information, see Minimum permissions for a service administrator.

This was a correction to the documentation only. The actual managed policy wasn't changed. For more information, see AWS managed policies for AWS Parallel Computing Service.

For more information, see AWS managed policies for AWS Parallel Computing Service.