aws-samples/aws-hpc-recipes

Contains example recipes that demonstrate how to build HPC systems using AWS services and solutions.

HPC Recipes for AWS

This repository contains example recipes that demonstrate how to build HPC systems using AWS ParallelCluster, Research and Engineering Studio on AWS, and other AWS products.

Getting Started

• If you are new to AWS ParallelCluster, you can watch a 5-minute explainer video.
• If you are new to Research and Engineering Studio on AWS, you can watch a 5-minute explainer video.
• You can launch an HPC cluster in the cloud with just a few clicks.
• Do you need help setting up HPC-ready networking? Choose from a simple example or a more advanced configuration.
• There are also examples of setting up HPC-ready filesystems on AWS for you to try and learn from.

To simply explore what's here, the recipes home page shows you every recipe. Here is an example of what you'll find:

This is a recipe named slurm_accounting_db. It's part of the db (database collection). The tags tell us its a core recipe (i.e. it's maintained by AWS staff) and that it pertains to Amazon RDS and AWS ParallelCluster.

Security

See CONTRIBUTING for more information.

Contributing

We encourage your to contributions to this collection. Read up on our contribution process and guidelines. Then, head to Get Started to begin.

License

This repository is licensed under the MIT-0 License. See the LICENSE file.

aws-hpc-recipes's Issues

Managed AD Linux AMI version issues

When the template is executed in a pipeline some time after initial deployment the Cloudformation changeset identifies that the AMI for the AdDomainAdminNode needs to be updated, because there's a newer version of the amznlinux image. An AMI change requires the server resource be destroyed and rebuilt, which is attempted.

The new node is launched, but experiences an error during the startup script. It's trying to create the Service Account user via adcli , but this returns an error because the user account already exists.

To workaround the issue we pass the AMI image version as a parameter with the type of 'AWS::EC2::Image::Id'.

[Bug] Recipe `hpc_large_scale` has a deprecated lambda function runtime

As of January 25th 2024, the Python 3.7 lambda function runtime has been deprecated by AWS. It is no longer maintained and functions cannot be created using it.

The hpc_large_scale recipe uses a custom resource based on a lambda function and the runtime configure is Python 3.7. The stack fails to deploy with the message that runtime chosen for the custom resource is invalid. Changing this parameter to Python 3.9 fixes the problem and does not introduce any incompatibility.

How to reproduce

1. Try to deploy the hpc_large_scale recipe
2. The CloudFormation stack will fail deployment with the message that the runtime is deprecated.

Expected behaviour
Successfully deploy the stack

security/public_certs CertificateNode has overly broad permissions

The InstanceRole for the CertificateNode in security/public_certs has administrative permissions:

aws-hpc-recipes/recipes/security/public_certs/assets/main.yaml

Lines 100 to 105 in 8410cf4

	- PolicyDocument:
	Statement:
	- Action: '*'
	Effect: Allow
	Resource: '*'
	PolicyName: All

Presumably left-over from development?

res-demo-stack.yaml deployment fails due to Node v14 Lambda runtime dependency

recipes/res/res_demo_env/assets/res-demo-stack.yaml line 135 includes https://research-engineering-studio-us-east-1.s3.amazonaws.com/releases/2024.01/ResearchAndEngineeringStudio.template.json which on line 3500 in the definition of Lambda function AWSCDKCfnUtilsProviderCustomResourceProviderHandlerCF82AA57 includes "Runtime": "nodejs14.x". NodeJS v14 is deprecated as a Lambda runtime and this causes this stack deployment to fail.

pcluster/slurm_accounting receipe fails due to S3 bucket permissions

The pcluster/slurm_accounting recipe relies upon the ability to access the CloudFormation template provided at:

https://aws-hpc-recipes.s3.${AWS::Region}.${AWS::URLSuffix}/${Version}/recipes/db/slurm_accounting_db/assets/serverless-database.yaml

When deploying the recipes/pcluster/slurm_accounting/assets/launch.yaml, the stack fails due to being unable to access this S3 bucket.