How to modify the start task of an Azure Batch Pool started from HPC Pack 2019

Christopher Quenneville 0 Reputation points
2024-08-03T23:35:26.7966667+00:00

I have an On-Premises HPC Pack 2019 Update 2 head node and Windows workstation nodes environment from which my organisation would like to burst to Linux VMs on Azure Batch.

I have followed correctly this documentation: https://video2.skills-academy.com/en-us/powershell/high-performance-computing/burst-to-azure-batch-with-microsoft-hpc-pack

I am able to start pools but the start task fails every time since HpcBatchAgent.sh included in the automatic deployment fails to install the .NET dependencies to run HpcSync. The script looks for OS releases names and installs the .NET dependencies if it sees "centos" and others. We have to use almalinux now since centos is discontinued, and the script doesn't have "almalinux" listed as an os release to look for and yum install the requisites.

I tried mannually modifying the script to append |"almalinux", and it works. The issue is that pools are created and destroyed from HPC Pack, and we can't establish a workflow in which we would have to make that change mannually everytime.

This issue was fixed for the On-Premises side on HPC Pack 2019 Update 2, now including rockylinux and almalinux. However, it seems that the Azure Batch scripts weren't updated. It seems old, listing ubuntu's latest version as 16 (not 24). I was on the original HPC Pack 2019 and I correctly updated to update 1 and 2, so maybe my issue is there?

Does anyone know how to modify the start task of an Azure Batch Pool started from HPC Pack 2019?

That start task downloads HpcBatchAgent.sh from the web using this link: (https://go.microsoft.com/fwlink/?linkid=852446) (https://download.microsoft.com/download/D/5/6/D56F1BBB-0103-4151-82DC-B54751D447A6/HpcBatchAgent.sh) and if I could just swap that url for my modified version of the script we would be alright.

Thanks a lot in advance.

Azure Batch
Azure Batch
An Azure service that provides cloud-scale job scheduling and compute management.
337 questions
{count} votes

Your answer

Answers can be marked as Accepted Answers by the question author, which helps users to know the answer solved the author's problem.