AWS Neuron Kernel Modules on OpenShift with KMM

2025-07-24T00:00:00+02:00

As part of my work at AWS, I've been diving deep into running AI/ML workloads with the AWS Neuron SDK on Inferentia and Trainium instances on Red Hat OpenShift Services on AWS (ROSA). What started as exploring concepts quickly turned into building actual tooling to solve real problems.

The challenge: how do you get AWS Neuron drivers working reliably across different OpenShift versions? The solution turned out to be more interesting than I initially thought.

I developed and published kmod-with-kmm-for-ai-chips-on-aws, a project that automates building Kernel Module Management (KMM) compatible container images for AWS Neuron drivers.

Running specialized hardware like AWS Inferentia and Trainium chips requires kernel modules. Kernel modules are tied to specific kernel versions, and new driver releases need to be built against different kernel versions. This creates ongoing maintenance work that grows with each supported platform version.

Red Hat's Kernel Module Management (KMM) operator solves this by automating kernel module lifecycle management. But you still need to build the container images that contain your modules for each kernel version.

That's where the Driver Toolkit comes in - it provides the exact kernel headers and build environment for each OpenShift release.

I fully automated this using GitHub Actions with nightly scans for new OpenShift releases, automated builds triggered by driver or mapping changes, and minimal container images containing only busybox and the compiled kernel module.

GitHub Actions proved surprisingly powerful for this automation. The dual tagging strategy works well - images get both kernel-specific and OpenShift-specific tags for different use cases. The build script also works locally for development, automatically detecting the environment and adapting accordingly.

This removes the manual work of tracking releases and building modules for each kernel version. The process now happens automatically. The approach is reusable for other specialized hardware drivers on OpenShift.

If you're running specialized hardware on OpenShift, KMM and the Driver Toolkit are powerful tools once you understand how to use them effectively.

Red Hat Certified Specialist in Red Hat OpenShift Service on AWS (ROSA)

2024-09-05T00:00:00+02:00

It was time to re-new and keep my RHCSA, and RHCE current. I was curious about the recently released RHCS in ROSA Exam. I passed the EX229 today and earned the Red Hat Certified Specialist in Red Hat OpenShift Service on AWS (ROSA) (Archive: [1], [2]) Certification!

It was kind of a logical choice to take this one. Fits pretty well into my current Role, tech focus and complements previously gained skills.

It wasn't my first Red Hat Exam but I can tell, it was one of the most challenging one. I used the Red Hat Courses DO120, DO280, CS220 and CS221 to prepare. AWS and Kubernetes Hands-On experience helps a lot. Also the muscle memory from last years CKAD, CKA and CKS marathon. I don't think I would've passed otherwise.

You have 3 hours to finish the exam, which sounds long but the time is just flying by. You need to be fast and 100% focused. Makes me even more proud that I aced it :D

Summary: A Red Hat Certified Specialist in ROSA is able to configure and manage Red Hat OpenShift clusters running on Amazon Web Services (AWS). The credential is earned after successfully passing the Red Hat Certified Specialist in ROSA (EX229).

Source & Copyright: https://www.credly.com

Skills

Amazon CloudWatch
Configure AWS
OpenShift Cluster on AWS
PrivateLink ROSA clusters
ROSA Clusters

Source & Copyright: https://www.credly.com

Certificate

Downloads
- Certificate (2024-09-05) (PDF, 361K)
- Badge (PNG, 55K)
Links
- Credly Badge (Archive: [1], [2])
- Verify Certificate

The Wombelix Post - OpenShift

AWS Neuron Kernel Modules on OpenShift with KMM

Red Hat Certified Specialist in Red Hat OpenShift Service on AWS (ROSA)

Skills

Certificate