Infra Request - OpenShift AI for openSUSE

[skriesch]

Project Name

openSUSE

Code Repository URL

https://github.com/opensuse

Project License

GNU General Public License version 2

Names of primary maintainers/project leads

Sarah Julia Kriesch: sarah.kriesch@opensuse.org
Fabian Vogt: fvogt@suse.com
Holger Rauch: hrauch@opensuse.org

Architectures/operating systems currently supported

openSUSE Tumbleweed/openSUSE Leap
s390x
x86
arm

Why are you interested in supporting the mainframe?

We want to continue with supporting the mainframe with the goal to make Linux on Z accessible for all without the requirement of Enterprise support. That includes also the support of AI, which requires updated Z hardware. Red Hat has approved the support of openSUSE on OpenShift with OpenShift Virtualization and OpenShift AI. Such an environment is required for a collaborative support and for debugging issues.

Which operating system(s) will you need access to?

Linux on s390x

Resources requested outside of the standard resources provided.

OpenShift cluster with OpenShift Virtualization and OpenShift AI:

1. OpenShift Virt requires 1 Control Plane and 1 Worker node directly on an LPAR -> 2 LPARs
2. The rest of the OpenShift cluster can go on 1 LPAR: 2 Control Planes and 2 Worker Nodes on KVM or z/VM
   We need access to the OpenShift Virt and AI interface to debug openSUSE on OpenShift for a collaborative support of both.
   It should have got enough resources for setting up as a minimum 1 VM via OpenShift Virt and running AI agents with openSUSE containers and openSUSE VMs via API requests to OpenShift AI.

How long will the infrastructure be needed?

1 year or less

Any other comments or questions

I have received architectures with system requirements for setting up OpenShift AI. You can reach out to me, that I can send it to you.

[jmertic]

Thank you @skriesch - we will add to the list and keep you posted.

[jmertic]

TAC voting in process via LFX Voting

[jmertic]

Hi @skriesch - the TAC has done further review, and there is a question of if this work is an open-source project; can you confirm? Could you also point to the specific repo for the code; the link given just points to the OpenSUSE GH org. Thanks!

[skriesch]

openSUSE is an open source project (community Linux distribution) and part of the Linux Distributions Working Group.
We include all, what is required in a Linux distribution. Therefore, you can use all gnu projects, Linux kernel and everything else, incl. their licenses.
Our Github org is this one: https://github.com/opensuse
There, we have got also own projects like snapper (for operating system snapshots), zypper (installer) and more (all what has been developed by us as an open source project on top and what is openSUSE specific).
If one single project is required, use our Linux kernel, which is the main foundation of a Linux distribution: https://github.com/openSUSE/kernel
If you need our project resources for building VMs, used for this case in OpenShift Virtualization, that is the correct link: https://github.com/openSUSE/obs-build/blob/master/build-vm-kvm
Our license for the operating system is in our wiki: https://en.opensuse.org/openSUSE:License

[jmertic]

Thanks @skriesch - I guess my question is for specifically 'OpenShift AI', which is the project the resources were requested for. Can you point me to the code repo for those compenents? Thanks!

[skriesch]

OpenShift AI is the single open source AI (integrated in Red Hat OpenShift), which is supported by IBM for Linux on Z.
In general, that is a Red Hat product on top of Red Hat OpenShift: https://www.redhat.com/en/products/ai/openshift-ai
Following open source projects are integrated:

1. llm-d: https://github.com/llm-d/llm-d (license Apache 2.0)
2. vllm: https://github.com/vllm-project/vllm (license Apache 2.0)
3. Jupyter notebook: https://github.com/jupyter/notebook (license BSD-3-Clause)
4. langchain: https://github.com/langchain-ai/langchain (license MIT)
5. langgraph: https://github.com/langchain-ai/langgraph (license MIT)
6. KServe: https://github.com/kserve/kserve (license Apache 2.0)
7. MLflow: https://github.com/mlflow/mlflow (license Apache 2.0)
   Following projects will need also access in the future (because part of OpenShift AI):
8. PyTorch: https://github.com/pytorch/pytorch (license BSD-style)
9. tensorflow: https://github.com/tensorflow/tensorflow (license Apache 2.0)
10. Pandas: https://github.com/pandas-dev/pandas (license BSD-3-Clause)

IBM does not support any other Kubernetes platform any more. Therefore, Red Hat wants to open the collaboration to community Linux distributions with community support on top of OpenShift. The Red Hat Product Manager for OpenShift AI said that I should say it at the Linux Distributions Working Group, that we need OpenShift clusters for testing OpenShift AI and fixing bugs on Linux distribution level. But also most Python projects will also require such an access.
A shared environment would be beneficial.

[jmertic]

Thanks @skriesch - so what would be developed/tested here? Is it these 10 projects only?

[skriesch]

We as „openSUSE“ have to test and verify the whole functionality of an operating system in collaboration (and in the future also integrating) AI. As a first step, we want to use the offer by Red Hat, that we can „use“ OpenShift AI. The background is that IBM has been contributing only to llvm upstream their hardware enablement for the Spyre Accelerator with the target „OpenShift AI“. All the IBM Python extensions are not upstream. They are provided only as „commercial“ plugins/extensions in IBM containers. Linux distributions don’t support that, and we can not guarantee the functionality with our operating system dependencies for that. Therefore, I was happy after FrOSCon 2025 with the offer, that we should be a supported operating system in OpenShift for using OpenShift AI.

Our responsibilities:

1. openSUSE as an operating system in OpenShift Virtualization and as containers
   The foundation, that we can use OpenShift AI is, that we can provide openSUSE as VMs and containers in OpenShift. OpenSUSE is already supported for OpenShift Virtualization on s390x: https://access.redhat.com/articles/4234591
   We are in a critical situation (already reported in April 2026), because we can not verify reported bugs without any kubevirt in kubernetes on s390x. KIND (Kubernetes in Docker) is not provided for s390x any more, and my own running version in the LinuxONE Community Cloud has been too old for kubevirt. The Single Node Installations of OpenShift (already available in the LinuxONE Community Cloud) are not made for kubevirt installations. They are crashing because of not enough resources.

Support process:

• IBM is converting our qemu images to kubevirt images
• openSUSE virtualization images based on that will be tested only upstream (based on OpenShift Virtualization) at the moment!
• If there is a bug in openSUSE, that will be reported to openSUSE.
• IBM, SUSE and openSUSE are collaborating on the solution.
• openSUSE has to be able to reproduce all bugs (related to problems within openSUSE in combination with AI and Virtualization) in OpenShift on s390x!
  Example bug, what was difficult to solve without this environment (SSH problem on operating system level):
  https://bugzilla.opensuse.org/show_bug.cgi?id=1259587
  IBM has to provide the required licenses for that. That was already a topic in September 2025 during our meeting (afer the offer at FrOSCon).
  @pleia2 Can you organize the required OpenShift licenses (incl. OpenShift AI), please?
  -> I will give a talk about the collaborative support process at the IBM TechXChange.

2. OpenShift AI API functionality
   Red Hat is providing OpenShift AI within OpenShift (as ready to go). The Spyre Cards with Accelerators (min. 4 of them) will be accessed via the referenced stack above (connection via vllm).
   openSUSE can use this technology via API and the provided API Gateway + KServe for scaling. So you can scale the vllm functionality to openSUSE VMs and openSUSE containers.
   Our „community“ responsibilities then:

• Responding on bugreports by OpenShift users using our openSUSE VMs/containers
• Reproducing and fixing bugs related to that in openSUSE and upstream (mainly Python topics for AI/ML)
• Testing Python dependencies and API functionalities (working closed with upstream projects, Red Hat and IBM)
• Doing tests with new open source AI/ML products in combination with Telum processors and IBM Spyre Accelerator
• Working with the Python community on enabling their open source projects for Telum processors and IBM Spyre Cards → IBM does not contribute these functionalities until now.