[Help] Managed Windows Node Groups creating Linux nodes?

[dan-tw]

What help do you need?

I'm currently creating a new cluster to test out Managed Windows Node Groups by doing the following:

# cluster.yaml
# An example of ClusterConfig containing Windows and Linux node groups to support Windows workloads
---
apiVersion: eksctl.io/v1alpha5
kind: ClusterConfig

metadata:
  name: windows-cluster
  region: ap-southeast-2

managedNodeGroups:
  - name: windows-managed-ng
    amiFamily: WindowsServer2022FullContainer
    instanceType: g4dn.2xlarge
    minSize: 1
    maxSize: 1

Using the following version of eksctl: 0.125.0

The result is a cluster with a single Managed Node group that is running an Amazon Linux 2 EKS node (AMI ID: ami-0760f8e90a1753cfa) running kubernetes 1.24

I would've thought that the above config would create a Windows Managed Node Group?

eksctl output also states that it is enabling Windows IP addressing, so it must've registered the AMI Family:

2023-01-16 15:52:26 [ℹ]  
2 sequential tasks: { create cluster control plane "windows-cluster", 
    2 sequential sub-tasks: { 
        2 sequential sub-tasks: { 
            wait for control plane to become ready,
            enable Windows IP address management,
        },
        create managed nodegroup "windows-managed-ng",
    } 
}

Any idea what's gone wrong?

[github-actions]

Hello dan-tw 👋 Thank you for opening an issue in eksctl project. The team will review the issue and aim to respond within 1-5 business days. Meanwhile, please read about the Contribution and Code of Conduct guidelines here. You can find out more information about eksctl on our website

[yuxiang-zhang]

It looks like because of

[!]  WindowsServer2022FullContainer does not ship with NVIDIA GPU drivers installed, hence won't support running GPU-accelerated workloads out of the box

The WindowsServer2022FullContainer AMI Family using GPU instances is not mapped to any corresponding AMI Type.
https://github.com/weaveworks/eksctl/blob/e13718238bc9fc394a701f1095868bb927d3fda4/pkg/cfn/builder/managed_nodegroup.go#L295-L297

[yuxiang-zhang]

Creating GPU instances with Windows AMI is not allowed, see #4238.

[dan-tw]

That explains it.

Perhaps eksctl should fail earlier, rather than spin up a cluster that does not actually do what is being requested?

For Windows, we are actually using our own custom AMI that does have GPU drivers installed and does advertise available GPU's to containers. I was mostly interested in seeing how the "Managed Windows Node Group" implementation worked for eksctl

[github-actions]

This issue is stale because it has been open 30 days with no activity. Remove stale label or comment or this will be closed in 5 days.

[github-actions]

This issue was closed because it has been stalled for 5 days with no activity.

[TiberiuGC]

Related to #7645.

Perhaps eksctl should fail earlier, rather than spin up a cluster that does not actually do what is being requested?

@dan-tw - I agree with you on this. I shall work on a fix for both issues.