Support Node Feature Discovery

[yankay]

What type of PR is this?
/kind feature

What this PR does / why we need it:

Node Feature Discovery is a Kubernetes add-on for detecting hardware features and system configuration.

It's widely used and is required by the GPU Operator.
And it's help to AI-optimized k8s cluster, ref #10622.

Which issue(s) this PR fixes:

Fixes #10854

Special notes for your reviewer:

Does this PR introduce a user-facing change?:

Support Node Feature Discovery

.

[yankay]

HI @mzaian, @VannTen
Would you please help to review the PR :-)

[VannTen]

Should we really add more components to Kubespray ? One the thing mentionned in #6400 was about reducing Kubespray scope and maintenance load.

NFD has an helm chart right ?

[yankay]

Should we really add more components to Kubespray ? One the thing mentionned in #6400 was about reducing Kubespray scope and maintenance load.

NFD has an helm chart right ?

Thanks @VannTen

I think there are 3 topics.

---

1 Does the Kubespray should upgrade the GPU support?"

GPU Support has been a feature of kubespray since #3304.
The PR is part of "GPU Operator" because it is a more popular and easier-to-use method for supporting GPUs. After the GPU Operator is supported, the old method of using GPUs should be deprecated.

So, The kubespray should upgrade the GPU Support with GPU Operator.

---

2 "Do the addons of kubespray need to be migrated to helm chats?"

There are great discussions at #6400. Maintenance of the addon's jinjia2 template is not reasonable work.

The Helm way of installing add-ons is great, like https://docs.k3s.io/helm. Using the helm to install add-ons would simplify this project and the maintenance burden.

In the long term, I want to choose the "Helm style" instead of the "Jinjia2 style. " If it works, I think some add-ons should be migrated to it.
PS. Customs CNI, Kubelet csr approver is "Helm style"

---

3 "Do the NFD add-ons need to be helm chats?"

Because there is no shared view of "Helm style." The PR uses the old way like the other add-ons, or the "helm style" is OK.

What do you think about that @VannTen ?

[VannTen]

I think there is some addons using helm currently (kubelet_csr_approver ?) optionally. However the helm_apps role has a bad interface (I wrote it T_T) and I plan to rewrite it in the mid-term (if I can pull that off this month that would be great). We should not block other things ofc. OTOH, for kubespray users, packaging NFD or GPU operator is mostly convenience, isn't it ? I mean, is the setup more involved than installing the helm chart ?

[yankay]

I think there is some addons using helm currently (kubelet_csr_approver ?) optionally. However the helm_apps role has a bad interface (I wrote it T_T) and I plan to rewrite it in the mid-term (if I can pull that off this month that would be great). We should not block other things ofc. OTOH, for kubespray users, packaging NFD or GPU operator is mostly convenience, isn't it ? I mean, is the setup more involved than installing the helm chart ?

Thanks @VannTen

Looking forward to the new helm role very much. 💕💕💕

Q: “Is the setup more involved than installing the helm chart ?”
A: In my opinion, the add-on (helm style or jinjia2 style) in Kubespray has some benefits over "helm charts without Kubespray. "

1. It's integrated and tested.
2. It's all based on Ansible; the users can use Ansible var xx_enabled to enable an addon.
3. The kubespray can do some prerequisite checks for the addon, but the helm chart cannot. So it makes it easy. like https://docs.nvidia.com/datacenter/cloud-native/gpu-operator/latest/getting-started.html#prerequisites.
4. Some var is important for many components, such as kube_proxy_strict_arp. It's useful for kube-proxy, kube-vip, and metallb. If the user installs kube-vip without kubespray, He will take care of it himself.
5. In an air-gap environment, Kubespray declares the image. The helm does not do it.

So, if the add-on has many system relationships, like CNI/GPU, Kubespray can make it easy to install it. But if the addon does not have system requirements, like a Java web application, there is no need to use Kubespray to install it.

[mzaian]

Thanks @yankay

/lgtm

[k8s-ci-robot]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: mzaian, yankay

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Needs approval from an approver in each of these files:

• OWNERS [mzaian,yankay]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment