Document canonical way to set priority for static pods

[stackbaek]

Per doc, static pods does not support priorityClassName field. However, when static pods are created with only priority values set to 2000001000 (that matches with system-node-critical) kubelet prints the following error:

forbidden: the integer value of priority (2000001000) must not be provided in pod spec; priority admission controller computed 0 from the given PriorityClass name" pod="kube-system/...

[thisisharrsh]

/language en
/sig docs

[thisisharrsh]

Page related to the issue: https://kubernetes.io/docs/concepts/scheduling-eviction/node-pressure-eviction/#pod-selection-for-kubelet-eviction

/sig node

[sftim]

What's the defect @stackbaek? Are you saying that the docs are incorrect, that there is a bug in the kubelet, or that there is a problem but you are not sure where?

[stackbaek]

The instruction to add priority to static pods is unclear from the doc. From the doc, it implies that I should just add spece.priority to the manifest (since priorityClassName was documented to be not supported). In reality, I had to add both priority and priorityClassName.

At the end of the day, I just want to confirm the canonical way to set priority on the static pods to ensure node critical pods are running under node pressure.

---

NOTE: I have not tried to set just priorityClassName on the static pods.

[sftim]

/retitle Document canonical way to set priority for static pods

[sftim]

/kind bug

/priority backlog
/triage accepted

[thisisharrsh]

The necessary information needs to be added, and the documentation should be corrected. I'd want to hear from more people too.

[ktvargo-ms]

/assign

[ktvargo-ms]

@SergeyKanzhelev I was able to create a static pod with priorityClassName and was planning on adding the example to the document. Let me know if anything else is needed or if that isn't expected behavior. I see you have this added for triage with Node bugs.

k get priorityclass
NAME VALUE GLOBAL-DEFAULT AGE
system-cluster-critical 2000000000 false 5d2h
system-node-critical 2000001000 false 5d2h

cat static-web.yaml
apiVersion: v1
kind: Pod
metadata:
name: static-web
labels:
role: myrole
spec:
containers:
- name: web
image: nginx
ports:
- name: web
containerPort: 80
protocol: TCP
priorityClassName: system-cluster-critical

k describe pod static-web-minikube

Name: static-web-minikube
Namespace: default
Priority: 2000000000
Priority Class Name: system-cluster-critical
Node: minikube/192.168.49.2
Start Time: Tue, 25 Jun 2024 15:18:25 -0400
Labels: role=myrole
Annotations: kubernetes.io/config.hash: d13d717ea915d4c4049f7783bb26358a
kubernetes.io/config.mirror: d13d717ea915d4c4049f7783bb26358a
kubernetes.io/config.seen: 2024-06-25T19:18:25.377885764Z
kubernetes.io/config.source: file
Status: Running

[ktvargo-ms]

@stackbaek and @sftim, I was able to confirm that using just priorityClassName on the static pods works (see example above). Since this is a concept page, we do want to add an example or not? Otherwise, it is just a quick update on the docs to use priorityClassName on the static pods as well.

[stackbaek]

@ktvargo-ms great! examples always boosts clarity and readability (at least for me).

So, I guess we don't even need the priority number, correct? Otherwise all SGTM!

[sftim]

@stackbaek and @sftim, I was able to confirm that using just priorityClassName on the static pods works (see example above). Since this is a concept page, we do want to add an example or not? Otherwise, it is just a quick update on the docs to use priorityClassName on the static pods as well.

Interesting (and to me surprising); however, let's show that this works for an arbitrary Pod priority and not just a built-in one. We should also show that it'll work OK for a static Pod that starts before the control plane is healthy.

[ktvargo-ms]

The original issue (improper instruction) has been resolved. Would these items be better addressed in the static pod page?

[ktvargo-ms]

@ktvargo-ms great! examples always boosts clarity and readability (at least for me).

So, I guess we don't even need the priority number, correct? Otherwise all SGTM!

Correct. Only the priorityClassName is required. The priority number is displayed when describing the node.

[sftim]

Only the priorityClassName is required. The priority number is displayed when describing the node.

I don't get how this works if the control plane isn't up yet. That's an important consideration.

[ktvargo-ms]

Created 'xyz-priority' and will use that an example

k get priorityclass
NAME VALUE GLOBAL-DEFAULT AGE
system-cluster-critical 2000000000 false 2m2s
system-node-critical 2000001000 false 2m2s
xyz-priority 1000000 false 47s

apiVersion: v1
kind: Pod
metadata:
name: static-web
labels:
role: myrole
spec:
containers:
- name: web
image: nginx
ports:
- name: web
containerPort: 80
protocol: TCP
priorityClassName: xyz-priority

k describe pod static-web-minikube
Name: static-web-minikube
Namespace: default
Priority: 1000000
Priority Class Name: xyz-priority
Node: minikube/192.168.49.2
Start Time: Tue, 02 Jul 2024 15:56:35 -0400
Labels: role=myrole
Annotations: kubernetes.io/config.hash: a8236a64037ee18291f146e8a8013147
kubernetes.io/config.mirror: a8236a64037ee18291f146e8a8013147
kubernetes.io/config.seen: 2024-07-02T19:56:35.763567682Z
kubernetes.io/config.source: file
Status: Running