intel gpu device plugin show allocatable resource error

[zhm-algo]

there are 4 Arc Pro B60 gpu in the system, and I allocate 2 B60 for my pod

but the allocable resource of “gpu.intel.com/xe” show no change

[tkatila]

Thanks for the report @zhm-algo! Which k8s version are you running?

I tried it quickly with 1.35 k3s cluster and there it indeed does not change the allocatable count. Weirdly, if I create another Pod that tries to allocate the same resources, it goes to Pending. So the resources seem to be correctly calculated, but the node's state is not showing correctly. From the plugin point of view, we haven't changed the logic afaik.

edit: NPU plugin is experiencing the same issue as well as the CPU resource count. I think, at least in my case, the issue is within the k8s components.

[zhm-algo]

we will confuse if the node state not showing correctly.
is it possible to show correctly ?
thanks

[tkatila]

The plugin itself doesn't update the counts, it's done by the cluster. i.e. kubelet and/or scheduler. For me, the node resources were "stuck", meaning also the cpu count didn't update.

Which k8s version and variant are you using?

[zhm-algo]

(base) root@node1:~# kubectl version
Client Version: v1.28.8
Kustomize Version: v5.0.4-0.20230601165947-6ce0bf390ce3
Server Version: v1.28.8