What happened?

I created a resource claim template to get "All" GPUs on a node:

apiVersion: resource.k8s.io/v1beta1
kind: ResourceClaimTemplate
metadata:
  name: all-gpus
spec:
  spec:
    devices:
      requests:
      - name: gpu
        deviceClassName: gpu.nvidia.com
        allocationMode: All

I then created a deployment that had a Pod that used that claim. The Pod was scheduled to a node. However, my DRA driver on that node was not running, so there were no resource slices for that node.

What did you expect to happen?

I expected the pod to not schedule, since there were no available devices meeting the request. "All" should mean "at least one".

How can we reproduce it (as minimally and precisely as possible)?

Create the resource claim template as shown and a deployment, with no DRA driver running. The pod will still schedule.

Anything else we need to know?

/wg device-management

Kubernetes version

$ kubectl version
Client Version: v1.32.0
Kustomize Version: v5.5.0
Server Version: v1.32.0-gke.1358000

Cloud provider

OS version

$ cat /etc/os-release
PRETTY_NAME="Debian GNU/Linux rodete"
NAME="Debian GNU/Linux rodete"
VERSION_CODENAME=rodete
ID=debian
HOME_URL="https://go/glinux"
SUPPORT_URL="https://go/techstop"
BUG_REPORT_URL="https://go/techstop"
$ uname -a
Linux jbelamaric.c.googlers.com 6.10.11-1rodete2-amd64 #1 SMP PREEMPT_DYNAMIC Debian 6.10.11-1rodete2 (2024-10-16) x86_64 GNU/Linux

Install tools

Container runtime (CRI) and version (if applicable)

Related plugins (CNI, CSI, ...) and versions (if applicable)