kubernetes: Scheduler ignores requests for unknown fully-qualified resources.

/kind bug /sig scheduling

What happened:

Scenario A:

• Create a pod that requires one example.com/foo resource.
• Observe that the pod is scheduled, despite there being no matching resources available in the cluster.

Scenario B:

• Create a pod that requires one kubernetes.io/foo resource.
• Observe that the pod is scheduled, despite there being no matching resources available in the cluster.

What you expected to happen:

I did not expect the pod to be scheduled and run. In both cases, I expected the pod to do one of the following:

• Remain in pending state, having failed a scheduler predicate for all available nodes.
• Fail validation. Since there is no facility in the scheduler or kubelet for checking feasibility of these resources, you could argue that they should have failed admission.

How to reproduce it (as minimally and precisely as possible):

See above.

Anything else we need to know?:

There is a pull request in progress that attempts to fix this problem for all resources outside of the default namespace: PR 48922: . Another patch will be needed to handle fully-qualified resources in the kubernetes.io domain.

Environment:

• Kubernetes version (use kubectl version): v1.7, recent master
• Cloud provider or hardware configuration**: Local cluster
• OS (e.g. from /etc/os-release): Linux Mint 18.1, CentOS 7
• Kernel (e.g. uname -a): 4.4, 3.10
• Install tools: ./hack/local-up-cluster.sh