top

LimitPodHardAntiAffinityTopology admission controller rejects any pod that specifies "hard" (RequiredDuringScheduling) anti-affinity with a TopologyKey other than unversioned.LabelHostname. Because anti-affinity is symmetric, without this admission controller, a user could maliciously or accidentally specify that their pod (once it has scheduled) should block other pods from scheduling into the same zone or some other large topology, essentially DoSing the cluster. In the future we will address this problem more fully by using quota and priority, but for now this admission controller provides a simple protection, on the assumption that the only legitimate use of hard pod anti-affinity is to exclude other pods from the same node.

Imports 5 package(s) ΒΆ

  1. k8s.io/kubernetes/pkg/client/clientset_generated/internalclientset
  2. k8s.io/kubernetes/pkg/admission
  3. k8s.io/kubernetes/pkg/api
  4. k8s.io/kubernetes/pkg/api/errors
  5. k8s.io/kubernetes/pkg/api/unversioned