1. Proposed title of this feature request

• Eviction Filter and Eviction Policy customization support for kube-descheduler operator.

2. What is the nature and description of the request?

• We think below customization options should be supported to apply kube-descheduler on our application pod
  A. nodeFit option : We found that there's infinite pod eviction happening when we apply SoftTopologyAndDuplicates profile. If other node is ready but not have enough resource to accommodate evicted pod, this infinite pod eviction happening every descheduling interval. We expect that nodeFit option will check target node resource availability and will not evict pod if other node can't accept eviction pod.
  B. labelSelector option: Our CNF have more than 20+ statefulset / deployment with more than 100+ pods are running on same namespace. We need to only apply descheduling operator to some part of our application not all pods in same namespace.
  C. Selection of eviction policy: Openshift kube descheduler defined it's own profile which have combination of native kube descheduler eviction policy. For example, after we apply SoftTopologyAndDuplicates profile, we can see that RemoveDuplicates and RemovePodsViolatingTopologySpreadConstraint eviction policies are applied. This restriction is quite blocker to use this operator because we only need to use RemovePodsViolatingTopologySpreadConstraint policy not RemoveDuplicates policy. so we think policy must be customized to use.

3. Why does the customer need this? (List the business requirements here)

• KDDI customer is asking us to use kube-descheduler when same application pods are running on same node after node failure happening. We found that without customized filtering options, we can't control detail behavior of this function which makes can't satisfy customer request.

4. List any affected packages or components.

• Descheduler