Why are tasks in the bound state not evicted during the reclaim action?

[wdxylb]

volcano/pkg/scheduler/actions/reclaim/reclaim.go

Line 189 in de47ad5

if taskOnNode.Status != api.Running || !taskOnNode.Preemptable {

Some Pods have successfully bound an NFS PersistentVolume (PVC status is Bound), yet because the NFS server is currently unreachable, they remain stuck in Pending and never transition to Running.

[JesseStutler]

Which version of volcano did you use? And what is your workloads template? More context needs to be provided, thanks

[wdxylb]

In Volcano 1.3.5, the reclaim action only selects Running tasks that are also marked Preemptable as victims. When the NFS server becomes unreachable, a task that uses a PVC (backed by an NFS PV) should normally be Running, yet in practice it remains in the Bound state.
Another task—residing in the reclaim-others-resource-pool Queue, which has the annotation volcano.sh/enable-reclaim-others: "true"—needs the same resources and therefore triggers the reclaim action.
However, the reclamation logic only considers Running and Preemptable victims; since the Bound task fails the check taskOnNode.Status != api.Running, it is skipped.
Consequently, the resources remain logically occupied while physically unused, producing the “cannot be reclaimed” phenomenon.

apiVersion: v1
kind: PersistentVolume
metadata:
  name: nfs-pv-dummy
spec:
  capacity:
    storage: 1Gi
  accessModes:
    - ReadWriteMany
  persistentVolumeReclaimPolicy: Retain
  nfs:
    server: 192.0.2.99   # unreachable
    path: /exports

---
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: nfs-pvc-dummy
spec:
  accessModes:
    - ReadWriteMany
  resources:
    requests:
      storage: 1Gi
  volumeName: nfs-pv-dummy  

---
apiVersion: batch.volcano.sh/v1alpha1
kind: Job
metadata:
  name: nfs-pending-job
  annotations:
    volcano.sh/preemptable: "true"
spec:
  minAvailable: 1
  schedulerName: volcano
  queue:  reclaimable-resource-pool
  tasks:
    - replicas: 1
      name: task
      template:
        spec:
          containers:
          - name: main
            image: busybox
            command: ["sleep", "3600"]
            volumeMounts:
            - name: vol
              mountPath: /data
          volumes:
          - name: vol
            persistentVolumeClaim:
              claimName: nfs-pvc-dummy
          restartPolicy: Never

---
apiVersion: scheduling.volcano.sh/v1beta1
kind: Queue
metadata:
  name: reclaimable-resource-pool
spec:
  affinity:
    nodeGroupAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
      - minikube
  priority: 0
  reclaimable: true

---
apiVersion: scheduling.volcano.sh/v1beta1
kind: Queue
metadata:
  name: reclaim-others-resource-pool
  annotations:
    volcano.sh/enable-reclaim-others: "true"
spec:
  affinity:
    nodeGroupAffinity:
      requiredDuringSchedulingIgnoredDuringExecution:
      - minikube
  priority: 100
  reclaimable: false