reclaim action's evict can not be canceled

[lowang-bh]

Description

Reclaim action use the ssn.Evict, which directly evict pod and can not be caceled when eviction is not helpful.

Steps to reproduce the issue

cluster with cpu = 3 and scheduler configmap is

apiVersion: v1
data:
  volcano-scheduler.conf: |
    actions: "enqueue, allocate, reclaim"
    tiers:
    - plugins:
      - name: priority
      - name: gang
        enablePreemptable: false
      - name: conformance
    - plugins:
      - name: overcommit
      - name: drf
        enablePreemptable: false
      - name: predicates
      - name: proportion
      - name: nodeorder
      - name: binpack
kind: ConfigMap

1. kubectl apply -f job-a.yaml

   kind: Job
   metadata:
   name: job-a
   spec:
   backoffLimit: 3
   completions: 3
   parallelism: 3
   template:
   metadata:
     annotations:
       scheduling.k8s.io/group-name: job-a-pg
       volcano.sh/preemptable: "true"
   spec:
     containers:
     - image: nginx:1.14.2
       imagePullPolicy: IfNotPresent
       name: nginx
       ports:
         - containerPort: 80
       resources:
         requests:
           cpu: 1000m
           memory: 200Mi
         limits:
           cpu: 1000m
           memory: 200Mi
     restartPolicy: Never
     terminationGracePeriodSeconds: 1
     schedulerName: volcano
   ---
   apiVersion: scheduling.volcano.sh/v1beta1
   kind: PodGroup
   metadata:
   annotations:
   scheduling.k8s.io/reclaimable: "true"
   name: job-a-pg
   namespace: default
   spec:
   minMember: 1
   queue: queue-a

2. kubectl apply -f job-a.yaml ➜ reclaim git:(master) ✗ cat job-b.yaml

   apiVersion: batch/v1
   kind: Job
   metadata:
   name: job-b
   spec:
   backoffLimit: 2
   completions: 2
   parallelism: 2
   template:
   metadata:
     annotations:
       scheduling.k8s.io/group-name: job-b-pg
       volcano.sh/preemptable: "true"
   spec:
     containers:
     - image: nginx:1.14.2
       imagePullPolicy: IfNotPresent
       name: nginx
       ports:
         - containerPort: 80
       resources:
         requests:
           cpu: 2000m
           memory: 200Mi
         limits:
           cpu: 2000m
           memory: 200Mi
     restartPolicy: Never
     terminationGracePeriodSeconds: 1
     schedulerName: volcano
   ---
   apiVersion: scheduling.volcano.sh/v1beta1
   kind: PodGroup
   metadata:
   annotations:
   scheduling.k8s.io/reclaimable: "true"
   name: job-b-pg
   namespace: default
   spec:
   minMember: 2
   queue: queue-b

Describe the results you received and expected

Reclaim action should use the Statement.Evict and can be caceled with Statement.Discard

What version of Volcano are you using?

master

Any other relevant information

No response

[Monokaix]

You mean the gang can't be met? I think the gang plugin should returen no victims if it's not met.

[lowang-bh]

We need to use statement.Evict instead of ssn.Evict. ssn package doesn't support transaction。

[lowang-bh]

I think the gang plugin should returen no victims if it's not met.

That's victim job will return nil if its gang cannot be met. The reclaimor job's gang is not met and should revert.