recreated pod doesn't get GPU memory

[cyyeh]

apiVersion: apps/v1
kind: Deployment
metadata:
  name: objectdetection-deployment
  namespace: iinumbers
  labels:
    deployment: objectdetection-deployment
spec:
  replicas: 2
  selector:
    matchLabels:
      app: objectdetection-server
  template:
    metadata:
      labels:
        app: objectdetection-server
    spec:
      # hostIPC is required for MPS communication
      hostIPC: true
      containers:
      - name: objectdetection-container
        image: 192.168.99.150:5000/objectdetectionapp:v0.1.0
        env:
        - name: CALLBACK_URL
          value: "http://facerecognition-service:8001"
        - name: PER_PROCESS_GPU_MEMORY_FRACTION
          value: "0.23"
        ports:
        - containerPort: 8888
        # Use virtual gpu resource here
        resources:
          limits:
            k8s.amazonaws.com/vgpu: 1
        volumeMounts:
        - name: nvidia-mps
          mountPath: /tmp/nvidia-mps
      volumes:
      - name: nvidia-mps
        hostPath:
          path: /tmp/nvidia-mps
---
apiVersion: v1
kind: Service
metadata:
  labels:
    service: objectdetection-service
  name: objectdetection-service
  namespace: iinumbers
spec:
  ports:
  - port: 8888
    targetPort: 8888
    nodePort: 30000
  selector:
    app: objectdetection-server
  type: NodePort

Above content is the yaml file for deployment and service. I've set replicas to be 2 for the deployment. However, as I destroyed one of the pods, I found that although another new pod was created, it didn't get any GPU memory from watching the result of nvidia-smi.