How can i add the Priorityclass to the TFjob？

[Wenshiqi222]

Since i wanna realize the gangscheduling with the Scheduler-Framework，i need to ensure that any pod belongs to the same PodGroup should has the same priority and different PodGroup should has different priority . i can add the podgroup in the "Labels” field，but since the "PriorityClass” belongs to Pod.Spec field， how can i add the PirorityClass to the PodGroup or the specific TFjob ？
Thanks for your help~

[gaocegege]

/cc @zw0610

[zw0610]

When you define spec for each ReplicaType, it's actually PodTemplateSpec which contains a field called PriorityClassName

A simple example could be:

Please note that in this example, different priority classes are assigned to different ReplicaType

apiVersion: kubeflow.org/v1
kind: MPIJob
metadata:
  name: tensorflow-mnist-elastic
spec:
  slotsPerWorker: 1
  cleanPodPolicy: Running
  mpiReplicaSpecs:
    Launcher:
      replicas: 1
      template:
        spec:
          priorityClassName: preserved-pc
          containers:
          - image: horovod/horovod:0.20.0-tf2.3.0-torch1.6.0-mxnet1.5.0-py3.7-cpu
            name: mpi-launcher
            command:
            - horovodrun
            args:
            - -np
            - "2"
            - --min-np
            - "1"
            - --max-np
            - "3"
            - --host-discovery-script
            - /etc/mpi/discover_hosts.sh
            - python
            - /examples/elastic/tensorflow2_mnist_elastic.py
            resources:
              requests:
                cpu: 1
                memory: 2Gi
              limits:
                cpu: 1
                memory: 2Gi
    Worker:
      replicas: 2
      template:
        spec:
          priorityClassName: preemptible-pc
          containers:
          - image: horovod/horovod:0.20.0-tf2.3.0-torch1.6.0-mxnet1.5.0-py3.7-cpu
            name: mpi-worker
            resources:
              requests:
                cpu: 2
                memory: 4Gi
              limits:
                cpu: 2
                memory: 4Gi

[Wenshiqi222]

When you define spec for each ReplicaType, it's actually PodTemplateSpec which contains a field called PriorityClassName

A simple example could be:

Please note that in this example, different priority classes are assigned to different ReplicaType

apiVersion: kubeflow.org/v1
kind: MPIJob
metadata:
  name: tensorflow-mnist-elastic
spec:
  slotsPerWorker: 1
  cleanPodPolicy: Running
  mpiReplicaSpecs:
    Launcher:
      replicas: 1
      template:
        spec:
          priorityClassName: preserved-pc
          containers:
          - image: horovod/horovod:0.20.0-tf2.3.0-torch1.6.0-mxnet1.5.0-py3.7-cpu
            name: mpi-launcher
            command:
            - horovodrun
            args:
            - -np
            - "2"
            - --min-np
            - "1"
            - --max-np
            - "3"
            - --host-discovery-script
            - /etc/mpi/discover_hosts.sh
            - python
            - /examples/elastic/tensorflow2_mnist_elastic.py
            resources:
              requests:
                cpu: 1
                memory: 2Gi
              limits:
                cpu: 1
                memory: 2Gi
    Worker:
      replicas: 2
      template:
        spec:
          priorityClassName: preemptible-pc
          containers:
          - image: horovod/horovod:0.20.0-tf2.3.0-torch1.6.0-mxnet1.5.0-py3.7-cpu
            name: mpi-worker
            resources:
              requests:
                cpu: 2
                memory: 4Gi
              limits:
                cpu: 2
                memory: 4Gi

Got it，really thanks for your help~