yaml example

example.yaml

---
# jiansuan job example
apiVersion: batch.jiansuan.tech/v1alpha1
kind: Job
metadata:
  labels:
    jiansuan.tech/jiansuan.label.name: "jiansuan-job-db8kyn-1668675887705"
    jiansuan.tech/jiansuan.label.username: "peiwenqian"
    jiansuan.tech/jiansuan.label.group: "test-group"
    jiansuan.tech/jiansuan.label.workload: "job"
    jiansuan.tech/jiansuan.label.framework: "pytorch"
  name: jiansuan-job-db8kyn-1668675887705
  namespace: jiansuan-default-ns
spec:
  maxRetry: 3
  frameworkType: pytorch
  minAvailable: 1
  maxAvailable: 1
  scheduler:
    name: volcano
    queue: "vc-queue-1"
  tasks:
  - replicas: 1
    name: master
    template:
      metadata:
        annotations:
          jiansuan.tech/jiansuan.job.role-name: normal
        labels:
          iansuan.tech/jiansuan.label.name: "jiansuan-job-db8kyn-1668675887705"
          jiansuan.tech/jiansuan.label.username: "peiwenqian"
          jiansuan.tech/jiansuan.label.group: "test-group"
          jiansuan.tech/jiansuan.label.workload: "job"
          jiansuan.tech/jiansuan.label.framework: "pytorch"
      spec:
        containers:
        - command:
          - /bin/bash
          - /nfs/common/jiansuan/scripts/current/offline-vj-pytorch-entrypoint.sh
          - xxxxx
          env:
          - name: JIANSUAN_USER
            value: luban
          - name: JIANSUAN_JOB_CREATOR
            value: peiwenqian
          image: image-url
          name: jiansuan-job-db8kyn-1668675887705-master
          resourceGroup: A100_80x8
          storageMount:
          - name: nfs-common
            type: hostPath
            readOnly: true
            mountPath: "/nfs/common"
            sourceDescription: "file:///mnt_jiansuan/common"
          - name: user-nas
            type: hostPath
            mountPath: "/nfs/private"
            sourceDescription: "file:///mnt_jiansuan/private/users/peiwenqian"
        nodeSelector:
          jiansuan_gpu_a100: "true"

---
# rendered volcano job example
apiVersion: batch.volcano.sh/v1alpha1
kind: Job
metadata:
  labels:
    jiansuan.tech/jiansuan.label.name: "jiansuan-job-db8kyn-1668675887705"
    jiansuan.tech/jiansuan.label.username: "peiwenqian"
    jiansuan.tech/jiansuan.label.group: "test-group"
    jiansuan.tech/jiansuan.label.workload: "job"
    jiansuan.tech/jiansuan.label.framework: "pytorch"
  name: jiansuan-job-db8kyn-1668675887705
  namespace: jiansuan-default-ns
spec:
  maxRetry: 3
  minAvailable: 1
  plugins:
    env: []
    svc: []
  policies:
  - action: RetryJob
    event: PodEvicted
  - action: RestartJob
    event: PodFailed
  queue: vc-queue-1
  schedulerName: volcano
  tasks:
  - maxRetry: 3
    minAvailable: 1
    name: master
    policies:
    - action: CompleteJob
      event: TaskCompleted
    replicas: 1
    template:
      metadata:
        annotations:
          jiansuan.tech/jiansuan.job.role-name: normal
        labels:
          iansuan.tech/jiansuan.label.name: "jiansuan-job-db8kyn-1668675887705"
          jiansuan.tech/jiansuan.label.username: "peiwenqian"
          jiansuan.tech/jiansuan.label.group: "test-group"
          jiansuan.tech/jiansuan.label.workload: "job"
          jiansuan.tech/jiansuan.label.framework: "pytorch"
      spec:
        containers:
        - command:
          - /bin/bash
          - /nfs/common/jiansuan/scripts/current/offline-vj-pytorch-entrypoint.sh
          - xxxxx
          env:
          - name: NVIDIA_DRIVER_CAPABILITIES
            value: all
          - name: JIANSUAN_USER
            value: luban
          - name: JIANSUAN_JOB_CREATOR
            value: peiwenqian
          - name: JIANSUAN_RESOURCE_NUM_CPU
            value: "96"
          - name: JIANSUAN_RESOURCE_NUM_GPU
            value: "8"
          - name: JIANSUAN_RESOURCE_NUM_MEM
            value: 180Gi
          - name: JIANSUAN_DISTRIBUTED_IDENTIFICATION
            value: jiansuan-job-db8kyn-1668675887705
          - name: JIANSUAN_DISTRIBUTED_NODE_COUNT
            value: "1"
          - name: JIANSUAN_DISTRIBUTED_TASK_ROLE
            value: master
          - name: JIANSUAN_FRAMEWORK
            value: pytorch
          image: image-url
          name: jiansuan-job-db8kyn-1668675887705-master
          resources:
            limits:
              cpu: "96"
              memory: 180Gi
              nvidia.com/gpu: "8"
            requests:
              cpu: "48"
              memory: 90Gi
              nvidia.com/gpu: "8"
          securityContext:
            capabilities:
              add:
              - ALL
            privileged: false
          volumeMounts:
          - mountPath: /dev/infiniband
            name: infiniband
          - mountPath: /sys/fs/cgroup
            name: cgroup
          - mountPath: /dev/shm
            name: dev-shm
          - mountPath: /nfs/common
            name: nfs-common
            readOnly: true
          - mountPath: /nfs/private
            name: user-nas
        nodeSelector:
          jiansuan_gpu_a100: "true"
        dnsPolicy: ClusterFirstWithHostNet
        hostNetwork: true
        imagePullSecrets:
        - name: jiansuan-registry-secret
        restartPolicy: Never
        volumes:
        - hostPath:
            path: /dev/infiniband
          name: infiniband
        - hostPath:
            path: /sys/fs/cgroup
          name: cgroup
        - hostPath:
            path: /mnt_jiansuan/common
          name: nfs-common
        - hostPath:
            path: /mnt_jiansuan/private/users/peiwenqian
          name: user-nas
        - emptyDir:
            medium: Memory
          name: dev-shm