HPCGame kube 集群使用指南¶

kubectl exec pod/<podname> -ti -- bash

kubectl port-forward pod/{podname} 10022:22

kubectl get lq
kubectl get cq

kubectl get lq

NAME      CLUSTERQUEUE    PENDING WORKLOADS   ADMITTED WORKLOADS
default   default-queue   1

kubectl get cq

NAME            COHORT    PENDING WORKLOADS
default-queue   hpcgame   6
judge-queue     hpcgame   0

kubectl describe lq
kubectl describe cq

kubectl describe cq judge-queue

krun init

krun host

krun host -n -I > hostfile

mpirun -hostfile hostfile <your_program>

apiVersion: v1
kind: Pod
metadata:
  name: test
  labels:
    name: test
spec:
  nodeSelector:
    hpc.lcpu.dev/partition: __REPLACE_NODE__
  containers:
    - name: test
      securityContext:
        capabilities:
          add: ["SYS_PTRACE", "IPC_LOCK"]
      image: crmirror.lcpu.dev/hpcgame/intel:latest
      resources:
        limits:
          memory: __REPLACE_LIMITS_MEMORY__
          cpu: __REPLACE_LIMITS_CPU__
      command:
        - sleep
        - inf
      # 添加 volumeMounts 配置
      volumeMounts:
        - name: my-pvc-volume  # 卷的名称,需要和下面volumes中的名称对应
          mountPath: /mnt/data # 容器内的挂载路径
  # 添加 volumes 配置
  volumes:
    - name: my-pvc-volume     # 卷的名称
      persistentVolumeClaim:
        claimName: my-pvc     # 已存在的PVC名称

apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: data
spec:
  resources:
    requests:
      storage: 64Gi
  accessModes:
    - ReadWriteMany
  storageClassName: wm2-nfs

---

apiVersion: jobset.x-k8s.io/v1alpha2
kind: JobSet
metadata:
  name: testMPIJob # Replace with your jobname
spec:
  network:
    enableDNSHostnames: true
    subdomain: testMPIJob
  replicatedJobs:
    - name: workers
      replicas: 1
      template:
        spec:
          backoffLimit: 0
          # completions 和 parallelism 必须相等，为 Pod 数
          completions: __REPLACE_POD_NUM__
          parallelism: __REPLACE_POD_NUM__
          template: # Describe a pod
            metadata:
              annotations:
                ssh-operator.lcpu.dev/inject: enabled
                lxcfs.lcpu.dev/inject: disabled
            spec:
              nodeSelector:
                # 与单 Pod 的节点选择器功能相同
                hpc.lcpu.dev/partition: __REPLACE_NODE__

              # 将下方内容解除注释，即可指定任务调度到不同节点
              # ------------ BEGIN ------------
              # affinity:
              #   podAntiAffinity:
              #     requiredDuringSchedulingIgnoredDuringExecution:
              #       - topologyKey: "kubernetes.io/hostname"
              #         labelSelector:
              #           matchExpressions:
              #             - key: jobset.x-k8s.io/name
              #               operator: In
              #               values:
              #                 - __REPLACE_JOBNAME__
              # ------------ END ------------

              containers: # Describe a container inside a pod
                - name: worker
                  # Pre-defined images provided by HPCGame:
                  #   crmirror.lcpu.dev/hpcgame/full:latest
                  #   crmirror.lcpu.dev/hpcgame/llvm:latest
                  #   crmirror.lcpu.dev/hpcgame/gcc:latest
                  #   crmirror.lcpu.dev/hpcgame/nvhpc:latest
                  #   crmirror.lcpu.dev/hpcgame/julia:latest
                  #   crmirror.lcpu.dev/hpcgame/base:latest
                  #   crmirror.lcpu.dev/hpcgame/intel:latest
                  #   crmirror.lcpu.dev/hpcgame/cuda:latest
                  #   crmirror.lcpu.dev/hpcgame/aocc:latest
                  #   crmirror.lcpu.dev/hpcgame/hpckit:latest
                  # Or any other images that could be get via a URL.
                  image: __REPLACE_CONTAINER_IMAGE__

                  # Note: replace this with command that you want to use.
                  # ```
                  # command:
                  #   - sleep
                  #   - inf
                  # ```
                  # Or
                  # `command: ["sleep", "inf"]`
                  # This is the default command that would keep the container to run sliently
                  # for the convenience of login
                  command:
                    - sleep
                    - inf
                  resources:
                    # HPC场景中，limits和requests中所有值必须相同
                    limits:
                      cpu: __REPLACE_LIMITS_CPU__ # E.g. 4
                      memory: __REPLACE_LIMITS_MEMORY__ # E.g. 8Gi
                      rdma.hpc.lcpu.dev/hca_cx5: 1
                    requests:
                      cpu: __REPLACE_REQUESTS_CPU__ # E.g. 4
                      memory: __REPLACE_REQUESTS_MEMORY__ # E.g. 8Gi
                      rdma.hpc.lcpu.dev/hca_cx5: 1
                  volumeMounts:
                    - name: __REPLACE_VOLUME_NAME__ # The name of volume defined in `volumes` section, see below
                      mountPath: __REPLACE_VOLUME_MOUNT_PATH__
              volumes:
                - name: __REPLACE_VOLUME_NAME__
                  persistentVolumeClaim: # From which PVC does this volume come from?
                    claimName: __REPLACE_PVC_NAME__