apiVersion: kubeflow.org/v1alpha2
kind: MPIJob
metadata:
  name: nccl-tests-efa
spec:
  slotsPerWorker: 1
  cleanPodPolicy: Running
  mpiReplicaSpecs:
    Launcher:
      replicas: 1
      template:
         spec:
          imagePullPolicy: IfNotPresent
          restartPolicy: OnFailure
          containers:
          - image: public.ecr.aws/w6p6i9i7/aws-efa-nccl-rdma:base-cudnn8-cuda11-ubuntu18.04
            name: nccl-test-launcher
            env:
             - name: LD_LIBRARY_PATH
               value: /opt/amazon/openmpi/lib:/opt/nccl/build/lib:/opt/amazon/efa/lib:/opt/aws-ofi-nccl/install/lib:/usr/local/nvidia/lib:$LD_LIBRARY_PATH
             - name: PATH
               value: $PATH:/opt/amazon/efa/bin
             - name: XLA_FLAGS
               value: "--xla_gpu_cuda_data_dir=/usr/local/cuda"
             - name: TF_XLA_FLAGS
               value: "--tf_xla_cpu_global_jit"
            command:
            - /opt/amazon/openmpi/bin/mpirun
            - --allow-run-as-root
            - --tag-output
            - -np
            - "2"
            - -bind-to
            - none
            - -map-by
            - slot
            - -x
            - PATH
            - -x
            - LD_LIBRARY_PATH
            - -x
            - XLA_FLAGS
            - -x
            - TF_XLA_FLAGS
            - -x
            - NCCL_DEBUG=INFO
            - -x
            - NCCL_ALGO=RING
            - -x
            - FI_EFA_USE_DEVICE_RDMA=1
            - -x
            - RDMAV_FORK_SAFE=1
            #- --mca
            #- pml
            #- ^cm
            - --oversubscribe
            - /tmp/nccl-tests/build/all_reduce_perf
            - -b
            - "1"
            - -e
            - 1G
            - -f
            - "2"
            - -t
            - "1"
            - -g
            - "1"
            - -c
            - "1"
            - -n
            - "1000"

    Worker:
      replicas: 2
      template:
        spec:
          imagePullPolicy: IfNotPresent
          containers:
          - image: public.ecr.aws/w6p6i9i7/aws-efa-nccl-rdma:base-cudnn8-cuda11-ubuntu18.04
            name: nccl-worker
            resources:
              limits:
                nvidia.com/gpu: 1
#                hugepages-2Mi: 1Gi
                ephemeral-storage: "8Gi"
#                vpc.amazonaws.com/efa: 1
                memory: 2Gi
              requests:
                nvidia.com/gpu: 1
#                hugepages-2Mi: 1Gi
                ephemeral-storage: "8Gi"
#                vpc.amazonaws.com/efa: 1
                memory: 2Gi