apiVersion: kubeflow.org/v1alpha2
kind: MPIJob
metadata:
  name: tensorflow-benchmarks-efa
spec:
  slotsPerWorker: 8
  cleanPodPolicy: Running
  mpiReplicaSpecs:
    Launcher:
      replicas: 1
      template:
         spec:
           containers:
           - image: 898739678081.dkr.ecr.us-west-2.amazonaws.com/efa-eks-benchmark:0.2gamma
             name: tensorflow-benchmarks-efa
             command: ["/bin/sh"]
             args: ["-c", "mpirun \
                      --allow-run-as-root \
                      -np 16 \
                      -x FI_PROVIDER=efa \
                      -x NCCL_DEBUG=INFO \
                      -x NCCL_ALGO=ring \
                      -x FI_EFA_TX_MIN_CREDITS=64 \
                      -x FI_EFA_ENABLE_SHM_TRANSFER=0 \
                      --mca plm_rsh_no_tree_spawn 1 \
                      --bind-to none -map-by slot --mca pml ob1 \
                      --mca btl_vader_single_copy_mechanism none \
                      --mca btl_tcp_if_exclude lo,docker0 \
                      python \
                      benchmarks/scripts/tf_cnn_benchmarks/tf_cnn_benchmarks.py \
                      --model=resnet50 \
                      --batch_size=256 \
                      --variable_update=horovod"]
    Worker:
      replicas: 2
      template:
        spec:
          containers:
          - image: 898739678081.dkr.ecr.us-west-2.amazonaws.com/efa-eks-benchmark:0.2gamma
            name: tensorflow-benchmarks-efa
            securityContext:
              privileged: true
            volumeMounts:
            - mountPath: /dev/infiniband/uverbs0
              name: infiniband-efa
            resources:
              limits:
                nvidia.com/gpu: 8
                hugepages-2Mi: 5120Mi
              requests:
                memory: 8000Mi
          volumes:
          - name: infiniband-efa
            hostPath:
              path: /dev/infiniband/uverbs0