pserver.yaml

apiVersion: extensions/v1beta1
kind: ReplicaSet
metadata: {name: fluid-ctr-pserver}
spec:
  replicas: 2
  template:
    metadata:
      labels: {paddle-job-pserver: fluid-ctr}
    spec:
      containers:
      - command: [paddle_k8s, start_fluid]
        env:
        - {name: GLOG_v, value: '0'}
        - {name: GLOG_logtostderr, value: '1'}
        - {name: TOPOLOGY, value: ''}
        - {name: TRAINER_PACKAGE, value: /workspace}
        - {name: PADDLE_INIT_NICS, value: eth2}
        - name: NAMESPACE
          valueFrom:
            fieldRef: {fieldPath: metadata.namespace}
        - name: POD_IP
          valueFrom:
            fieldRef: {fieldPath: status.podIP}
        - name: POD_NAME
          valueFrom:
            fieldRef: {fieldPath: metadata.name}
        - name: PADDLE_CURRENT_IP
          valueFrom:
            fieldRef: {fieldPath: status.podIP}
        - {name: PADDLE_JOB_NAME, value: fluid-ctr}
        - {name: PADDLE_IS_LOCAL, value: '0'}
        - {name: PADDLE_TRAINERS_NUM, value: '2'}
        - {name: PADDLE_PSERVERS_NUM, value: '2'}
        - {name: FLAGS_rpc_deadline, value: '36000000'}
        - {name: ENTRY, value: cd /workspace/ctr && python train.py --is_local 0 --cloud_train 1}
        - {name: PADDLE_PORT, value: '30236'}
        - {name: LD_LIBRARY_PATH, value: '/usr/local/lib:/usr/local/nvidia/lib64:/usr/local/rdma/lib64:/usr/lib64/mlnx_ofed/valgrind'}
        - {name: PADDLE_TRAINING_ROLE, value: PSERVER}
        - {name: TRAINING_ROLE, value: PSERVER}
        image: wopeizl/paddle_ctr_distribute
        imagePullPolicy: Always
        name: pserver
        volumeMounts:
        - {mountPath: /mnt/seqdata, name: seqdata}
        resources:
          limits: {cpu: '10', memory: 30Gi}
          requests: {cpu: '1', memory: 100M}
      hostNetwork: true
      imagePullSecrets:
      - {name: regcred}
      volumes:
      - hostPath: {path: /home/work/}
        name: seqdata