resnet_serving_v1.yaml

apiVersion: extensions/v1beta1
kind: Deployment
metadata:
  name: resnet-serving
  labels:
    app: resnet-serving
    version: v1
spec:
  revisionHistoryLimit: 3
  progressDeadlineSeconds: 30
  strategy:
    rollingUpdate:
      maxUnavailable: 0
    type: RollingUpdate
  template:
    metadata:
      labels:
        app: resnet-serving
        version: v1
    spec:
      containers:
      - name: tensorflow-serving
        image: tensorflow/serving:latest
        command:
        - tensorflow_model_server
        - --port=9000
        - --rest_api_port=9001
        - --model_config_file=/serving/models/config/models.conf
        - --file_system_poll_wait_seconds=$(FS_POLL_WAIT_SECONDS)
        env:
        - name: FS_POLL_WAIT_SECONDS
          value: "60"
        - name: AWS_ACCESS_KEY_ID
          valueFrom:
            secretKeyRef:
              name: s3-storage-creds
              key: access_id
        - name: AWS_SECRET_ACCESS_KEY
          valueFrom:
            secretKeyRef:
              name: s3-storage-creds
              key: access_key
        - name: AWS_REGION
          value: us-east-2
        - name: S3_ENDPOINT
          value: s3.us-east-2.amazonaws.com
        ports:
        - containerPort: 9000
          protocol: TCP
        - containerPort: 9001
          protocol: TCP
        volumeMounts:
        - mountPath: /serving/models/config
          name: tf-serving-models-config
      volumes:
      - name: tf-serving-models-config
        configMap:
          name: tf-serving-models-config
---
apiVersion: v1
kind: Service
metadata:
  name: resnet-serving
  labels:
    app: resnet-serving
spec:
  ports:
  - name: grpc
    port: 9000
    protocol: TCP
  - name: rest
    port: 9001
    protocol: TCP
  selector:
    app: resnet-serving
---
apiVersion: v1
kind: ConfigMap
metadata:
  name: tf-serving-models-config
data:
  models.conf: |+
    model_config_list: {
      config: {
        name: "resnet",
        base_path: "s3://ml-models-repository/resnet",
        model_platform: "tensorflow",
        model_version_policy: {
          specific: {
            versions: 1
          }
        }
      }
    }