web & 无NodePort

创建配置文件

cat >tomcat_deploy.yaml<<EOF
apiVersion: apps/v1
kind: Deployment
metadata:
  name: mytomcat
spec:
  replicas: 1
  selector:
    matchLabels:
      app: mytomcat
  template:
    metadata:
      name: mytomcat
      labels:
        app: mytomcat
    spec:
      containers:
      - name: mytomcat
        image: tomcat:8
        ports:
        - containerPort: 8080
EOF

应用配置文件

kubectl apply -f tomcat_deploy.yaml

获取pod地址

kubectl get pods

通过pod地址访问

curl pod ip:pod port

tensorflow & NodePort

创建deployment配置文件

cat >tensorflow_deploy.yaml<<EOF
apiVersion: apps/v1
kind: Deployment
metadata:
  name: tensorflow
spec:
  replicas: 1
  selector:
    matchLabels:
      app: tensorflow 
  template:
    metadata:
      name: tensorflow
      labels:
        app: tensorflow
    spec:
      containers:
      - name: tensorflow
        imagePullPolicy: IfNotPresent
        image: tensorflow1_18:v2
        ports:
        - containerPort: 6081
EOF

创建service配置文件

cat >tensorflow_service_nodeport.yaml<<EOF
apiVersion: v1
kind: Service
metadata:
  name: front-tensorflow
  labels: 
    app: tensorflow
spec:
  ports: 
  - port: 6081
    targetPort: 6081
    nodePort: 30000
  selector: 
    app: tensorflow
  type: NodePort
EOF

应用配置文件

kubectl apply -f tensorflow_deploy.yaml
kubectl apply -f tensorflow_service_nodeport.yaml

使用浏览器访问应用，打开NodeIp:NodePort页面

tensorflow & ingress

见第二章，《安装ingress》

指定node部署

先给node打标签

kubectl label nodes dsai slave=136

cat >tomcat_deploy_node.yaml<<EOF
apiVersion: apps/v1
kind: Deployment
metadata:
  name: mytomcat
spec:
  replicas: 1
  selector:
    matchLabels:
      app: mytomcat
  template:
    metadata:
      name: mytomcat
      labels:
        app: mytomcat
    spec:
      containers:
      - name: mytomcat
        image: tomcat:8
        ports:
        - containerPort: 8080
      nodeSelector:
        slave: "136"
EOF

应用配置文件

kubectl apply -f tomcat_deploy_node.yaml

GPU调度(隔离模式)

cat >deepo_deploy_node.yaml<<EOF
apiVersion: apps/v1
kind: Deployment
metadata:
  name: deepo
spec:
  replicas: 1
  selector:
    matchLabels:
      app: deepo
  template:
    metadata:
      name: deepo
      labels:
        app: deepo
    spec:
      containers:
      - name: deepo
        imagePullPolicy: IfNotPresent
        image: wenfengand/deepo:testd
        resources:
          limits:
            nvidia.com/gpu: 1 # 必须为整数
      nodeSelector:
        slave: "136"
EOF
kubectl apply -f deepo_deploy_node.yaml

运行kubectl get pods找到pod名称，与container进行交互

kubectl exec -it deepo-6ffccbdb4d-bwh99 /bin/bash

运行nvidia-smi查看GPU数量

在gpu node节点查看所有GPU数量

可以看出在pods中gpu数量为1块，而slave节点中gpu数量为2块，k8s完成gpu调度。

GPU调度（共享模式）

cat >deepo_deploy_node_share.yaml<<EOF
apiVersion: apps/v1
kind: Deployment
metadata:
  name: deepo
spec:
  replicas: 1
  selector:
    matchLabels:
      app: deepo
  template:
    metadata:
      name: deepo
      labels:
        app: deepo
    spec:
      containers:
      - name: deepo
        imagePullPolicy: IfNotPresent
        image: wenfengand/deepo:testd
      nodeSelector:
        slave: "136"
EOF
kubectl apply -f deepo_deploy_node_share.yaml

使用与隔离模式时相同的命令，可知容器内可见的GPU数量为2，与node中的GPU数量相同。

参考

• k8s调度-指定node