Wareless Group Blog

# 创建一个 Kubernetes 的 Namespace
apiVersion: v1
kind: Namespace
metadata:
  name: detection

---

# 创建一个 PV
apiVersion: v1
kind: PersistentVolume
metadata:
  name: infer-worker
  labels:
    type: local
    app: infer-worker
  namespace: detection
spec:
  storageClassName: local-path
  capacity:
    storage: 10Gi
  volumeMode: Filesystem
  accessModes:
    - ReadWriteMany
  hostPath:
    # 指定了本机对应的存储位置
    path: /raid/model-weights

---

# 创建了一个 PVC
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: infer-worker
  labels:
    app: infer-worker
  namespace: detection
spec:
  storageClassName: local-path
  accessModes:
    - ReadWriteMany
  resources:
    requests:
      storage: 10Gi

---

# 创建了一个 ConfigMap（http-server）
apiVersion: v1
kind: ConfigMap
metadata:
  name: http-server
  labels:
    app: http-server
  namespace: detection
data:
  # 指定的是环境变量
  # infer worker 对应的 URL
  WORKER_URL: tcp://infer-worker:5001 # 这里指定的域名是按照 kubernetes 的规则来的

---

# 创建了一个 ConfigMap （infer-worker)
apiVersion: v1
kind: ConfigMap
metadata:
  name: infer-worker
  labels:
    app: infer-worker
  namespace: detection
data:
  # 指定环境变量
  # 指定 FTP 连接 IP（域名）
  FTP_CONNECT_IP: ftp.server
  # 指定 权重文件位置（容器中的路径）
  WEIGHT_FILE: /model/best.pt

---

# 创建了一个 http-server 的 部署（deployment）
apiVersion: apps/v1
kind: Deployment
metadata:
  name: http-server
  namespace: detection
spec:
  selector:
    matchLabels:
      app: http-server
  replicas: 1 # POD 实例个数
  strategy:   # 创建策略
    type: Recreate
  template: # POD 模板
    metadata:
      labels:
        app: http-server
    spec:
      containers: # 容器参数
      - name: http-server # 容器名称
        image: http-server # 镜像名称
        args: # 命令行参数
        - '--bind=0.0.0.0:8080' # 指定绑定 HTTP 的位置
        - '--timeout=1000' # 防止超时
        imagePullPolicy: "Never" # 组织拉取镜像（离线部署，如果是在线部署，可以选择 Always）
        envFrom: # 设置环境变量
          - configMapRef: # 并从 config-map http-server 中获取，上面 data 中的内容就会变成环境变量和相关的值。
              name: http-server
        ports: # 容器导出的端口
        - containerPort: 8080

---

# 创建了一个 http-server 的 部署（deployment） （同上）
apiVersion: apps/v1
kind: Deployment
metadata:
  name: infer-worker
  namespace: detection
spec:
  selector:
    matchLabels:
      app: infer-worker
  replicas: 1
  strategy:
    type: Recreate
  template:
    metadata:
      labels:
        app: infer-worker
    spec:
      hostAliases:
      - hostnames: [ "ftp.server" ]
        ip: "192.168.1.100"
      containers:
      - name: infer-worker
        image: infer-worker
        imagePullPolicy: "Never"
        resources: # 配置环境资源限制
          limits:
            nvidia.com/gpu: 1 # 分配一个 GPU 来使用
        ports:
        - containerPort: 5001
        volumeMounts:  # 将 上面创建的 PVC 绑定并挂载到容器的 /model 文件夹中
          - mountPath: /model
            name: model
      volumes: # 声明的挂载的卷，采用上面创建的 PVC
        - name: model
          persistentVolumeClaim:
            claimName: infer-worker

---

# 配置 http-server 的 Service
apiVersion: v1
kind: Service
metadata:
  name: http-server
  labels:
    app: http-server
  namespace: detection
spec:
  selector:
    app: http-server
  ports: # 指定端口， 从 8080 绑定到 8080
  - port: 8080
    targetPort: 8080

---

# 配置 infer-worker 的 Service
apiVersion: v1
kind: Service
metadata:
  name: infer-worker
  labels:
    app: infer-worker
  namespace: detection
spec:
  selector:
    app: infer-worker
  ports:
  - port: 5001 # 指定端口，从 5001 绑定到 5001
    targetPort: 5001

---

# 配置 ingress 暴漏 http 服务
apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
  name: proxy
  namespace: detection
spec:
  rules:
  - host: detection # http 服务的域名
    http:
      paths:
      - path: /detect # http 服务的路径
        pathType: Exact # http url 的绑定方式
        backend:
          service: # 指定转发到的 server 和对应的端口
            name: http-server
            port:
              number: 8080