# 备模 / 小模型占位：演示多 Deployment + 路由分流
apiVersion: apps/v1
kind: Deployment
metadata:
  name: vllm-qwen-small
  namespace: gpu-infer
  labels:
    app: vllm-qwen-small
    model: qwen-small
spec:
  replicas: 1
  selector:
    matchLabels:
      app: vllm-qwen-small
  template:
    metadata:
      labels:
        app: vllm-qwen-small
        model: qwen-small
    spec:
      nodeSelector:
        nvidia.com/gpu.present: "true"
      tolerations:
        - key: nvidia.com/gpu
          operator: Equal
          value: "true"
          effect: NoSchedule
      containers:
        - name: vllm
          image: harbor.example.local/ai/vllm-openai:v0.8.5
          args:
            - --model
            - /models/qwen-small
            - --served-model-name
            - qwen-small
            - --host
            - 0.0.0.0
            - --port
            - "8000"
          ports:
            - name: http
              containerPort: 8000
          resources:
            requests:
              cpu: "4"
              memory: 24Gi
              nvidia.com/gpu: "1"
            limits:
              nvidia.com/gpu: "1"
              memory: 32Gi
          readinessProbe:
            httpGet:
              path: /health
              port: http
            initialDelaySeconds: 45
            periodSeconds: 10
          volumeMounts:
            - name: model
              mountPath: /models/qwen-small
              readOnly: true
      volumes:
        - name: model
          persistentVolumeClaim:
            claimName: pvc-qwen-small
---
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: pvc-qwen-small
  namespace: gpu-infer
spec:
  accessModes: ["ReadOnlyMany"]
  resources:
    requests:
      storage: 40Gi
  storageClassName: nfs-models
---
apiVersion: v1
kind: Service
metadata:
  name: vllm-qwen36-27b
  namespace: gpu-infer
spec:
  selector:
    app: vllm-qwen36-27b
  ports:
    - name: http
      port: 8000
      targetPort: http
---
apiVersion: v1
kind: Service
metadata:
  name: vllm-qwen-small
  namespace: gpu-infer
spec:
  selector:
    app: vllm-qwen-small
  ports:
    - name: http
      port: 8000
      targetPort: http
---
# 对 207 LiteLLM 暴露：内网 Ingress（示例域名）
apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
  name: vllm-ingress
  namespace: gpu-infer
  annotations:
    nginx.ingress.kubernetes.io/proxy-body-size: "32m"
    nginx.ingress.kubernetes.io/proxy-read-timeout: "300"
spec:
  ingressClassName: nginx
  rules:
    - host: vllm-qwen.gpu-k8s.example.local
      http:
        paths:
          - path: /
            pathType: Prefix
            backend:
              service:
                name: vllm-qwen36-27b
                port:
                  number: 8000
    - host: vllm-small.gpu-k8s.example.local
      http:
        paths:
          - path: /
            pathType: Prefix
            backend:
              service:
                name: vllm-qwen-small
                port:
                  number: 8000