deployments.yaml

apiVersion: apps/v1
kind: Deployment
metadata:
  name: endpoint-picker
  labels:
    app: endpoint-picker
spec:
  replicas: 1
  selector:
    matchLabels:
      app: endpoint-picker
  template:
    metadata:
      labels:
        app: endpoint-picker
    spec:
      serviceAccountName: endpoint-picker
      terminationGracePeriodSeconds: 130
      containers:
      - name: epp
        image: quay.io/vllm-d/gateway-api-inference-extension/epp:latest
        imagePullPolicy: IfNotPresent
        args:
        - -poolName
        - "${POOL_NAME}"
        - -v
        - "4"
        - --zap-encoder
        - "json"
        - -grpcPort
        - "9002"
        - -grpcHealthPort
        - "9003"
        ports:
        - containerPort: 9002
        - containerPort: 9003
        - name: metrics
          containerPort: 9090
        livenessProbe:
          grpc:
            port: 9003
            service: inference-extension
          initialDelaySeconds: 5
          periodSeconds: 10
        readinessProbe:
          grpc:
            port: 9003
            service: inference-extension
          initialDelaySeconds: 5
          periodSeconds: 10
        env:
          - name: KVCACHE_INDEXER_REDIS_ADDR
            value: ${REDIS_HOST}:${REDIS_PORT}
          - name: HF_TOKEN
            valueFrom:
              secretKeyRef:
                name: ${HF_SECRET_NAME}
                key: ${HF_SECRET_KEY}