apiVersion: apps/v1
kind: Deployment
metadata:
  name: {{ include "llm-app.fullname" . }}
spec:
  replicas: {{ .Values.replicaCount }}
  selector:
    matchLabels: {{- include "llm-app.selectorLabels" . | nindent 6 }}
  template:
    metadata:
      labels: {{- include "llm-app.selectorLabels" . | nindent 8 }}
    spec:
      containers:
        - name: vllm-server
          # Image entrypoint is already `vllm serve`; args start with the model tag.
          image: "{{ .Values.image.repository }}{{ if .Values.image.digest }}@{{ .Values.image.digest }}{{ else }}:{{ .Values.image.tag }}{{ end }}"
          imagePullPolicy: {{ .Values.image.pullPolicy }}
          args:
            - {{ .Values.model.name | quote }}
            - "--host"
            - "0.0.0.0"
            - "--port"
            - {{ .Values.server.port | quote }}
            - "--served-model-name"
            - {{ .Values.model.alias | quote }}
            - "--max-model-len"
            - {{ .Values.model.maxModelLen | quote }}
            - "--dtype"
            - {{ .Values.model.dtype | quote }}
            {{- with .Values.server.extraArgs }}
            {{- toYaml . | nindent 12 }}
            {{- end }}
          env:
            - name: HF_HOME
              value: /cache/huggingface
            - name: VLLM_CPU_KVCACHE_SPACE
              value: "2"
            {{- if gt (int .Values.server.ompThreads) 0 }}
            - name: OMP_NUM_THREADS
              value: {{ .Values.server.ompThreads | quote }}
            {{- end }}
          ports:
            - name: http
              containerPort: {{ .Values.server.port }}
              protocol: TCP
          readinessProbe:
            httpGet:
              path: /health
              port: http
            # vLLM CPU cold-start is ~2 min + HF download on first boot.
            initialDelaySeconds: 60
            periodSeconds: 10
            timeoutSeconds: 5
            failureThreshold: 180
          livenessProbe:
            httpGet:
              path: /health
              port: http
            initialDelaySeconds: 600
            periodSeconds: 30
            timeoutSeconds: 5
            failureThreshold: 6
          resources: {{- toYaml .Values.resources | nindent 12 }}
          volumeMounts:
            - name: cache
              mountPath: /cache
            - name: shm
              mountPath: /dev/shm
      volumes:
        - name: cache
          emptyDir:
            sizeLimit: {{ .Values.modelCache.sizeLimit }}
        - name: shm
          emptyDir:
            medium: Memory
            sizeLimit: 1Gi