# NEX-Sprachausgabe (Piper, CPU) auf gpu-pve52, Host-Netz Port 8179 (ufw: nur appsrv08).
apiVersion: apps/v1
kind: Deployment
metadata:
  name: nex-tts
  namespace: nex
  labels: {app: nex-tts}
spec:
  replicas: 1
  strategy: {type: Recreate}
  selector:
    matchLabels: {app: nex-tts}
  template:
    metadata:
      labels: {app: nex-tts}
    spec:
      hostNetwork: true
      dnsPolicy: ClusterFirstWithHostNet
      nodeSelector:
        kubernetes.io/hostname: gpu-pve52
      containers:
        - name: piper
          image: knaster/nex-tts-piper:1.8.0
          imagePullPolicy: Never
          env:
            - {name: TTS_PORT, value: "8179"}
            - {name: PIPER_VOICES_DIR, value: /voices}
            - {name: PIPER_VOICES, value: "de_DE-thorsten-high,de_DE-kerstin-low"}
            - {name: PIPER_DEFAULT_VOICE, value: de_DE-thorsten-high}
            - {name: PCM_SAMPLE_RATE, value: "24000"}
            - {name: TTS_MAX_CHARS, value: "5000"}
            - {name: TTS_PARALLEL, value: "2"}
          resources:
            limits: {cpu: "3", memory: 2Gi}   # ein Kern bleibt für whisper-server und das System
            requests: {cpu: "500m", memory: 512Mi}
          securityContext:
            readOnlyRootFilesystem: true
          volumeMounts:
            - {name: voices, mountPath: /voices, readOnly: true}
            - {name: tmp, mountPath: /tmp}
          startupProbe:
            httpGet: {path: /health, port: 8179, host: 127.0.0.1}
            periodSeconds: 5
            failureThreshold: 24
          readinessProbe:
            httpGet: {path: /health, port: 8179, host: 127.0.0.1}
            periodSeconds: 30
          livenessProbe:
            httpGet: {path: /health, port: 8179, host: 127.0.0.1}
            periodSeconds: 30
            failureThreshold: 3
      volumes:
        - name: voices
          hostPath: {path: /srv/models/piper, type: Directory}
        - name: tmp
          emptyDir: {sizeLimit: 64Mi}