homelab/immich/k8s/machine-learning.yaml

apiVersion: v1
kind: Service
metadata:
  name: immich-machine-learning
  namespace: immich
spec:
  selector:
    app: immich-machine-learning
  ports:
    - name: http
      port: 3003
      targetPort: 3003
---
apiVersion: apps/v1
kind: Deployment
metadata:
  annotations:
    reloader.stakater.com/auto: "true"
  name: immich-machine-learning-deployment
  namespace: immich
  labels:
    app: immich-machine-learning
spec:
  replicas: 1
  selector:
    matchLabels:
      app: immich-machine-learning
  strategy:
    type: Recreate
  template:
    metadata:
      labels:
        app: immich-machine-learning
    spec:
      containers:
        - name: immich-machine-learning
          image: ghcr.io/immich-app/immich-machine-learning:v3
          envFrom:
            - configMapRef:
                name: immich-config
            - secretRef:
                name: immich-secrets
          ports:
            - name: http
              containerPort: 3003
          volumeMounts:
            - name: model-cache
              mountPath: /cache
          # The first request pulls a model over the internet, so a cold start
          # is slower than a container start.
          startupProbe:
            httpGet:
              path: /ping
              port: http
            failureThreshold: 60
            periodSeconds: 5
            timeoutSeconds: 5
          readinessProbe:
            httpGet:
              path: /ping
              port: http
            periodSeconds: 10
            timeoutSeconds: 5
          livenessProbe:
            httpGet:
              path: /ping
              port: http
            initialDelaySeconds: 30
            periodSeconds: 30
            timeoutSeconds: 5
          # Same reasoning as the server: the request covers the idle cost
          # only, because the node has no spare cores to schedule against.
          # Recognition is the burst - a busy import wants both cores.
          resources:
            requests:
              cpu: "100m"
              memory: "1Gi"
            limits:
              cpu: "2000m"
              memory: "3Gi"
      volumes:
        - name: model-cache
          persistentVolumeClaim:
            claimName: immich-model-cache-pvc
---
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: immich-model-cache-pvc
  namespace: immich
spec:
  accessModes:
    - ReadWriteOnce
  resources:
    requests:
      storage: 2Gi