apiVersion: apps/v1
kind: Deployment
metadata:
  name: pipeline
  namespace: pipeline
  labels:
    app.kubernetes.io/name: pipeline
spec:
  # Keep at 1. The WebUI stores jobs in a local file and renders by spawning the
  # CLI child process; there is no shared state or distributed locking, so extra
  # replicas would each see a private jobs.json and could double-run jobs.
  replicas: 1
  strategy:
    type: Recreate            # RWO PVCs can't move while the old pod owns them
  selector:
    matchLabels:
      app.kubernetes.io/name: pipeline
  template:
    metadata:
      labels:
        app.kubernetes.io/name: pipeline
    spec:
      securityContext:
        runAsNonRoot: true
        runAsUser: 1000        # `node` user baked into the image
        runAsGroup: 1000
        fsGroup: 1000          # chowns mounted PVCs so uid 1000 can write them
      # Uncomment once you've created the Harbor pull secret (see README).
      # imagePullSecrets:
      #   - name: harbor-creds
      containers:
        - name: pipeline
          # ⚠️ Replace with your Harbor registry + project. See README.
          image: harbor.corp.shuidi.tech/library/pipeline:latest
          imagePullPolicy: IfNotPresent
          ports:
            - name: http
              containerPort: 3000
          envFrom:
            - configMapRef:
                name: pipeline-env
            - secretRef:
                name: pipeline-env-secret
          # Let in-flight renders finish on rolling update / drain. A long
          # render still exceeds this; interrupted jobs are auto-marked failed
          # on the next pod start (startup sweep in job-store).
          terminationGracePeriodSeconds: 60
          startupProbe:
            httpGet:
              path: /api/health/live
              port: http
            failureThreshold: 30
            periodSeconds: 10
          livenessProbe:
            httpGet:
              path: /api/health/live
              port: http
            initialDelaySeconds: 10
            periodSeconds: 30
            timeoutSeconds: 5
            failureThreshold: 3
          readinessProbe:
            httpGet:
              path: /api/health/ready
              port: http
            initialDelaySeconds: 5
            periodSeconds: 10
            timeoutSeconds: 5
          # Video rendering (Remotion/Chrome + ffmpeg) is CPU/memory heavy.
          # Tune to your cluster.
          resources:
            requests:
              cpu: "1000m"
              memory: 2Gi
            limits:
              cpu: "4000m"
              memory: 6Gi
          volumeMounts:
            - name: output
              mountPath: /app/output
            - name: jobs
              mountPath: /app/jobs
      volumes:
        - name: output
          persistentVolumeClaim:
            claimName: pipeline-output
        - name: jobs
          persistentVolumeClaim:
            claimName: pipeline-jobs