service.yaml

# Knative service manifest for Google Cloud Run.
#
# Deploy with:
#   gcloud run services replace deploy/cloudrun/service.yaml \
#     --region "$GCP_REGION" --project "$GCP_PROJECT"
#
# The `image` below is a placeholder — deploy/cloudrun/deploy.sh substitutes the
# real pushed tag before applying. Edit the resource values to taste.
apiVersion: serving.knative.dev/v1
kind: Service
metadata:
  name: document-ocr
  labels:
    app: document-ocr
  annotations:
    run.googleapis.com/ingress: all
spec:
  template:
    metadata:
      annotations:
        # OCR is CPU-bound and serialized inside the process (Semaphore(1)), so
        # one request per instance — let Cloud Run scale out horizontally.
        autoscaling.knative.dev/minScale: '0'
        autoscaling.knative.dev/maxScale: '10'
        # CPU is only allocated during requests (scale-to-zero friendly). The
        # first request after a cold start downloads/loads models, so allow a
        # generous startup window via the readiness probe below.
        run.googleapis.com/cpu-throttling: 'true'
    spec:
      containerConcurrency: 1
      timeoutSeconds: 300
      containers:
        - image: IMAGE_PLACEHOLDER
          ports:
            - name: http1
              containerPort: 8000
          resources:
            limits:
              cpu: '2'
              memory: 2Gi
          startupProbe:
            httpGet:
              path: /ready
            # Models load on startup; /ready returns 503 until they are ready.
            periodSeconds: 5
            failureThreshold: 30
            timeoutSeconds: 3
          livenessProbe:
            httpGet:
              path: /health
            periodSeconds: 30