Kubernetes Deployment AI Prompt

Write Kubernetes manifests for deploying this ML model serving application at production scale.

1. Deployment manifest:
   - Replicas: {{min_replicas}} initial
   - Resource requests and limits:
     - CPU: request={{cpu_request}}, limit={{cpu_limit}}
     - Memory: request={{memory_request}}, limit={{memory_limit}}
     - GPU: nvidia.com/gpu: 1 (if GPU-based)
   - Rolling update strategy: maxUnavailable=0, maxSurge=1 (zero-downtime deploys)
   - Liveness probe: GET /health every 10s, failure threshold 3
   - Readiness probe: GET /ready every 5s (only route traffic when model is loaded)
   - Startup probe: GET /ready with longer timeout for slow model loading

2. Horizontal Pod Autoscaler (HPA):
   - Scale based on: CPU utilization target {{cpu_target}}% or custom metric (requests per second)
   - Min replicas: {{min_replicas}}, max replicas: {{max_replicas}}
   - Scale-down stabilization: 5 minutes to prevent thrashing

3. Service and Ingress:
   - ClusterIP Service for internal traffic
   - Ingress with TLS termination, rate limiting, and timeout settings

4. ConfigMap and Secret management:
   - Non-sensitive config in ConfigMap (model path, log level, batch size)
   - Sensitive config in Secrets (API keys, database credentials)
   - Mount secrets as environment variables, not files

5. Pod disruption budget:
   - minAvailable: {{min_available}} to prevent all pods being evicted simultaneously

6. Namespace and RBAC:
   - Dedicated namespace for ML serving
   - ServiceAccount with minimal permissions

Return: Deployment, HPA, Service, Ingress, ConfigMap, and PodDisruptionBudget manifests.