API Deployment on EKS

# namespace.yaml
apiVersion: v1
kind: Namespace
metadata:
  name: mlops
  labels:
    app.kubernetes.io/name: retail-api
---

# serviceaccount.yaml
apiVersion: v1
kind: ServiceAccount
metadata:
  name: retail-api-sa
  namespace: mlops
  annotations:
    eks.amazonaws.com/role-arn: arn:aws:iam::842676018087:role/eks-sagemaker-access-role
---

# deployment.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
  name: retail-api
  namespace: mlops
  labels:
    app: retail-api
spec:
  replicas: 2
  selector:
    matchLabels:
      app: retail-api
  template:
    metadata:
      labels:
        app: retail-api
    spec:
      serviceAccountName: retail-api-sa
      containers:
      - name: retail-api
        image: 842676018087.dkr.ecr.ap-southeast-1.amazonaws.com/mlops/retail-api:latest
        ports:
        - containerPort: 8000
        env:
        - name: PORT
          value: "8000"
        - name: AWS_DEFAULT_REGION
          value: "ap-southeast-1"
        - name: MODEL_PACKAGE_GROUP
          value: "retail-price-sensitivity-models"
        resources:
          requests:
            memory: "512Mi"
            cpu: "250m"
          limits:
            memory: "1Gi"
            cpu: "500m"
        livenessProbe:
          httpGet:
            path: /health
            port: 8000
          initialDelaySeconds: 30
          periodSeconds: 10
        readinessProbe:
          httpGet:
            path: /health
            port: 8000
          initialDelaySeconds: 10
          periodSeconds: 5
---

# service.yaml
apiVersion: v1
kind: Service
metadata:
  name: retail-api-service
  namespace: mlops
  labels:
    app: retail-api
spec:
  selector:
    app: retail-api
  ports:
  - name: http
    port: 80
    targetPort: 8000
  type: LoadBalancer

# hpa.yaml
apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: retail-api-hpa
  namespace: mlops
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: retail-api
  minReplicas: 2
  maxReplicas: 5
  metrics:
  - type: Resource
    resource:
      name: cpu
      target:
        type: Utilization
        averageUtilization: 60

# Deploy all manifests in order
kubectl apply -f namespace.yaml
kubectl apply -f serviceaccount.yaml
kubectl apply -f deployment.yaml
kubectl apply -f service.yaml
kubectl apply -f hpa.yaml

# Kiểm tra trạng thái pods
kubectl get pods -n mlops

# Kiểm tra service và load balancer
kubectl get svc -n mlops

# Kiểm tra horizontal pod autoscaler  
kubectl get hpa -n mlops

# Kiểm tra logs của pod
kubectl logs -l app=retail-api -n mlops --tail=50

# Lấy URL của LoadBalancer
kubectl get svc retail-api-service -n mlops -o jsonpath='{.status.loadBalancer.ingress[0].hostname}'

# Test health check endpoint  
curl http://[LOAD_BALANCER_URL]/health

# Test API documentation
curl http://[LOAD_BALANCER_URL]/docs

# Test prediction endpoint với data format thật
curl -X POST http://[LOAD_BALANCER_URL]/predict \
  -H "Content-Type: application/json" \
  -d '{
    "BASKET_SIZE": "M",
    "BASKET_TYPE": "MIXED", 
    "STORE_REGION": "LONDON",
    "STORE_FORMAT": "LS",
    "SPEND": 125.50,
    "QUANTITY": 3,
    "PROD_CODE_20": "FOOD",
    "PROD_CODE_30": "FRESH"
  }'

# Port forward service đến localhost (nếu LoadBalancer chưa ready)
kubectl port-forward service/retail-api-service 8080:80 -n mlops

# Test qua port forward
curl http://localhost:8080/health

# Kiểm tra model info endpoint
curl http://[LOAD_BALANCER_URL]/model/info

# Kiểm tra model metrics từ SageMaker Registry  
curl http://[LOAD_BALANCER_URL]/model/metrics

# Expected response: Accuracy 84.7%, F1-Score 83.2% từ Registry

# Load test với data format đúng
for i in {1..100}; do
  curl -X POST http://[LOAD_BALANCER_URL]/predict \
    -H "Content-Type: application/json" \
    -d '{"BASKET_SIZE":"M","BASKET_TYPE":"MIXED","STORE_REGION":"LONDON","STORE_FORMAT":"LS","SPEND":125.50,"QUANTITY":3,"PROD_CODE_20":"FOOD","PROD_CODE_30":"FRESH"}' &
done

# Theo dõi HPA scaling
kubectl get hpa retail-api-hpa -n mlops -w

# Theo dõi pods được scale up (từ 2 → max 5)
kubectl get pods -n mlops -w

# Xóa tất cả resources trong namespace mlops
kubectl delete namespace mlops

# Hoặc xóa từng resource riêng lẻ
kubectl delete deployment retail-api -n mlops
kubectl delete service retail-api-service -n mlops
kubectl delete hpa retail-api-hpa -n mlops
kubectl delete serviceaccount retail-api-sa -n mlops

# Kiểm tra LoadBalancer đã bị xóa
aws elbv2 describe-load-balancers --query 'LoadBalancers[?contains(LoadBalancerName, `k8s-mlops`)].LoadBalancerArn'

# List images trong repository
aws ecr describe-images --repository-name mlops/retail-api --region ap-southeast-1

# Xóa specific image tag
aws ecr batch-delete-image \
  --repository-name mlops/retail-api \
  --image-ids imageTag=v3 \
  --region ap-southeast-1

# Xóa tất cả images
aws ecr batch-delete-image \
  --repository-name mlops/retail-api \
  --image-ids "$(aws ecr describe-images --repository-name mlops/retail-api --region ap-southeast-1 --query 'imageDetails[].imageDigest' --output text | tr '\t' '\n' | sed 's/.*/imageDigest=&/')" \
  --region ap-southeast-1

# Kiểm tra không còn pods nào
kubectl get pods -n mlops

# Kiểm tra không còn services nào
kubectl get svc -n mlops

# Kiểm tra LoadBalancer đã bị terminate
aws elbv2 describe-load-balancers --query 'LoadBalancers[?contains(LoadBalancerName, `k8s-mlops`)]'

# Example PVC for model storage
apiVersion: v1
kind: PersistentVolumeClaim
metadata:
  name: model-storage
  namespace: mlops
spec:
  accessModes:
    - ReadWriteOnce
  resources:
    requests:
      storage: 10Gi
  storageClassName: gp3

resources:
  requests:
    memory: "256Mi"    # Start smaller
    cpu: "100m"        # Minimal CPU request
  limits:
    memory: "512Mi"    # Reasonable limit
    cpu: "250m"        # Allow bursting

# Node affinity for cost optimization
affinity:
  nodeAffinity:
    preferredDuringSchedulingIgnoredDuringExecution:
    - weight: 100
      preference:
        matchExpressions:
        - key: kubernetes.io/instance-type
          operator: In
          values: ["t3.micro", "t3.small"]  # Prefer cheaper instances

# Use single ALB for multiple services
apiVersion: networking.k8s.io/v1
kind: Ingress
metadata:
  name: shared-alb
  annotations:
    kubernetes.io/ingress.class: alb
spec:
  rules:
  - http:
      paths:
      - path: /api/v1/*
        backend:
          service:
            name: retail-api-service
            port:
              number: 80
      - path: /admin/*
        backend:
          service:
            name: admin-service
            port:
              number: 80

# Monitor pod resource usage
kubectl top pods -n mlops

# Check actual vs requested resources
kubectl describe pod <pod-name> -n mlops

# Monitor HPA behavior
kubectl get hpa -w -n mlops

# Check LoadBalancer usage
aws elbv2 describe-load-balancers --names <alb-name>

# ELB costs
aws ce get-cost-and-usage \
  --time-period Start=2024-01-01,End=2024-01-31 \
  --granularity MONTHLY \
  --metrics BlendedCost \
  --group-by Type=DIMENSION,Key=SERVICE \
  --filter '{"Dimensions":{"Key":"SERVICE","Values":["Amazon Elastic Load Balancing"]}}'

# EC2 costs for nodes
aws ce get-cost-and-usage \
  --time-period Start=2024-01-01,End=2024-01-31 \
  --granularity MONTHLY \
  --metrics BlendedCost \
  --group-by Type=DIMENSION,Key=INSTANCE_TYPE