Containers e serverless em arquitetura enterprise

Lambda:
  Runtime managed, ideal event-driven, cold start milisegundos
  Limites: 15min, 10GB RAM, /tmp 10GB
  Container image support (até 10GB)
  SnapStart pra Java reduzir cold start

ECS Fargate:
  Containers sem EC2 (serverless compute)
  Task definitions, services, autoscaling
  Ideal long-running APIs, batch jobs

ECS on EC2:
  Mais controle (GPU, custom AMI, Spot)
  Mais operational overhead

EKS Fargate:
  Kubernetes sem gerir nodes (limites: sem DaemonSets,
  alguns operators exigem EC2)

EKS on EC2 (+ Karpenter):
  K8s completo com scaling por Karpenter (fora do ASG)

App Runner:
  Git/ECR → container rodando com HTTPS, zero infra

Batch:
  Queue de jobs batch, escolhe compute (Fargate/EC2/Spot)
  automaticamente. Perfeito pra ML training, rendering

# Exemplo CDK: ECS Fargate service autoescalado
const cluster = new ecs.Cluster(this, 'C', { vpc });

const taskDef = new ecs.FargateTaskDefinition(this, 'T', {
  cpu: 512, memoryLimitMiB: 1024,
});
taskDef.addContainer('api', {
  image: ecs.ContainerImage.fromRegistry('111.dkr.ecr.us-east-1.amazonaws.com/api:v1.2'),
  logging: ecs.LogDrivers.awsLogs({ streamPrefix: 'api' }),
  portMappings: [{ containerPort: 8080 }],
});

const svc = new ecsPatterns.ApplicationLoadBalancedFargateService(this, 'Svc', {
  cluster, taskDefinition: taskDef,
  desiredCount: 2, minHealthyPercent: 100,
  circuitBreaker: { rollback: true },
});

svc.service.autoScaleTaskCount({ minCapacity: 2, maxCapacity: 50 })
  .scaleOnCpuUtilization('cpu', { targetUtilizationPercent: 60 });