Quiz

B — Railway or Fly.io.

Railway and Fly.io can deploy a Dockerized Spring Boot application in minutes with zero infrastructure configuration. EKS requires days of setup (VPC, node groups, IAM roles, add-ons, ingress controllers); ECS Fargate still requires VPC, IAM task roles, ALB, and ECR configuration; Lightsail is simple but still needs an image pushed to a registry and a manual container service setup. When time-to-live is the constraint, a PaaS that builds straight from the repository wins.

B — ECS Fargate is serverless; EKS requires managing node groups and the full K8s API surface.

With ECS Fargate, AWS provisions and patches the underlying compute — you define a task (CPU, memory, container image, env vars) and AWS runs it. With EKS, you pay for the control plane (~$73/month) and manage node groups yourself: instance types, AMI updates, node draining during upgrades, and all Kubernetes add-ons. At 10 developers, the ECS operational simplicity typically outweighs the loss of the Kubernetes ecosystem.

C — False. EKS control plane + minimum node group exceeds the total cost of simpler alternatives.

The EKS control plane is $0.10/hour × 730 hours ≈ $73/month before a single container runs. A minimal node group (2 × t3.medium) adds about $61/month, and a load balancer and a small RDS database bring the class's cost model to roughly $180/month — against roughly $15–25/month for the same three services on Fly.io, Railway or a Lightsail VM (October 2026 list prices, us-east-1). (The course project uses EKS anyway because its goal is to learn Kubernetes.) Beyond cost, EKS requires significant operational knowledge (IAM roles for service accounts, VPC CNI networking, ingress controller setup) that distracts from learning the application-level concepts the course targets.

B — K3s is a lightweight, full-API-compatible Kubernetes distribution for on-prem, edge, or cost-sensitive deployments.

K3s runs the complete Kubernetes API in a binary under 100 MB, replacing etcd with SQLite by default and stripping out cloud-provider integrations. You would choose K3s over EKS when you need Kubernetes on bare-metal servers (a university data center, a factory floor, a Raspberry Pi cluster), when internet connectivity is unreliable, or when the ~$73/month EKS control plane fee is unjustifiable. The trade-off is that you own all operations — upgrades, backups, and HA setup — with no AWS managed support.

B — Amazon ECS Express Mode.

AWS closed App Runner to new customers on 30 April 2026 (existing customers keep it, without new features) and recommends ECS Express Mode as the replacement. Express Mode has no extra charge — you pay for the Fargate tasks, the load balancer and the other resources it creates — and because everything it creates is visible and editable in your account, the team can later "graduate" to full ECS Fargate without migrating. Unlike App Runner, it does not build from source: the CI pipeline must build and push the image to ECR first.

B — Deploy to additional regions; Fly.io's Anycast routing directs users to the nearest instance.

Fly.io assigns each application an Anycast IP address. When a European user connects, their request is routed by BGP to the nearest Fly edge node that runs your container — typically lhr (London) or ams (Amsterdam). Adding regions is one command, e.g. fly scale count 3 --region gru,lhr,ams. The trade-off: stateless services scale globally with zero effort, but stateful services (databases, sessions) require deliberate replication or global consistency strategies — the classic Fly Postgres is an app you operate yourself, and the newer Managed Postgres removes that work at a higher fixed price.

C — Private domain for Lightsail-to-Lightsail; VPC peering for RDS.

Lightsail does not use standard AWS VPC networking. Each container service gets one public endpoint plus a private domain <service-name>.service.local, reachable from other Lightsail resources in the same Region — enough for gateway → account. Resources outside Lightsail, such as an RDS instance in a regular VPC, are reachable only after enabling Lightsail VPC peering with the default VPC of the Region, a step many tutorials skip. There are no security groups or network policies to fine-tune either, which is one reason teams outgrow Lightsail.

B — The presence of a dedicated platform engineer is the decisive factor.

At 40 microservices, both ECS Fargate and EKS are technically capable. The real question is operational ownership. EKS requires someone who understands Kubernetes networking (CNI, services, ingress), manages node group upgrades, configures RBAC, installs and maintains add-ons (Karpenter, ArgoCD, external-secrets-operator), and responds to cluster incidents. Without a dedicated platform engineer, these tasks fall on feature developers, slowing product delivery. ECS Fargate covers most needs of a typical microservice system with no control-plane fee and no nodes to operate — and is the right answer for most teams at this scale unless GitOps, multi-tenancy, or cross-cloud portability are hard requirements.

B — Fixed costs dominate at small scale.

Three small Fargate tasks (0.25 vCPU / 0.5 GB) cost about $27/month; the ALB's hourly fee, capacity units and public IPv4 addresses cost about $30, and the smallest RDS instance about $14 — all billed by the hour whether or not requests arrive. EKS would add a $73 control plane and still need a load balancer. This is why fixed-fee platforms lose at three services and win at fifty, and why forgotten NAT gateways, load balancers and clusters are the classic surprise on a student's bill.

C — Stay, and define the migration triggers.

Each migration should be driven by a concrete pain point, not by speculation or résumé value. EKS would add a fixed control-plane fee, node operations and Kubernetes expertise the team does not need yet — hours of platform work that ship no feature. Keeping the application 12-factor makes a later move a deployment exercise rather than a rewrite, so waiting costs little.

C — The managed, highly available control plane.

K3s is a CNCF-certified distribution, so your Deployments, Services and Ingresses work unchanged. What EKS's $0.10/hour buys is a control plane that AWS runs across several Availability Zones, patches and backs up. With K3s on one VM, losing the VM loses the cluster and every workload on it; self-hosting moves the work to you, not just the cost away.