Choosing: ECS or EKS?
Team: “Let’s go K8s. It’s the standard. Portability. Resume experience.”
Three months later, we calculate the cost.
You get it, right?
Trendy, modern, all that stuff…
Infrastructure on AWS. Microservices, containers!
This is normal.
Two options basically:
- ECS Fargate: AWS-specific, managed
- EKS: Kubernetes, “industry standard” (you feel the sarcasm, right?)
People choose EKS. Arguments:
- “Everyone knows K8s, easier to hire”
- “Portability: move to GCP in a day”
- “Flexibility: any deployment options”
Sounds reasonable? Let’s count.
After 3 Months
EKS control plane: $73/month. Fixed.
Minimum 3 nodes for stability. Even for single service on t3.small.
What we have:
- Ingress: NGINX or ALB Controller. Setup, updates, monitoring.
- Monitoring: Prometheus + Grafana or buy Datadog.
- Logging: Fluentd/Fluent Bit → CloudWatch/S3.
- RBAC: Service accounts, IRSA, namespace isolation.
- DNS: CoreDNS configuration, debugging.
- Updates: Cluster, nodes, addons, CNI. Every quarter.
In ECS:
- Ingress: ALB, done.
- Monitoring: CloudWatch out of box.
- Logging: one parameter.
- RBAC: IAM task role, done.
- DNS: Route 53, works.
- Updates: AWS does it.
Engineer time: 20-30% on cluster maintenance. Not features.
About “Portability”
“Move to GCP tomorrow” sounds nice. Reality:
Managed K8s differs:
- EKS: AWS Load Balancer Controller
- GKE: Google Cloud Load Balancer
- Bare metal: different Ingress
Pure K8s for portability:
- No managed LBs → more complex
- No IAM integration → another auth layer
- No provider storage → own provisioner
1.5-2x more complexity.
Questions:
- Really migrating? When? Why not now?
- What criteria trigger the move?
- Will self-limitation pay off?
Most companies never migrate. Vendor lock-in scarier in theory.
Analogy
Saw a story: boss decided to work directly with factories in China to avoid middleman markup (15%).
Container sat in port for one week due to forgotten customs documents. Penalties cost as much as two months of supplies with a middleman.
Same with EKS:
You “avoid” vendor lock-in, get “flexibility”. But you pay with engineer time, stability, development speed.
The middleman (in this case AWS ECS) takes a “markup” but solves problems for you.
ECS Advantages
- Deploy in 10 minutes
- CloudWatch monitoring
- Autoscaling in 5 minutes
- AWS handles updates
Cost: only containers. No control plane, no “minimum 3 nodes”.
Limitations: AWS lock-in, less flexibility.
Takeaway
“K8s is the standard” ≠ argument.
ECS: simpler, cheaper and faster in dev. EKS: flexible but expensive to maintain.
Total cost ($ + team time) prevents surprises.
“Simpler” often beats “correct but complex”.
P.S. “Flexibility” and “portability” sound good in presentations. Reality: 20-30% team time. Worth it?