/Kubernetes on AWS
Kubernetes on AWS with Amazon EKS, ready for production
We design, migrate and operate Amazon EKS clusters for teams that genuinely need Kubernetes. Scaling with Karpenter, GitOps deployments with Argo CD and Helm, pod-level security, observability and planned upgrades, all in Terraform and in your repository: no magic, and no clusters nobody dares to touch.
/Diagnosis
Signs it is time
A cluster nobody wants to upgrade
Your EKS version is nearing the end of standard support, add-ons are outdated and every upgrade gets postponed because nobody knows what might break.
Expensive, underused nodes
Oversized node groups, untuned requests and little use of Spot or Graviton mean cluster costs grow faster than traffic.
Manual deploys with kubectl
Changes are applied by hand or with scripts, there is no source of truth in Git, and the real cluster state drifts from what the manifests say.
Many teams, one cluster
Several teams share the cluster without properly isolated namespaces, quotas, network policies or per-service permissions, so one team's mistake hits everyone.
/What we deliver
Concrete deliverables, in your repository
EKS architecture
VPC design, clusters per environment, managed add-ons, a node strategy with Karpenter or EKS Auto Mode, and a multi-tenancy model.
Clusters in Terraform
Clusters, node pools, IAM and add-ons defined as code, reproducible in every environment and versioned in your repository.
GitOps with Argo CD and Helm
A deployment repository with Helm charts or Kustomize, Argo CD sync, promotion between environments and rollback from Git.
Cluster security
EKS Pod Identity, access entries, network policies, Pod Security Standards, image scanning in ECR and secrets in AWS Secrets Manager.
Upgrade plan and execution
A per-version upgrade runbook, deprecated API review, add-on upgrades and a blue/green cluster strategy when it makes sense.
/How we work
Our process
Cluster assessment
We review versions, add-ons, nodes, costs, security and deployment practices with read-only access, or your requirements if you are starting from scratch.
Design and roadmap
We define the target architecture, confirm whether EKS is the right choice over ECS and prioritize changes by risk and impact.
GitOps implementation
We build or refactor in Terraform, install Argo CD and migrate workloads in waves, validating at every step.
Ongoing operations
Regular upgrades, capacity tuning, cost reviews and cluster incident support, on an ongoing basis if you need it.
Kubernetes on AWS: when EKS is the right call
Kubernetes solves a lot, and it costs a lot to operate. Before building a cluster, we check whether Amazon ECS would solve the problem with less effort. EKS makes sense when there are many services and teams, when you need Kubernetes-native tooling such as operators, a service mesh or ML tooling, or when portability across environments is a real requirement.
If the answer is EKS, the next step is deciding how much you manage yourself. EKS Auto Mode hands compute, scaling and several critical add-ons to AWS, in exchange for less customization. A cluster with Karpenter and managed add-ons gives you more control. We document the trade-off with estimated operating costs and review it with you before building.
Scaling and cost with Karpenter, Spot and Graviton
Karpenter replaces fixed node groups with a provisioner that picks the right instance type for pending pods and consolidates nodes when there is spare capacity. Configured with NodePools that mix Spot, On-Demand and Graviton, it cuts waste without sacrificing stability.
Real savings also depend on the applications. We tune requests and limits with usage data, configure the Horizontal Pod Autoscaler or KEDA to scale on business metrics, and define PodDisruptionBudgets so consolidation does not hurt availability. With per-namespace tags, each team's cost becomes visible.
GitOps with Argo CD, Helm and an optional service mesh
In a well-run cluster, Git is the single source of truth. Argo CD continuously syncs the desired state, detects drift and lets you roll back with a commit. Services are packaged with Helm or Kustomize, and promotion between environments follows the same pull request flow as application code.
A service mesh like Istio, in ambient mode or with sidecars, brings mTLS between services, traffic policies and detailed telemetry. It also adds complexity. We introduce it only when there is a concrete need, such as service-to-service encryption for compliance or advanced canary deployments, never as part of a standard bundle.
Zero-downtime EKS upgrades
Each Kubernetes version on EKS has a standard support window and then moves to extended support, which carries an additional per-cluster fee. Postponing upgrades piles up both risk and spend. So we set an upgrade cadence with a per-version runbook: deprecated API review, add-on compatibility, control plane upgrade and node rotation.
For major changes or heavily customized clusters, we use a blue/green strategy: we stand up a new cluster in Terraform, sync the applications with Argo CD and shift traffic gradually. If something fails, traffic goes back to the previous cluster, which stays up until the migration is closed out.
/FAQ
Kubernetes on Amazon EKS: frequently asked questions
Another question? Write to hello@nubent.com and an engineer will answer.
What is the difference between Amazon EKS and Amazon ECS?
Both run containers on AWS. ECS is AWS's own orchestrator, simpler to operate and tightly integrated with other AWS services. EKS is managed Kubernetes, with its standard API, tooling and portability, but it demands more expertise and maintenance. For a handful of services ECS is usually enough; for large platforms with many teams, EKS wins.
What is Karpenter and why use it on EKS?
Karpenter is an open source node provisioner that launches instances based on what pending pods actually need, choosing type, size and purchase option. It also consolidates nodes when there is spare capacity. Compared to traditional node groups, it scales faster and makes better use of Spot and Graviton, which helps lower cluster costs.
How often should you upgrade an EKS cluster?
Kubernetes ships new versions several times a year, and EKS keeps each one in standard support for a limited period, followed by extended support at an extra cost. The recommended approach is to plan upgrades regularly, one version at a time, with a runbook tested in lower environments before touching production.
Should we use EKS Auto Mode?
EKS Auto Mode is a good option if you want AWS to manage compute, scaling, load balancing and some add-ons, and you prefer less operational work. If you need very specific node configurations, custom AMIs or fine-grained control over every component, a cluster with Karpenter and managed add-ons may fit better. We evaluate it against your requirements.
Can you migrate our applications to Kubernetes on AWS?
Yes. We containerize applications if needed, write the Helm charts, set up Argo CD and migrate in waves with performance validation and a planned rollback. If you are coming from another Kubernetes provider or a self-managed cluster, we adapt ingress, storage, identity and observability to AWS-native services.
Related services
Platform Engineering
An internal developer platform on AWS with golden paths, Backstage and Terraform self-service, for autonomous teams that ship quickly and safely.
Learn moreAWS Application Development
Backends, APIs, frontends on CloudFront and Amplify, databases and CI/CD: cloud-native software built on AWS and ready to run in production.
Learn moreObservability and SRE
CloudWatch, OpenTelemetry and Grafana, SLOs with error budgets, actionable alerts, incident management and postmortems to run AWS on data.
Learn moreDevOps as a Service
An ongoing external DevOps team for your AWS infrastructure: Terraform, CI/CD, security, costs and on-call coverage, on a monthly subscription.
Learn moreNext step
Let's talk about your Kubernetes on Amazon EKS project
30 minutes with an engineer. Concrete next steps, whether or not you work with us.