Kubernetes
Cluster design, upgrades, security hardening, GitOps and monitoring for Kubernetes on AWS, Azure, Google Cloud, Hetzner or your own servers. Your developers release faster, and upgrades become routine work.
We reply within one business day.
Any cloud
EKS, AKS, GKE, Hetzner, on-prem
GitOps
every change through Git and Argo CD
Assessment first
a clear plan before anything is built
Commit → pipeline → GitOps → cluster, with monitoring on every step
Commit → pipeline → GitOps → cluster, with monitoring on every step
Works with
Kubernetes
Helm
Argo CD
Rancher / RKE2
Docker
Prometheus
Grafana
Terraform / OpenTofu
AWS
Google Cloud
Hetzner
In short
We design, build, upgrade and support Kubernetes clusters: managed services (EKS, AKS, GKE) or self-managed, including vanilla upstream Kubernetes with kubeadm on bare metal or VMs, k3s and RKE2/Rancher. Work covers networking (Cilium, Calico, Gateway API), service mesh with Istio or Linkerd, autoscaling with HPA, VPA, KEDA and Cluster Autoscaler or Karpenter, storage (Rook-Ceph, Longhorn), RBAC and policy hardening, GitOps with Argo CD, monitoring, backups and tested version upgrades. We can hand over to your team or run the platform for you.
Services
Managed or self-hosted clusters with networking, ingress, storage and separate environments, built from Terraform.
Upstream Kubernetes with kubeadm on your own servers or VMs: HA control plane, etcd backups, Cilium or Calico, MetalLB, and Rook-Ceph or Longhorn storage. No vendor distribution required.
Horizontal Pod Autoscaler on CPU, memory or custom metrics, Vertical Pod Autoscaler for right-sized requests, KEDA for queue- and event-driven scaling, and Cluster Autoscaler or Karpenter for nodes.
Mutual TLS between services, traffic splitting for canary releases, retries and timeouts, and service-level observability. We also tell you when a mesh is overkill.
NGINX, Traefik or Envoy-based gateways, Gateway API routes, cert-manager for TLS and external-dns.
Tested Kubernetes version upgrades, API deprecation fixes and add-on updates, with a rollback plan.
RBAC, network policies, Pod Security Standards, Kyverno or OPA Gatekeeper policies, image scanning, External Secrets with Vault, and audit logging.
Velero backups, persistent volumes, and databases on Kubernetes done safely, or kept outside when that is wiser.
Right-sized requests and limits, node pools and spot or cheaper nodes, so you stop paying for idle capacity.
We run the platform: upgrades, patching, monitoring and on-call, with response times in writing.
Examples
Applications packaged with Helm, deployed through GitOps, with staging and production that behave the same.
Clusters several versions behind brought up to date one step at a time, without downtime for users.
Hardening and documentation for ISO 27001, SOC 2 or a customer security questionnaire.
The same workloads on simpler, cheaper infrastructure, with monitoring and backups from day one.
Workers scale on queue length, web pods on requests per second, and nodes appear only while they are needed.
Every service-to-service call encrypted and authorised, plus canary releases and traffic metrics per service.
How to start
Architecture, security and cost review
Autoscaling and upgrade path
Prioritised fix list
About 1 week
Cluster(s) built from code
GitOps with Argo CD
Monitoring and backups
Handover and training
Upgrades and patching
Monitoring and alerting
Capacity reviews
Response times in writing
Process
What runs where, what hurts, and what the cluster costs.
In phases, from code, with staging first.
Training and runbooks, or we operate it for you.
Why Kubernetes
Kubernetes is a standard way to run containers across any infrastructure. Done well, it makes releases routine and capacity elastic. Done badly, it is expensive complexity, which is why we start with an assessment.
Failed containers restart, unhealthy ones are replaced and traffic only reaches healthy instances, without someone being paged.
Pods scale out on CPU, memory or business metrics (HPA, KEDA), right-size themselves (VPA) and nodes are added or removed automatically.
Rolling, blue-green and canary deployments with instant rollback; with a service mesh you can shift 1% of traffic at a time.
The same manifests run on a laptop, AWS, Azure, Hetzner or your own servers, which keeps you portable and avoids lock-in.
Many services packed onto fewer machines with resource limits, so you pay for capacity you actually use.
Logging, monitoring, security policies and secrets handled the same way for every service, instead of per server.
An honest note before you book.
A handful of services with steady traffic: VMs or a PaaS can be simpler and cheaper.
No one to own the platform: unless we run it for you, someone has to keep it upgraded.
A single monolith that is not containerised yet: start with containers and CI/CD first.
FAQ
Straight answers. Ask us anything else on a call.
Not always. For a few services, VMs or a PaaS can be simpler and cheaper. We tell you which in the assessment, and design for where you will be in two years.
Upstream Kubernetes with kubeadm gives full control and no vendor dependency, at the cost of more operational work. k3s and RKE2 are lighter to run; OpenShift and Rancher add management layers. We pick by team size, compliance needs and who will operate it.
Only if you need mutual TLS everywhere, fine-grained traffic control or service-level telemetry across many services. For a few services, Linkerd or plain ingress plus network policies is simpler. We tell you which in the assessment.
HPA adds or removes pods when load changes; VPA adjusts how much CPU and memory each pod requests. They should not both act on CPU and memory for the same workload. A common setup is HPA (or KEDA) for scaling and VPA in recommendation mode to keep requests right-sized.
Managed control planes save effort on the big clouds. On Hetzner or your own hardware, k3s or RKE2 are reliable and much cheaper. We compare total cost including the time to run it.
Yes. Most of our Kubernetes work starts with an existing cluster that needs upgrading, hardening or tidying up.
Related
Next step
Send a few details, and pick up to 3 times if you would like a call. We reply within one business day.
You talk to the expert who would do the work
Written scope and price after the call
No obligation