A Kubernetes setup a small team can actually run

Written By: BrainSoft In DevOps

Kubernetes has a reputation for needing a dedicated team. For a lot of what people run on it, that reputation is earned by choice, not by the tool. A three-person team can operate a cluster comfortably if they decide early what they are not going to do.

Let someone else run the control plane

Managed Kubernetes from a cloud provider removes the part that is genuinely hard to operate: etcd, API server upgrades, certificate rotation. The cost is small next to one engineer-week spent recovering a self-managed control plane. Unless you have a firm requirement to run on your own hardware, take the managed option and spend the saved time on your application.

Pick a small set of add-ons and stop

A cluster needs a few things beyond the defaults. It does not need everything in the ecosystem. A workable baseline:

An ingress controller, so external traffic has one well-understood entry point.
Automatic TLS certificates.
A metrics stack and a way to read logs. Managed offerings are fine here too.
A secrets approach that is not "commit them to the repo".

Service mesh, multi-cluster federation, custom operators and GitOps controllers are all defensible later. On day one they are surface area you have to understand before you can debug anything.

Make deploys reproducible early

Keep your manifests in version control and render them with one tool, whether that is Helm, Kustomize or plain templating. The goal is that anyone on the team can look at the repo and know exactly what is running. Apply from CI, not from a laptop.

Set requests and limits, and mean them

Most cluster instability on small teams comes from workloads with no resource requests fighting for the same node. Set a request that reflects normal use and a limit that catches a runaway process. Add a couple of pod disruption budgets for anything that should not drop to zero during a node upgrade. That is most of what keeps a small cluster calm.

Know how you would rebuild it

Write down, and ideally script, the steps to recreate the cluster and redeploy every workload from scratch, then run through it once on a throwaway cluster. A team that has done this treats the cluster as replaceable infrastructure. A team that has not treats it as a pet, and the reluctance to touch it grows over time.

Frequently asked questions

Is Kubernetes overkill for a small team?

Not if you already run more than a couple of services that need to scale or deploy independently. For one or two applications, a simpler platform is usually the better choice.

Should we run Kubernetes ourselves or use a managed service?

Use managed unless you have a firm reason to self-host. Managed control planes remove the hardest part of the operational burden for a small team.

How many people do you need to run a small cluster?

One engineer who understands it well, with clear runbooks, is usually enough. The real risk is knowledge concentrated in one head, not a lack of headcount.

We help small teams set up clusters they can actually run themselves — see our DevOps services or talk to us about your setup.

#DevOps