The layers
What a cluster is actually made of
Each layer is a thing your team now owns. The point of the drawing is that the workload at the top is the reason you are here, and everything under it is the cost you took on to run it.
01
The workload
Your services, the reason the platform exists. Everything below is in service of keeping these running.
- Containers
- Deployments
02
Scheduling and networking
Where the cluster earns its keep — and where a misconfiguration is hardest to see until traffic finds it.
- Scheduler
- CNI
- Ingress
03
Security and policy
RBAC, network policy and admission control — the layer that turns "it runs" into "it is safe to run".
- RBAC
- Policy
04
Nodes and upgrades
The machines and the quarterly upgrade nobody enjoys, which is where an unmaintained cluster quietly becomes a liability.
- Node pools
- Upgrades
The step-back option
For some estates the honest recommendation is a managed container service or plain virtual machines. We say so when the workload does not earn the cluster.
A cluster nobody upgrades is not a platform; it is an unpatched host with extra steps.
Why teams call
The cluster someone set up and then left
The most common engagement is not a greenfield design. It is a cluster a departed engineer stood up, that works well enough that nobody touches it, is three versions behind on upgrades, and holds a security posture nobody has looked at since it launched. The value is not in rebuilding it; it is in making it safe, upgradable and operable by the team that has to live with it.
The second most common is a team that adopted Kubernetes because it was the default answer, and is now paying its operational tax on a workload that would run happily on a managed service. There the honest deliverable is a migration away from the cluster, and we would rather write that recommendation than sell an operations contract for complexity the estate never needed.
The failures that matter
What breaks a cluster in production, and when
A cluster looks healthy until one of these arrives, and each is an operational discipline rather than a configuration value.
A node runs out of resources under load
Graceful degradation rather than a cascade, because the limits were sized before the peak, not during it.
Requests, limits and autoscaling set from real usage, not guessed, with headroom that matches the traffic pattern.
An upgrade is overdue and now risky
A quarterly non-event, which is the only healthy state for cluster upgrades.
A staged upgrade path with a tested rollback, rehearsed in a copy of the cluster before it touches production.
A workload can reach more than it should
Blast radius contained by design, checked against a benchmark rather than assumed.
Network policy and RBAC scoped so a compromised pod cannot become a compromised cluster.
The one person who understood it leaves
A platform the team owns, rather than one that owns a single engineer.
A runbook, an on-call rotation and a handover built during the engagement, not promised after it.
Our role
What we set up and what we hand back
- 01
The assessment comes first
Before any build we say whether the cluster earns its cost for your estate. A recommendation to simplify is a valid, and sometimes the only honest, outcome.
- 02
Hardening with evidence
We harden to a recognised benchmark and keep the evidence, so a later audit or a customer security review has something to read rather than a claim to trust.
- 03
Operability, handed over
The runbook, the on-call model and the upgrade rehearsal stay with your team, because a platform you cannot operate without us is a dependency, not a capability.
Whether to run Kubernetes at all stays your decision; we give you the assessment to make it well, and if the answer is yes, a cluster your own team can operate and upgrade without us.
Common questions
What buyers ask
Would you ever tell us not to use Kubernetes?+
Often. For a handful of services with steady traffic, a managed container service or plain virtual machines carries a fraction of the operational cost and none of the upgrade burden. Kubernetes earns its complexity when you have many services, real scaling variance, or a platform team to run it. Where you do not, recommending it anyway would be selling you our comfort rather than your outcome.
Can you take over a cluster we already have?+
Yes, and it is the most common request. We start with a health and security review — version drift, RBAC, network policy, backup and the upgrade path — then agree a plan to get it current and keep it there. The deliverable includes the runbook and on-call model your team needs to operate it, because inheriting a cluster nobody documented is exactly the situation we are usually called to fix.
How do you keep the cluster compliant with local requirements?+
By hardening to a recognised benchmark and keeping the evidence, and by placing data where your obligations require. Kubernetes does not change what a regulator or a customer security review asks for; it changes where the controls live. We map the cluster’s controls to the requirement your organisation is held to and retain the proof, so the platform helps a review rather than becoming its hardest chapter.
In this cluster
Next step
Review your platform with us
A two-week assessment that ends with a recommendation to adopt, right-size or step back from Kubernetes, costed against your actual workload.

