Managed Kubernetes or self-managed? How to weigh the choice
With managed Kubernetes the provider runs the control plane; you remain responsible for workloads, updates and security.
With managed Kubernetes the provider runs the control plane; when you self-manage, you do everything yourself. Managed reduces management work, but you remain responsible for workloads, access management and updates. Self-managing gives more control and requires considerably more expertise and time.
Who is it for: DevOps teams and technical buyers choosing between a managed Kubernetes service and their own cluster.
What does the provider manage and what do you keep doing yourself?
Kubernetes consists of a control plane (the decision-makers, including the database holding the cluster state) and worker nodes where you run your applications. With managed Kubernetes, the provider runs and maintains the control plane. The worker nodes can also be managed, depending on the offering.
You remain responsible for what runs in the cluster: container images, permissions, network policy, secrets, resource limits and monitoring. The shared responsibility model differs per provider. Ask for it explicitly, so you know who picks up what in an incident.
What does self-managing offer, and what does it cost?
Your own cluster gives you a choice of versions, network plugins, storage solutions and location, including on your own hardware or in a particular legal environment. That is relevant if you have strict requirements on data location or need non-standard hardware.
On the other hand, you take care of updates, certificates, backup of the cluster state, high availability of the control plane and security patches yourself. Include those hours in your cost comparison: staff are often the largest item. A formula helps: total = infrastructure + management hours x hourly cost + cost of an incident x probability.
Which questions do you ask a provider?
A managed offering is only an advantage if it matches your requirements. Ask concrete questions and put the answers side by side.
- How long are Kubernetes versions supported and who carries out upgrades?
- Is the control plane set up redundantly and what does the agreement say about that?
- Which custom extensions and network options may I use?
- Where are the cluster and data located, and who has access to them?
- How do I export my configuration if I want to switch?
How do you limit dependence on a provider?
Keep your manifests and infrastructure description standard and avoid features that exist only with one provider, unless the gain is clear. Periodically test a reinstall on another cluster. That keeps switching feasible, even if you choose managed.
Responsibilities compared
| Component | Managed | Self-managed |
|---|---|---|
| Control plane | Provider | You |
| Platform updates | Shared, provider often carries them out | You plan, test and carry them out |
| Workloads and images | You | You |
| Network and storage choice | Limited to the offering | Free choice |
| Required expertise | Using Kubernetes | Using and managing Kubernetes |
| Switching | Depends on the extra features used | Depends on your own set-up |
Who is it not ideal for?
- Managed is less suitable if you need your own hardware or a specific location that the provider does not offer.
- Self-managing is less suitable for small teams without cluster management experience and with limited availability outside office hours.
- If you only run a few applications, it may be better to first consider whether you need Kubernetes at all.
Frequently asked questions
Is managed Kubernetes less secure?
Not by definition. Security is a shared task; the provider takes care of the control plane, you take care of configuration and workloads. Ask what the provider can demonstrate about its part.
Can I later switch from self-managed to managed?
That is possible if your configuration is standard. Plan a migration with a parallel cluster and switch traffic over gradually.
What are worker nodes managed by the provider?
These are servers for which the provider takes care of the operating system and patches, so you can focus on the containers.
What about monitoring?
Monitoring of applications remains your task. Ask which data about the platform itself is visible to you.
