DevOps Software Configuration Management

Supporting Applications Across Customer-Managed Kubernetes Clusters

Running an application in a Kubernetes cluster managed by its vendor is different from supporting it across clusters controlled by customers. In customer-managed environments, the application provider may not set the Kubernetes version, cluster policies, networking, storage, or upgrade schedule. Yet customers still expect help when an installation fails, an upgrade causes problems, or a workload behaves unexpectedly.

In guidance published November 19, 2026, Adam Zahorscak of Fairwinds argues that providers need to treat this variation as a normal part of their support model. The goal is not to make every cluster identical, but to define the conditions under which the application is supported and give teams a consistent way to diagnose issues.

Document the cluster requirements

Two customers may both report that they run Kubernetes while using substantially different platforms. One might use a managed public-cloud service; another might operate its own infrastructure. Their clusters can differ in Kubernetes versions, networking, ingress controllers, storage classes, image registries, security policies, observability tools, and access controls.

Those differences can prevent an otherwise valid deployment from working. A customer’s admission policy might reject a required permission, storage may not meet the application’s needs, or the cluster may be on a Kubernetes version the provider has not tested or no longer supports. Without clear requirements, customers can encounter these constraints only after a failed installation or during an incident.

Providers should document the application’s supported Kubernetes versions and distributions, required permissions and cluster capabilities, and expectations for networking, ingress, DNS, certificates, secrets, and storage. Resource requirements also matter. So does an escalation checklist that tells customers which logs, events, metrics, and configuration details support teams need. Zahorscak recommends making diagnostic commands available where possible, so customers can collect useful evidence without guessing what to run.

These details establish a support boundary: what the application provider can investigate and what depends on the customer’s platform or its configuration. A defined operating model can keep each deployment from becoming a fresh discovery exercise.

Set expectations for Kubernetes upgrades

Customer organizations often cannot upgrade Kubernetes as quickly as an application roadmap might assume. They may need to coordinate maintenance windows, compliance reviews, internal dependencies, and validation of cluster add-ons, integrations, and workloads. Even when an upgrade is planned, the platform team may need time to establish that the broader environment remains compatible.

Compatibility also changes over time. Kubernetes APIs can be deprecated and removed, and compatibility can vary across control-plane and node components. The Kubernetes project supports only its three most recent minor releases, according to Zahorscak’s guidance. Application providers therefore need a policy that identifies currently supported versions, explains when support will end, and gives customers notice before a version is dropped.

The policy should also address what happens when a new application release requires a newer Kubernetes capability. If a customer cannot upgrade immediately, providers should explain what is incompatible and why, and document any available workaround. Treating “upgrade Kubernetes” as the only response can leave customers without a practical path forward.

Put Kubernetes knowledge in customer support

Support teams do not need every employee to operate production clusters, but customer-facing staff need enough Kubernetes expertise to distinguish application defects from deployment or cluster issues. That distinction helps identify which team must act and what information is needed.

For example, a pod in Pending status could reflect insufficient capacity, resource requests that do not fit available nodes, scheduling constraints, or a policy that blocks placement. An image pull failure could stem from registry access, credentials, network connectivity, or image policy enforcement. These symptoms point to different investigations and may involve the application owner, platform team, security group, or infrastructure team.

Customer-facing specialists should request relevant diagnostics early, review Kubernetes events and pod status, and examine appropriate logs. They also need to recognize when platform, security, networking, or infrastructure teams must participate, and explain the issue in terms those stakeholders can act on. When engineering escalation is necessary, the report should include enough environmental context to make the investigation useful.

This coordination matters because the person who opens a support case may not own the cluster. Application teams, platform teams, security groups, and infrastructure teams can have separate responsibilities for deployment, policies, networking, storage, and identity. Support should help move the investigation between those groups rather than assuming one customer contact can resolve every cluster-level issue.

Make variation part of the operating model

Fairwinds’ guidance centers on preparation: define supported environments, document requirements, test the Kubernetes versions and deployment conditions the provider commits to supporting, and maintain a clear upgrade and deprecation policy. Kubernetes expertise should be available through onboarding, deployment, escalation, and ongoing support.

When a provider manages the Kubernetes layer alongside the customer, troubleshooting may take place against a more clearly defined operating foundation. For providers supporting customer-managed clusters, however, the practical requirement is to account for environmental differences from the start. Clear boundaries and diagnostic processes help customers understand what is supported and give support teams a more reliable basis for resolving problems.