kubemark

Testing how Kubernetes manages thousands of machines can require more hardware than a development team has. Kubemark simulated worker nodes, letting engineers measure the services that accept requests, schedule work and track cluster state. Its simulated nodes did not run real application containers, so it could answer control-plane scaling questions without measuring application performance. This entry explains the removed upstream implementation and the separate provider still encountered in existing test environments.

Current guidance

On October 6, 2026, Kubernetes merged removal of Kubemark from its master branch; the removal explanation calls it unmaintained. This is explicit lifecycle evidence, not merely a missing URL. The change does not erase Kubemark from older release tags.

The separate Cluster API Provider Kubemark still documents fake machines whose nodes only pretend to run scheduled pods, with its own version matrix and an explicit alpha warning. That provider is not evidence that Kubernetes continues to maintain the removed implementation. For an existing CAPK setup, pin the provider, Cluster API, Kubernetes and Kubemark image combination documented together, and establish how its Kubemark dependency is maintained before upgrading.

Hollow-node tests address control-plane questions such as API request latency, object churn, scheduler behavior and reconciliation under simulated node counts. Document the control-plane hardware and load generator, and keep tests isolated from production. Simulated pod counts do not demonstrate application throughput, storage performance or network capacity; those need real worker-node benchmarks. This review has not run a scale test or verified continued compatibility of the external provider after the upstream removal.