Skip to main content
Use this guide when HPC Verification tests don’t start in a CoreWeave Kubernetes Service (CKS) cluster. A controller that starts successfully doesn’t confirm that a GPU test ran. The test still requires an idle Node with all GPUs available. Check the verification result in the Node Details dashboard.

​
The verification controller Pod stays Pending

If no tests start, check the controller Pod before changing Node taints. You need kubectl access to the cluster and permission to list Pods, get Pod details, and list events in the cw-hpc-verification Namespace.
1

List the verification Pods

Look for a Pod with -ctrl- in its name that remains Pending.
2

Inspect the controller Pod's scheduling events

Replace [CONTROLLER-POD-NAME] with the controller Pod’s name:
In Events, look for FailedScheduling messages that identify untolerated taints or unavailable resources.
3

Contact Support if a customer-applied taint blocks the controller

If a customer-applied NoSchedule taint blocks the controller, contact CoreWeave Support to check the controller configuration. Include the following details:
  • Cluster name.
  • Controller Pod name.
  • Scheduling events.
  • Approximate time of the verification request.
Existing workflows can retain an older controller configuration even after CoreWeave updates the workflow template.
Leave CoreWeave-managed taints unchanged. Don’t delete verification workflows or Pods to retry a health check. Deleting an existing request doesn’t guarantee a replacement request. Administrator
Last modified on November 3, 2026