NVIDIA / NVIDIA/gpu-operator

[Feature]: Ignore already cordoned nodes during driver upgrades

Open
#2,820 0 comments 0 reactions 1 assignee View on GitHub

@JunAr7112 is already working on this.

Since Sep 10, 2026.

feature lifecycle/frozen needs-triage
Dominant language
Go
Stars
2.9k
Forks
552
Avg merge
2d 4h
Merged PRs (30d)
90

Description

Requestor: @kvalliyurnatt

Summary

Ignore nodes that are already cordoned (not by any of the GPU operator components) during upgrades from upgrade controller/k8s-driver-manager

Motivation

If there is a node that has already been cordoned (not by any of the GPU operator components), then that means the node is not accepting any workloads at the moment, and there is no guarantee of the state that node is in, it could be down or in an unresponsive state. So it feels like the upgrade controller/k8s-driver-manager should ignore nodes that are already cordoned and not take any action on such nodes until they are uncordoned.

Proposal

We will need a way differentiate nodes that are cordoned by the GPU operator components, for which we could add an annotation on the nodes when the upgrade controller/k8s-driver-manager cordons a node(I believe the upgrade controller already does this) and then uncordon nodes based on that annotation. Then the upgrade controller/k8s-driver-manager can ignore nodes that are in a cordoned state but don't have the annotation on them.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.