Cordoning scaled down nodes by Cluster Autoscaler

[Daniel Kłobuszewski]

Hi all,

Current behavior in Cluster Autoscaler is to taint nodes before scaling them down to prevent workloads from getting scheduled while the underlying VM is getting removed. There is a feature flag that additionally cordons (i.e. sets "unschedulable" bit on the node object) such nodes, which also covers pods with wildcard toleration. We are planning to flip that feature flag to true by default in CA 1.34 and eventually to remove it in a future release (issue, PR). If you have any concerns regarding this change, please comment on the linked issue.

[Michael McCune]

thanks for the /heads up/ Daniel!

peace o/

On Tue, Jun 3, 2025 at 10:09 AM 'Daniel Kłobuszewski' via Autoscaling Kubernetes <kubernetes-si...@googlegroups.com> wrote:

Hi all,

Current behavior in Cluster Autoscaler is to taint nodes before scaling them down to prevent workloads from getting scheduled while the underlying VM is getting removed. There is a feature flag that additionally cordons (i.e. sets "unschedulable" bit on the node object) such nodes, which also covers pods with wildcard toleration. We are planning to flip that feature flag to true by default in CA 1.34 and eventually to remove it in a future release (issue, PR). If you have any concerns regarding this change, please comment on the linked issue.

--
You received this message because you are subscribed to the Google Groups "Autoscaling Kubernetes" group.
To unsubscribe from this group and stop receiving emails from it, send an email to kubernetes-sig-auto...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/kubernetes-sig-autoscaling/CAD3gtR8nGgnuBNYGFAxu9R4MV%2BqL_tMydFoFkJ5qWRytNM%3DX7w%40mail.gmail.com.