New nodes don't have parallel image pull configured at the kubelet level

[juanmaia]

Version

Karpenter Version: v0.0.0 (command doesn't work for me) but probably latest.

Kubernetes Version: v1.32.3

Expected Behavior

Images should be pulled in parallel in nodes created by Karpenter. Basically no waiting time should be seen for image pulling.

Actual Behavior

Images are being pulled in series in nodes created by Karpenter, and in parallel for nodes created manually or by cluster autoscaler (VMSS).

When pods in a new node is pulling images, I'm seeing this message often:
Successfully pulled image "foo:bar" in 2.786s (10m16.315s including waiting). Image size: 40538901 bytes.

The waiting part made me realize that parallel image pulling is not set by default. When inspecting the kubelet config for nodes that Karpenter created, I don't see the config:
"serializeImagePulls"

And while inspecting nodes from regular node pools (VMSS) I do see that it's set to false:

 "serializeImagePulls": false

Steps to Reproduce the Problem

In a new node created by Karpenter, check in the kubelet config if it has:

 "serializeImagePulls": false

Resource Specs and Logs

I'm using NAP, so no access to logs.

Community Note

• Please vote on this issue by adding a 👍 reaction to the original issue to help the community and maintainers prioritize this request
• Please do not leave "+1" or "me too" comments, they generate extra noise for issue followers and do not help prioritize the request
• If you are interested in working on this issue or have submitted a pull request, please leave a comment