Leftover Network Interfaces after scaling down

[mbevc1]

Version

Karpenter: v0.5.4

Kubernetes: v1.21

Expected Behavior

Scale up and down and Karpenter provisions and cleans up resources correctly.

Actual Behavior

When scaling down there are some Network Interfaces left unattached from removed Pods/Nodes.

Steps to Reproduce the Problem

Deploy Karpenter and demo inflate Deployment. Scale it up and then down to 0. Not always there seems to be leftover Network Interfaces unattached after scaling down.

Resource Specs and Logs

n/a

[ellistarn]

Hmm -- this looks like it might be an AWS VPC CNI bug. Can you file an issue over there: https://github.com/aws/amazon-vpc-cni-k8s

[suket22]

@MBevc - were you able to file an issue against the VPC CNI? The CNI is responsible for attaching, detaching and deleting ENIs on worker nodes I'm going to resolve this unless there's something else we can help with

[mbevc1]

Sorry, just opened the issue upstream as well: aws/amazon-vpc-cni-k8s#1825

[mbevc1]

Identified to be related to aws/amazon-vpc-cni-k8s#1223 upstream is working on. Sharing for prosperity here.

[ellistarn]

Thanks for following up. Feel free to reopen if you identify anything remaining in karpenter.