filter non-ASG nodes by tag

[universam1]

Describe the feature

Karpenter nodes are not bound to a Autoscaling Group. Filter on tags via check-asg-tag-before-draining is scoped to ASG only, so it is not possible to filter nodes such nodes based on tags.

aws-node-termination-handler/pkg/monitor/sqsevent/sqs-monitor.go

Lines 356 to 365 in 2048e92

	if m.CheckIfManaged {
	if nodeInfo.AsgName == "" {
	// If ASG tags are not propagated we might need to use the API
	// to retrieve the ASG name
	nodeInfo.AsgName, err = m.retrieveAutoScalingGroupName(nodeInfo.InstanceID)
	if err != nil {
	return nil, fmt.Errorf("unable to retrieve AutoScaling group: %w", err)
	}
	}
	if nodeInfo.Tags[m.ManagedAsgTag] == "" {

Is the feature request related to a problem?

AFAIK pre - filtering the Spot ITN Events is not possible because EC2 Spot Instance Interruption Warning do not contain tags of the instance where we could filter the rule on

I have several clusters in one account and thus I am missing an option in NTH to filter for cluster respective nodes only.

Describe alternatives you've considered

none

[AustinSiu]

Hi @universam1! Can you help confirm the ask? It sounds like you want the ability for NTH to ignore nodes that are not associated with an ASG, using the same ManagedAsgTag we use today for nodes in ASGs (name might need to be revisited...).

I have to bound NTH to the cluster respective nodes only.

Can you help clarify what you mean, assuming this is your current workaround to the issue?

[universam1]

@AustinSiu Thank you for asking - actually the opposite 😄

I want NTH to manage non-ASG nodes that are created by Karpenter. Plus I need to filter based on tags to identify the cluster-nodes.
However, ManagedAsgTag filter is not applicable for non-ASG nodes, see above.

assuming this is your current workaround to the issue?

Sorry if I was unclear, I have no workaround for this problem.

[bwagner5]

👋 @universam1 you are right and this is a regression. We used to have an AssumeASGTagPropagation configuration that would result in no ASG lookup, but that was removed recently. We should add that back.

37e9899#r78210273

[pangorgo]

I've just encountered a similar issue. I just wanted to post a potential workaround to tag your instance with aws:autoscaling:groupName, but I forgot aws: prefix is reserved and the user can't assign such a tag.

We are looking forward to the solution. @bwagner5 not sure what Priority:Critical means. Can you shed some light on a possible timeframe for a resolution? Not sure if we should switch to IMDS or just wait for the fix, so any estimation (days, weeks, months?) would be very useful.

[bwagner5]

@pangorgo likely a week or so to get the change in and tested. A current workaround is to rollback to v1.16.2

[pangorgo]

Thanks @bwagner5 I've just tested it with version v1.16.2 and it works 🎉
I just had to add ASSUME_ASG_TAG_PROPAGATION=true

[snay2]

This has been released in NTH app version v1.17.0. Helm chart version v0.19.0.