Add documentation in using Queue Processor with Karpenter

[aldredb]

Describe the feature
I'm using NTH with Karpenter. Since Karpenter does not use ASG, I require documentation in using Queue Processor with Karpenter e.g. using ASSUME_ASG_TAG_PROPAGATION

Is the feature request related to a problem?
N/A

Describe alternatives you've considered
N/A

[stevehipwell]

I thought Karpenter dealt with it's own node lifecycle? @ellistarn @bwagner5 should be able to answer this definitively.

[aldredb]

@stevehipwell As of now, Karpenter need to rely on NTH for spot interruption handling

[stevehipwell]

@aldredb I thought that was a pre GA priority but you're right (aws/karpenter-provider-aws#702).

[bwagner5]

Karpenter does not handle Spot ITNs with the full 2 minutes to drain workloads. It also doesn't handle Rebalance Recommendations which would give workloads longer to drain. Karpenter does cordon and drain nodes when an operator deletes a nodes w/ kubectl or Karpenter itself terminates a node due to no pods running on it. But Karpenter will not react to AWS terminating a node like in the case of Spot ITNs.

We are still looking to add spot handling natively to Karpenter, but NTH will need to be installed for now.

[bwagner5]

I think we should add the docs to Karpenter instead of NTH since Karpenter has a nicer docs site. Once we get it into Karpenter, we can link it off of NTH's docs.

[sidewinder12s]

I've got a custom autoscaler as well that does not use ASGs, this would be useful to document since it should be similar requirements.

[olemarkus]

I am curious why you wouldn't want NTH to deal with spot terminations also together with karpenter? Karpenter's area of responsibility within a cluster is already pretty huge for a controller and I think NTH does its job pretty well.

[olemarkus]

Most clusters will be mixed Karpenter and CAS-based anyway given that most people cannot run Karpenter on the Control Plane, so NTH is most likely needed anyway.

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. If you want this issue to never become stale, please ask a maintainer to apply the "stalebot-ignore" label.

[github-actions]

This issue was closed because it has become stale with no activity.

[tamirhad]

When using NTH with SQS queue, ENABLE_SPOT_INTERRUPTION_DRAINING is always false when deploying with helm.
Is it expected? It's not something we need to be true?

[stevehipwell]

@tamirhad this is expected as I replied on the other thread that you asked this question on. Despite some AWS docs referring to this value when also specifying queue mode, ENABLE_SPOT_INTERRUPTION_DRAINING is only used for IMDS mode.

[universam1]

can we reopen this issue please? I actually have issues to properly configure NTH for Karpenter nodes #654

[aavileli]

please update karpentar documentation that users need to use NTH(with Queue Processor) for aws api termination for spot and on-demnad instances

[snay2]

As @bwagner5 mentioned, this would fit better in the Karpenter documentation.

I have opened an issue on the Karpenter repo and recommend that the discussion continue there: aws/karpenter-provider-aws#2159