Autoscaler: Cluster-autoscaler does not work with VPC-Endpoint (AWS/EKS)

[maust]

We do have a quite locked-down network in AWS (no internet-connectivity at all). Access to AWS services only via VPC endpoints and on-premise systems via DirectConnect. At the same time we would like to use IAM roles.

Kubernetes: 1.14
Cluster-Autoscaler: 1.14.7

When using cluster-autoscaler it cannot fetch credentials via STS using the IAM role. To my understanding the issue is caused by cluster-autoscaler not using the regional STS endpoint (https://sts.eu-central-1.amazonaws.com) but instead the global (https://sts.amazonaws.com). With VPC-Endpoints it is not possible to replace the global enpoint.

With github.com/aws/aws-sdk-go v1.25.18
(see https://github.com/aws/aws-sdk-go/blob/master/CHANGELOG.md) configuration of regional STS endpoints was introduced by setting env AWS_STS_REGIONAL_ENDPOINTS=regional

I tried setting the region as env AWS_REGION and AWS_STS_REGIONAL_ENDPOINTS but still the global endpoint is used.

After looking at the 1.14 branch for cluster-autoscaler, it looks like v1.23.22 is used (see
https://github.com/kubernetes/autoscaler/blob/cluster-autoscaler-release-1.14/cluster-autoscaler/vendor/github.com/aws/aws-sdk-go/CHANGELOG.md)

I also checked the other cluster-autoscaler branches:

• 1.14 github.com/aws/aws-sdk-go v1.23.22
• 1.15 github.com/aws/aws-sdk-go v1.16.26
• 1.16 github.com/aws/aws-sdk-go v1.23.12
• 1.17 github.com/aws/aws-sdk-go v1.23.18
• master github.com/aws/aws-sdk-go v1.28.2

So I would assume that supporting such a use case would be possible by upgrading the aws-sdk-go version to >= v1.25.18 - let me know if I can be of help.

logs:
E0213 16:05:54.390164 1 aws_manager.go:259] Failed to regenerate ASG cache: cannot autodiscover ASGs: WebIdentityErr: failed to retrieve credentials
caused by: RequestError: send request failed
caused by: Post https://sts.amazonaws.com/: dial tcp 54.239.29.25:443: i/o timeout
F0213 16:05:54.390200 1 aws_cloud_provider.go:330] Failed to create AWS Manager: cannot autodiscover ASGs: WebIdentityErr: failed to retrieve credentials
caused by: RequestError: send request failed
caused by: Post https://sts.amazonaws.com/: dial tcp 54.239.29.25:443: i/o timeout

Attached you can find the kubernetes deployment yaml.
deployment.txt

[ajohnstone]

This has been updated in master 3 months ago, but no release. Any idea when a new release would be cut?

https://github.com/kubernetes/autoscaler/commit/af6f3258d6a1ebf7bf939ad2fc65de4cf8e2a9cb

[Jeffwan]

/assign

[Jeffwan]

Thanks. I can help resolve this issue and request newer release. I think what we can help is bump the SDK version and then user can mount env AWS_STS_REGIONAL_ENDPOINTS=regional. SDK client will pick up env and resolve right endpoint. Is that correct?

[ajohnstone]

@Jeffwan That is correct, It's resolved in the version that is already in master.

-   github.com/aws/aws-sdk-go v1.23.18
+   github.com/aws/aws-sdk-go v1.28.2

See go.mod in master here.
https://github.com/kubernetes/autoscaler/commit/af6f3258d6a1ebf7bf939ad2fc65de4cf8e2a9cb#diff-5b1211f36242f6afe85bdb0062369dc3R16

The upstream fix was here https://github.com/aws/aws-sdk-go/pull/2779/files
Resolved in aws-sdk-go "Release v1.25.18 (2019-10-23)"
https://github.com/aws/aws-sdk-go/blob/master/CHANGELOG.md#sdk-enhancements-13

[ajohnstone]

Fixes #2532

[Jeffwan]

@ajohnstone Thanks. I plan to have a few cherry-pick recently, I will make the change and include this in the new release.

[Jeffwan]

Just make changes on 1.15. I will make the changes for rest of the version

[Jeffwan]

em.. Sorry we only have following versions to support this case.

https://github.com/kubernetes/autoscaler/releases/tag/cluster-autoscaler-1.15.6
https://github.com/kubernetes/autoscaler/releases/tag/cluster-autoscaler-1.18.1

1.14, 1.16 and 1.17 changes is not included in this release. Change will be merged and you can build one image for short term. If you need any help, let me know

[maust]

We are just migrating to 1.15, thanks a lot. If I can find time I will extend the AWS documentation.

[maust]

Successful tested with 1.15.6, thanks again.

I opened https://github.com/kubernetes/autoscaler/pull/3052

From my side this could be closed, not sure if you want to keep it open until the other versions support it.

[Jeffwan]

I will leave it open to track changes in other branches. @maust Thanks for the contribution. I will review the doc change

[haofeif]

Hi @Jeffwan any plan to fix them in other version (1.16.x -> 1.17.x)?
thanks.

[Jeffwan]

1.17.x change has been merged. I get some feedbacks on 1.16.x and I will fix it before next release. @haofeif

I also update PR and address feedbacks for 1.16 change. Once it get merged, next release will pick it up. https://github.com/kubernetes/autoscaler/pull/3003

[ramdesh]

Hello, is there any info on when this will be released in a 1.16.x or 1.17.x release? Thanks

[Jeffwan]

1.16.6, 1.17.3 have been released. Please download latest version. I will close the issue. Thanks everyone for all your feedbacks

[Jeffwan]

/close

[k8s-ci-robot]

@Jeffwan: Closing this issue.

In response to this:

/close

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.