[EKS] [EFS CSI driver add-on]: deleteAccessPointRootDir flag not working

[kordian2k]

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

Tell us about your request

The EFS CSI driver add-on allows to set the configuration flag deleteAccessPointRootDir. If set to true it should automatically delete files on EFS (not only the access points) once the corresponding PVC is removed. However, the files are not deleted. Moreover, the add-on does not behave like the corresponding helm chart with the controller.deleteAccessPointRootDir=true option set.

In particular, the efs-plugin container within the efs-csi-controller pod starts with the argument --delete-access-point-root-dir=false

Additional context Tested with the following setup EKS: 1.27 EFS add-on: v1.5.8-eksbuild.1 Node group: BOTTLEROCKET_x86_64, 1.14.3-764e37e4, No Fargate nodes present

[dims]

@kordian2k the EFS CSI Driver is based off of the upstream kubernetes community repository here - https://github.com/kubernetes-sigs/aws-efs-csi-driver, So can you please raise this issue there?

[kordian2k]

@dims it seems the problem is within the issue not in the upstream driver. The latter works with the deleteAccessPointRootDir flag if installed via the helm chart.

[mikestef9]

We are working on fix which will be deployed with version 1.6 of add-on

[shandr]

@mikestef9 , I'm on v1.7.0-eksbuild.1 version, and it's still not working.

In the deployment I see that - --delete-access-point-root-dir=false

[kordian2k]

I can confirm that the issue is still not resolved on 1.7.0.

[relaxdiego]

This bug is still present in v1.7.1-eksbuild.1

[tsoe77]

The issue still exists in v1.7.4-eksbuild.1

[vijaybandari]

@dims Looking at the helm chart the value is defined under controller object, but the add-on configuration schema doesn't have it under controller, rather it's at root level (deleteAccessPointRootDir).

Fixing configuration schema should fix this.

Due to this issue, we have bunch of orphan access points costing us $.

[DrVaskee]

Still present in v1.7.5-eksbuild.2

We also have a lot of access points that should have been deleted a long time ago.

[mjsoyeon]

This issue has been fixed in latest version of driver(v1.7.6) with updated schema.