Closed raulgdUOC closed 6 days ago
Hi @raulgdUOC , Thank you for bringing this issue to our attention. We have investigated the problem where persistent volumes were getting stuck during deletion and have identified the cause. To address this, we recently made updates to the CSI provisioner sidecars and adjusted the RBAC permissions specifically for managing persistent volumes. This should fix the issue.
Hi @mskanth972, thanks for your response,
I have changed the version of the images and updated the ClusterRole, but the problem still persists with the same logs:
Looks like it can't mount the efs at that access point created after delete the PVC and then, it get stucked with the PV in RECLAIM POLICY: Delete
and STATUS: Released
with the error showd in the first message and the logs showed before.
The procedure to test the feature deleteAccessPointRootDir: true
is the next:
deleteAccessPointRootDir: true
and CSI provisioner sidecars and adjusted the RBAC permissionsAm I doing something wrong, or is this procedure not the intended functionality for deleteAccessPointRootDir: true
?
When I delete the PVC, the content inside that PV should be deleted along with its access point.
For the above we need to enable this parameter deleteAccessPointRootDir: true
for sure.
Can I know the command you are running for deleting PVC and also where you are assigning this parameter reclaimPolicy: Delete
?
For the above we need to enable this parameter
deleteAccessPointRootDir: true
for sure. Can I know the command you are running for deleting PVC and also where you are assigning this parameterreclaimPolicy: Delete
?
I am using kubectl delete pvc efs-claim
to delete the PVC.
I am assigning the parameter reclaimPolicy: Delete
in the Storage Class, but I also tested it by doing a patch
with the storage class set to Retain
.
Hello @mskanth972,
The root problem was that when the controller tried to mount the EFS, it couldn't reach the EFS because the Security Group only allowed traffic with the CIDR of my worker nodes. The network of the Pod is different from that of the node, so the controller couldn't mount the EFS or delete the access point and its content, causing it to get stuck in the process. To solve this, I added the label hostNetwork: true
, and it worked.
Additionally, the information you provided about the ClusterRole
also worked to delete the PV. Without the patch
in the ClusterRole
, the PV got stuck in the terminating stage.
/kind bug
What happened? When I try to delete the PVC, its PV does not get deleted (with
reclaimPolicy: Delete
), and its access point remains. The controller starts to generate logs trying to mount that access point, but it can't mount it.What you expected to happen? When I delete the PVC, the content inside that PV should be deleted along with its access point.
How to reproduce it (as minimally and precisely as possible)? I have been using this to deploy the storage class, pod, and PVC -> https://github.com/kubernetes-sigs/aws-efs-csi-driver/tree/master/examples/kubernetes/dynamic_provisioning/specs
I used Helm to deploy it. In the values, the most important change that I made is: controller.deleteAccessPointRootDir=true
Anything else we need to know?:
Environment
kubectl version
): v1.28.9-eksPlease also attach debug logs to help us better diagnose
This is when I create a new PV with the storage class
After delete the POD and the PVC, I start to see this logs:
This is a description of a PV from an another test that I have made. Some parts may not agree with the previous logs. This is the result after delete the PVC.