operator: DaemonSet "csi-node-driver" is invalid

Expected Behavior

All pods are successfully launched by the operator.

Current Behavior

When the operator tries to create the csi-node-driver Deployment, it gets an error:

DaemonSet.apps "csi-node-driver" is invalid: [spec.template.spec.volumes[2].hostPath.path: Required value, spec.template.spec.containers[0].volumeMounts[3].name: Not found: "kubelet-dir", spec.template.spec.containers[0].volumeMounts[3].mountPath: Required value]

Possible Solution

N/A

Steps to Reproduce (for bugs)

  1. Have an EKS 1.21 cluster
  2. Deploy v3.24.3 of the helm chart with installation.kubernetesProvider=EKS

Context

The csi-node-driver Deployment failing to create causes the calico-node pods to be endlessly created and terminated.

Your Environment

  • Operating System and version: EKS 1.21

About this issue

  • Original URL
  • State: closed
  • Created 4 years ago
  • Comments: 32 (10 by maintainers)

Most upvoted comments

I can confirm that updating the (numerous) CRDs manually allows me to upgrade to the v3.24.5 release in the same cluster where I originally encountered this problem

@Josh-Tigera We believe to have found the issue.

Helm v3 does not handle CRDs upgrade by default. It means that every changes made in crds folder of the Helm charts are not upgraded and only apply at first install (that’s why a clean install is OK for you)

When I applied manually the latest crds folders of the v3.24.5 charts everything worked.

I believe you should either:

  • release a separate charts to handle CRDs upgrade
  • Tell user to upgrade CRDs separatly from Helm (that’s what kube-prometheus-stack is doing from example)
  • Have the CRDs packaged in templates but that would cause issues at first install because no guarantee that installation CRD would be installed before the actual installation object.

Here in my case I didn’t have the new spec kubeletVolumePluginPath in my CRD because it was not up to date

We are seeing the same error on EKS 1.22 with platform version eks.6 and the calico status is degraded. However, it doesn’t appear to be restarting the calico-node pods.