Charm configures "Topology=True" preventing volume mounts

[Peter Jose De Sousa]

Hello,

When deploying the cinder-csi charm, the cinder controller plugin is configured with Topology enabled = true. This prevents volumes from mounting in the case where nodes are setup in Host Aggregate groups, e.g. AZ1, AZ2, and so on.

This is because the PV is forced onto a "nova" AZ:

kubectl get -o yaml pv pvc-2b992195-0afb-4033-ab45-e371a671bfee
apiVersion: v1
kind: PersistentVolume
metadata:
  annotations:
    pv.kubernetes.io/provisioned-by: cinder.csi.openstack.org
    volume.kubernetes.io/provisioner-deletion-secret-name: ""
    volume.kubernetes.io/provisioner-deletion-secret-namespace: ""
  creationTimestamp: "2024-03-26T13:07:40Z"
  finalizers:
  - kubernetes.io/pv-protection
  name: pvc-2b992195-0afb-4033-ab45-e371a671bfee
  resourceVersion: "254584"
  uid: 7af9c1f5-f781-48ce-821f-5924006f0266
spec:
  accessModes:
  - ReadWriteOnce
  capacity:
    storage: 3Gi
  claimRef:
    apiVersion: v1
    kind: PersistentVolumeClaim
    name: task-pv-claim
    namespace: default
    resourceVersion: "254580"
    uid: 2b992195-0afb-4033-ab45-e371a671bfee
  csi:
    driver: cinder.csi.openstack.org
    fsType: ext4
    volumeAttributes:
      storage.kubernetes.io/csiProvisionerIdentity: 1711392938167-8081-cinder.csi.openstack.org
    volumeHandle: f2b028de-5960-43a5-bbec-a6db22ccc11d
  nodeAffinity:
    required:
      nodeSelectorTerms:
      - matchExpressions:
        - key: topology.cinder.csi.openstack.org/zone
          operator: In
          values:
          - nova

[Workaround]

This is forced by the CSI. To work around this, disable the feature gate:

kubectl edit deploy -n kube-system csi-cinder-controllerplugin

  template:
    metadata:
      creationTimestamp: null
      labels:
        app: csi-cinder-controllerplugin
    spec:
      containers:
      - args:
        - --csi-address=$(ADDRESS)
        - --timeout=3m
        - --leader-election=true
        - --default-fstype=ext4
        env:
        - name: ADDRESS
          value: /var/lib/csi/sockets/pluginproxy/csi.sock
        image: rocks.canonical.com:443/cdk/sig-storage/csi-attacher:v4.2.0
        imagePullPolicy: IfNotPresent
        name: csi-attacher
        resources: {}
        terminationMessagePath: /dev/termination-log
        terminationMessagePolicy: File
        volumeMounts:
        - mountPath: /var/lib/csi/sockets/pluginproxy/
          name: socket-dir
      - args:
        - --csi-address=$(ADDRESS)
        - --timeout=3m
        - --default-fstype=ext4
        - --feature-gates=Topology=false
        - --extra-create-metadata
        - --leader-election=true

The correct, long term solution here is to be able to configure topology via the charm or for the charm to automatically configure the topology.

Thanks,
Peter

[Peter Jose De Sousa (pjds)]

Subscribing field high as its affectecting a customer deployment and prevents PV attachment.

[Adam Dyess (addyess)]

https://github.com/canonical/cinder-csi-operator/pull/5#pullrequestreview-1964015287