Uploaded image for project: 'OpenShift Bugs'
  1. OpenShift Bugs
  2. OCPBUGS-4697

Cinder volumes are not provisioned when Cinder AZ and Nova AZ do not match


    • ?
    • Rejected
    • False
    • Hide



      Description of problem:

      When setting cluster monitoring to use the standard-csi storage class for prometheus in an environment with three Nova AZs (az1, az2, az3) and one Cinder AZ (nova), the prometheus pods hang in Pending status as the CSI driver is failing to create the Cinder volume due to an incorrect AZ specification.

      Version-Release number of selected component (if applicable):

      Tested on 4.10.20, 4.11.9, and 4.12.0-ec.5

      How reproducible:

      In an environment with Nova AZs but not Cinder AZs (common for environments using Ceph without DCN), deploy an OCP cluster and then attempt to set the storage for prometheus to use the standard-csi storage class.

      Steps to Reproduce:

      1. Deploy OpenShift with default shift-on-stack configurations
      2. Observe that ignore-volume-az is set to yes by default
      3. Modify the cluster-monitoring-config ConfigMap to set a prometheus volumeClaimTemplate that uses the standard-csi storage class
      4. Observe that prometheus pods are stuck in Pending status
      5. View events on the created PVC and observe Cinder API responses indicating an incorrect AZ provided in the volume creation call

      Actual results:

      Volume creation fails and pods are stuck in Pending status

      Expected results:

      Volume is created without an AZ specified to Cinder (the in-tree behavior) and the pods start normally

      Additional info:

      Tested on OSP 16.2.3

            sfinucan@redhat.com Stephen Finucane
            aaustin@redhat.com Andrew Austin Byrum
            Jon Uriarte Jon Uriarte
            0 Vote for this issue
            6 Start watching this issue