Loading...

XML

Word

Printable

Type: Bug
Resolution: Done-Errata
Priority: Major
Fix Version/s: rhos-18.0.6
Affects Version/s: rhos-18.0.0
Component/s: cinder-operator
Labels:
None

Story Points:
3
Blocked:
False
Blocked Reason:

Hide

None

Show
None
Ready:
False
Docs Approval:
No Docs Impact
Fixed in Build:
cinder-operator-container-1.0.7-6
Regression:
None
Release Note Text:

Hide
.ExtraMounts can process a per-instance propagation when the pod is prefixed with an arbitrary name

When `uniquePodNames` is `true`, every Cinder pod (and in general each component and service) is prefixed by a pseudo-random string. With this update, ExtraMounts can process a per-instance propagation when the pod is prefixed with an arbitrary name.

Show
.ExtraMounts can process a per-instance propagation when the pod is prefixed with an arbitrary name When `uniquePodNames` is `true`, every Cinder pod (and in general each component and service) is prefixed by a pseudo-random string. With this update, ExtraMounts can process a per-instance propagation when the pod is prefixed with an arbitrary name.
Release Note Type:
Bug Fix
Release Note Status:
Done
Git Pull Request:
https://github.com/openstack-k8s-operators/cinder-operator/pull/463, https://github.com/openstack-k8s-operators/cinder-operator/pull/464
Intelligence Requested:
Market:
Errata Link:
https://errata.engineering.redhat.com/advisory/146727

Severity:
Moderate

SFDC Cases Counter:
SFDC Cases Open:
SFDC Cases Links:

When uniquePodNames is set to true, every Cinder Pod (but in general each component and service) is prefixed by a pseudo random string. This affects the per-instance propagation, because the old way based on strings.TrimPrefix is not valid anymore.

In a DCN deployment we prefer to propagate secrets to pods by matching the instance AZ name. Example 1 below results in pods whose name match az0 getting the secret ceph-conf-az-0 and pods who's name patch az1 getting the secret ceph-conf-az-0 and so on. The example below works for Glance pods but only works for Cinder pods if uniquePodNames is set to false. As a workaround customers should set uniquePodNames false as in Example 2 until this bug is resolved. uniquePodNames is only needed if the storage backend uses NFS.

Example 1

apiVersion: core.openstack.org/v1beta1
kind: OpenStackControlPlane
spec: 
  extraMounts: 
  - extraVol: 
    - extraVolType: Ceph
      mounts: 
      - mountPath: /etc/ceph
        name: ceph0
        readOnly: true
      propagation: 
      - az0
      volumes: 
      - name: ceph0
        projected: 
          sources: 
          - secret: 
              name: ceph-conf-az-0
    - extraVolType: Ceph
      mounts: 
      - mountPath: /etc/ceph
        name: ceph1
        readOnly: true
      propagation: 
      - az1
      volumes: 
      - name: ceph1
        projected: 
          sources: 
          - secret: 
              name: ceph-conf-az-1

Example 2

apiVersion: core.openstack.org/v1beta1
kind: OpenStackControlPlane
<...>
spec: 
  cinder: 
    uniquePodNames: false   # workaround https://issues.redhat.com/browse/OSPRH-11240
    enabled: true
    apiOverride: 
      <...>