Summary: Galera pods fail to start in dual-stack IPv4/IPv6 OpenStack deployments

Description: When deploying OpenStack with dual-stack IPv4/IPv6 networking, the Galera database pods enter CrashLoopBackOff state and fail to form a proper cluster quorum. This affects both the main Galera cluster (openstack-galera) and cell clusters (openstack-cell1-galera).

Environment:

• OpenStack Platform: RHOSO 18 / OSP 18
• Network Configuration: Dual-stack IPv4/IPv6
• Deployment Topology: NFV OVS-DPDK-SRIOV with IPv6
• Kubernetes/OpenShift version: v1.31.6

Steps to Reproduce:

1. Deploy OpenStack control plane with dual-stack IPv4/IPv6 configuration
2. Deploy Galera database clusters as part of the control plane
3. Observe pod status with oc get pods -n openstack | grep galera

que hacen estos pods?
openstack-cell1-galera-0                                                 0/1     CrashLoopBackOff   1280 (4m51s ago)   2d19h
openstack-cell1-galera-1                                                 0/1     CrashLoopBackOff   1279 (46s ago)     2d19h
openstack-cell1-galera-2                                                 1/1     Running            6 (28h ago)        2d19h
openstack-galera-0                                                       0/1     CrashLoopBackOff   1286 (4m5s ago)    2d19h
openstack-galera-1                                                       0/1     CrashLoopBackOff   1278 (4m3s ago)    2d19h
openstack-galera-2                                                       1/1     Running            6 (12h ago)        2d19h 

Actual Results:

• Galera pods remain in CrashLoopBackOff state (e.g., 2d19h duration)
• Only 1 out of 3 pods in each cluster successfully starts
• Cluster quorum cannot be established (requires 2/3 nodes minimum)
• Pods affected:
  • openstack-galera-0, openstack-galera-1, openstack-galera-2
  • openstack-cell1-galera-0, openstack-cell1-galera-1, openstack-cell1-galera-2

Expected Results: All Galera pods should start successfully and form a healthy 3-node cluster with proper quorum.

Root Cause: The Galera StatefulSet/Service resources are not configured with the proper ipFamilyPolicy for dual-stack environments. The services (openstack-galera and openstack-cell1-galera) default to single-stack mode, preventing proper cluster communication in IPv4/IPv6 environments.

Workaround: Manually patch the Galera service resources to add dual-stack configuration:

{{# Patch the main Galera service
oc -n openstack patch service openstack-galera --type='merge' -p '{
"spec":

}'

1. Patch the cell1 Galera service
   oc -n openstack patch service openstack-cell1-galera --type='merge' -p ' { "spec": \{ "ipFamilyPolicy": "PreferDualStack", "ipFamilies": ["IPv6", "IPv4"] }

   }'

1. Verify the patches were applied
   oc -n openstack get service openstack-galera -o jsonpath='{.spec.ipFamilyPolicy} {"\n"} {.spec.ipFamilies}

   {"n"}
   '
   oc -n openstack get service openstack-cell1-galera -o jsonpath='{.spec.ipFamilyPolicy}

   {"\n"}{.spec.ipFamilies}{"n"}

   '

1. Monitor pod recovery (may need to force restart)
   oc -n openstack get pods | grep galera

1. If pods don't recover automatically, force restart by deleting them
   oc -n openstack delete pod openstack-galera-0 openstack-galera-1 openstack-galera-2
   oc -n openstack delete pod openstack-cell1-galera-0 openstack-cell1-galera-1 openstack-cell1-galera-2}}
   Proposed Solution: The OpenStack operator should support configuring ipFamilyPolicy through the Galera template configuration in the OpenStackControlPlane CR. This would allow users to properly configure dual-stack networking without requiring manual patches to the generated service resources.

Example of desired template configuration:

{{galera:
templates:
openstack:
ipFamilyPolicy: PreferDualStack
ipFamilies:

• IPv6
• IPv4
  openstack-cell1:
  ipFamilyPolicy: PreferDualStack
  ipFamilies:
• IPv6
• IPv4}}
  Impact:

• Severity: High/Critical
• Priority: High
• Blocks deployment of OpenStack in dual-stack IPv4/IPv6 environments
• Affects all database-dependent services (Nova, Neutron, Glance, Cinder, Heat, etc.)
• No functional control plane without working Galera clusters
• Manual workaround is required for every deployment and may not persist through operator reconciliation

Additional Information:

• This configuration follows standard Kubernetes dual-stack networking practices
• The ipFamilyPolicy: PreferDualStack setting allows the service to work with both IP families
• Currently not exposed through the operator's template interface for Galera resources
• The operator should propagate this configuration from the OpenStackControlPlane CR to the generated Service resources