-
Bug
-
Resolution: Done
-
Normal
-
rhos-18.0.0
-
None
-
2
-
False
-
-
False
-
?
-
?
-
?
-
?
-
None
-
-
-
Moderate
the openstackclient crd supports passing a nodeselector https://github.com/openstack-k8s-operators/openstack-operator/blob/main/apis/client/v1beta1/openstackclient_types.go#L53 . It should be possible to set it via the openstackclient section in OpenStackControlPlane CR.
current ctlplane openstackclient section:
# oc get osctlplane controlplane -o yaml | yq .spec.openstackclient
{
"template": {
"openStackConfigMap": "openstack-config",
"openStackConfigSecret": "openstack-config-secret"
}
}
Pod is scheduled to master-1
# oc get pods -o wide |grep client ceph-client-edpm-deployment-post-ceph-openstack-edpm-7fzx8 0/1 Completed 0 2d19h 192.168.20.134 master-1 <none> <none> openstackclient 1/1 Running 0 2d19h 192.168.20.89 master-1 <none> <none>
Patch the ctlplane to set a nodeSelector for the openstackclient
# oc patch osctlplane controlplane --type='merge' --patch '{"spec": {"openstackclient": {"template": { "nodeSelector": {"kubernetes.io/hostname": "master-0"}}}}}' openstackcontrolplane.core.openstack.org/controlplane patched # oc get osctlplane controlplane -o yaml | yq .spec.openstackclient { "template": { "nodeSelector": { "kubernetes.io/hostname": "master-0" }, "openStackConfigMap": "openstack-config", "openStackConfigSecret": "openstack-config-secret" } }
There is a bug right now that after setting the nodeSelector, the openstackclient pod won't get re-created with the nodeSelector. When deleting the openstackclient the new pod will receive the nodeSelector
# oc delete pod openstackclient -n openstack
# oc get pods openstackclient -o yaml | yq .spec.nodeSelector
{
"kubernetes.io/hostname": "master-0"
}
# oc get pods -o wide |grep openstackclient
openstackclient 1/1 Running 0 5m19s 192.168.16.204 master-0 <none> <none>
