Support cluster name determination via label for ClusterClass compatibility (#165)

* Support cluster name determination via label for ClusterClass compatibility

Cluster API, when using the ClusterClass approach, automatically sets the
`cluster.x-k8s.io/cluster-name` label on the OpenStackCluster resource. It also
generates the OpenStackCluster `metadata.name` by appending five random characters
(e.g. `<cluster-name>-pr2wr`). As a result, relying solely on the resource name
to determine the cluster name caused deletion to fail, since the expected
cluster name could not be matched with existing resources, e.g. LBs.

This commit updates the cluster name detection logic to first check for the
presence of the `cluster.x-k8s.io/cluster-name` label. If the label is found,
its value is used as the cluster name. If the label is not present, the
OpenStackCluster's `metadata.name` is used as a fallback.

This change maintains backward compatibility and does not introduce any
breaking behavior for clusters that do not use ClusterClass.

Signed-off-by: Matej Feder <[email protected]>

* Apply suggestions from code review

Co-authored-by: Matt Anson <[email protected]>

---------

Signed-off-by: Matej Feder <[email protected]>
Co-authored-by: Matt Anson <[email protected]>

Author: matofeder

README.md

@@ -66,17 +66,22 @@ Cluster API `Cluster`, from being removed until the finalizer is removed.
 (specifically, it waits for the `deletionTimestamp` to be set, indicating that a deletion
 has been requested), at which point it uses the credential from
 `OpenStackCluster.spec.identityRef` to remove any dangling resources that were created by
-the OCCM or Cinder CSI with the same cluster name as the cluster being deleted. Once all
-the resources have been deleted, the finalizer is removed.
+the OCCM or Cinder CSI with the same cluster name as the cluster being deleted. 
+The cluster name is determined by the `cluster.x-k8s.io/cluster-name` label on the
+OpenStackCluster resource, if present.
+If the label is not set, the name of the OpenStackCluster resource (`metadata.name`) is
+used instead.
+Once all the resources have been deleted, the finalizer is removed.
 
 > **WARNING**
 >
 > The cluster name of the OCCM and Cinder CSI **must** be set to the `metadata.name`
-> of the `OpenStackCluster` resource.
+> of the OpenStackCluster resource, or to the value of the `cluster.x-k8s.io/cluster-name`
+> label if it is present on the OpenStackCluster resource.
 >
 > For instance, the `openstack-cluster` chart from the
 > [capi-helm-charts](https://github.com/azimuth-cloud/capi-helm-charts) ensures that this happens
-> automatically.
+> automatically and sets the OpenStackCluster's `metadata.name` for OCCM and Cinder CSI.
 
 The advantage of this approach vs. a task that runs before the cluster deletion is started
 is that the external resource deletion happens _after_ all the machines have been deleted,

capi_janitor/openstack/operator.py

@@ -334,13 +334,17 @@ async def wrapper(**kwargs):
 
 @kopf.on.event(CAPO_API_GROUP, "openstackclusters")
 @retry_event
-async def on_openstackcluster_event(name, namespace, meta, spec, logger, **kwargs):
+async def on_openstackcluster_event(name, namespace, meta, labels, spec, logger, **kwargs):
     """
     Executes whenever an event occurs for an OpenStack cluster.
     """
     # Get the resource for manipulating OpenStackClusters at the preferred version
     capoapi = await ekclient.api_preferred_version(CAPO_API_GROUP)
     openstackclusters = await capoapi.resource("openstackclusters")
+    # Use the value of the `cluster.x-k8s.io/cluster-name` label as the cluster name if it exists,
+    # otherwise, fall back to the OpenStackCluster resource name.
+    clustername = labels.get("cluster.x-k8s.io/cluster-name", name)
+    logger.debug(f"cluster name that will be used for cleanup: '{clustername}'")
 
     finalizers = meta.get("finalizers", [])
     # We add a custom finalizer to OpenStack cluster objects to
@@ -401,7 +405,7 @@ async def on_openstackcluster_event(name, namespace, meta, spec, logger, **kwarg
             clouds,
             cloud_name,
             cacert,
-            name,
+            clustername,
             volumes_annotation_value == VOLUMES_ANNOTATION_DELETE,
             remove_appcred and len(finalizers) == 1
         )