Cleanup in-memory node state on node removal

Signed-off-by: Patryk Diak <[email protected]>

Author: kyrtapz

pkg/cloudprovider/aws.go

@@ -447,6 +447,9 @@ func sharedCredentialsFileFromDirectory(dir string) (string, error) {
 	return f.Name(), nil
 }
 
+func (a *AWS) CleanupNode(nodeName string) {
+}
+
 // newConfigForStaticCreds is copied verbatim from:
 // https://github.com/openshift/cluster-ingress-operator/blob/1600a0e349ef075fcb52ab65b33445e256358ab8/pkg/util/aws/shared_credentials_file.go#L52
 func newConfigForStaticCreds(accessKey string, accessSecret string) []byte {

pkg/cloudprovider/azure.go

@@ -667,6 +667,12 @@ func (a *Azure) getNodeLock(nodeName string) *sync.Mutex {
 	return a.nodeLockMap[nodeName]
 }
 
+func (a *Azure) CleanupNode(nodeName string) {
+	a.nodeMapLock.Lock()
+	defer a.nodeMapLock.Unlock()
+	delete(a.nodeLockMap, nodeName)
+}
+
 func getNameFromResourceID(id string) string {
 	return id[strings.LastIndex(id, "/"):]
 }

pkg/cloudprovider/cloudprovider.go

@@ -63,6 +63,10 @@ type CloudProviderIntf interface {
 	// The cpicIPs parameter is a set of IP addresses that are
 	// managed by CloudPrivateIPConfigs and should be excluded from capacity calculations.
 	GetNodeEgressIPConfiguration(node *corev1.Node, cpicIPs sets.Set[string]) ([]*NodeEgressIPConfiguration, error)
+
+	// CleanupNode removes any internal state associated with the node.
+	// This should be called when a node is deleted.
+	CleanupNode(nodeName string)
 }
 
 // CloudProviderWithMoveIntf is additional interface that can be added to cloud

pkg/cloudprovider/cloudprovider_fake.go

@@ -69,3 +69,6 @@ func (f *FakeCloudProvider) GetNodeEgressIPConfiguration(node *corev1.Node, cpic
 	}
 	return nil, nil
 }
+
+func (f *FakeCloudProvider) CleanupNode(nodeName string) {
+}

pkg/cloudprovider/gcp.go

@@ -335,3 +335,9 @@ func (g *GCP) getNodeLock(nodeName string) *sync.Mutex {
 	}
 	return g.nodeLockMap[nodeName]
 }
+
+func (g *GCP) CleanupNode(nodeName string) {
+	g.nodeMapLock.Lock()
+	defer g.nodeMapLock.Unlock()
+	delete(g.nodeLockMap, nodeName)
+}

pkg/cloudprovider/openstack.go

@@ -1015,3 +1015,6 @@ func getNodeInternalAddrs(node *corev1.Node) (net.IP, net.IP) {
 	}
 	return v4Addr, v6Addr
 }
+
+func (o *OpenStack) CleanupNode(nodeName string) {
+}

pkg/controller/node/node_controller.go

@@ -87,6 +87,8 @@ func NewNodeController(
 				controller.Enqueue(newN)
 			}
 		},
+		// Enqueue removals for cleanup purposes only
+		DeleteFunc: controller.Enqueue,
 	})
 	if err != nil {
 		return nil, err
@@ -103,6 +105,8 @@ func (n *NodeController) SyncHandler(key string) error {
 		// // A lister can only return ErrNotFound, which means: the Node
 		// resource no longer exist, in which case we stop processing.
 		klog.Infof("corev1.Node: '%s' in work queue no longer exists", key)
+		// Clean up any cloud provider state associated with this node
+		n.cloudProviderClient.CleanupNode(key)
 		return nil
 	}