Fix importer issue after leader controller restart

1. Reset common area's cluster manager so it can be started again when member cluster disconnect with leader and reconnect. Otherwise, it will raise an error `can't accept new runnable as stop procedure is already engaged` 2. Always update the ResourceImport cache for importer so it can handle deletion event correctly. Signed-off-by: Lan Luo <[email protected]>
antrea-io · Apr 7, 2022 · 4368965 · 4368965
1 parent 03b3f2b
commit 4368965
Show file tree

Hide file tree

Showing 3 changed files with 16 additions and 5 deletions.
diff --git a/multicluster/controllers/multicluster/commonarea/leader_elector.go b/multicluster/controllers/multicluster/commonarea/leader_elector.go
@@ -122,6 +122,8 @@ func (r *remoteCommonAreaManager) setElectedLeader(cluster RemoteCommonArea) {
 	}
 	r.electedLeaderCluster = cluster
 	if cluster != nil {
-		cluster.StartWatching()
+		if err := cluster.StartWatching(); err != nil {
+			klog.ErrorS(err, "Failed to start watching events")
+		}
 	}
 }
diff --git a/multicluster/controllers/multicluster/commonarea/remote_common_area.go b/multicluster/controllers/multicluster/commonarea/remote_common_area.go
@@ -147,7 +147,6 @@ func NewRemoteCommonArea(clusterID common.ClusterID, clusterSetID common.Cluster
 	if e != nil {
 		return nil, e
 	}
-
 	remote := &remoteCommonArea{
 		Client:                  remoteClient,
 		ClusterManager:          mgr,
@@ -409,6 +408,17 @@ func (r *remoteCommonArea) StopWatching() {
 	}
 	r.managerStopFunc()
 	r.managerStopFunc = nil
+
+	// Reset ClusterManager so this common area can be started again when it's reconnected.
+	mgr, err := ctrl.NewManager(r.config, ctrl.Options{
+		Scheme:             r.scheme,
+		MetricsBindAddress: "0",
+		Namespace:          r.Namespace,
+	})
+	if err != nil {
+		klog.ErrorS(err, "Error to reset manager for RemoteCommonArea", "Cluster", r.ClusterID)
+	}
+	r.ClusterManager = mgr
 }
 
 func (r *remoteCommonArea) GetStatus() []multiclusterv1alpha1.ClusterCondition {

diff --git a/multicluster/controllers/multicluster/commonarea/resourceimport_controller.go b/multicluster/controllers/multicluster/commonarea/resourceimport_controller.go
@@ -210,7 +210,6 @@ func (r *ResourceImportReconciler) handleResImpUpdateForService(ctx context.Cont
 			klog.ErrorS(err, "Failed to update imported Service", "service", svcName.String())
 			return ctrl.Result{}, err
 		}
-		r.installedResImports.Update(*resImp)
 	}
 
 	if !apiequality.Semantic.DeepEqual(svcImp.Spec, svcImpObj.Spec) {
@@ -221,8 +220,8 @@ func (r *ResourceImportReconciler) handleResImpUpdateForService(ctx context.Cont
 			klog.ErrorS(err, "Failed to update ServiceImport", "serviceimport", svcImpName.String())
 			return ctrl.Result{}, err
 		}
-		r.installedResImports.Update(*resImp)
 	}
+	r.installedResImports.Update(*resImp)
 	return ctrl.Result{}, nil
 }
 
@@ -324,8 +323,8 @@ func (r *ResourceImportReconciler) handleResImpUpdateForEndpoints(ctx context.Co
 			klog.ErrorS(err, "Failed to update MCS Endpoints", "endpoints", epNamespaced.String())
 			return ctrl.Result{}, err
 		}
-		r.installedResImports.Update(*resImp)
 	}
+	r.installedResImports.Update(*resImp)
 	return ctrl.Result{}, nil
 }