update: monitoring stack only apply if it is managed cluster

- keep managed cluster logic as-is
- create redhat-ods-monitoring namespace only in managed if dsic set monitoring to enabled
- do not create redhat-ods-monitoring namespace in self-managed any more even monitoring is enabled
- remove role rolebinding and servicemonitor in upgrade for self-managed
- do not apply role rolebinding networkpoliy and servicemonitor for monitoring in clean install for self-managed

Signed-off-by: Wen Zhou <[email protected]>

Author: zdtsw

controllers/dscinitialization/dscinitialization_controller.go

@@ -234,35 +234,24 @@ func (r *DSCInitializationReconciler) Reconcile(ctx context.Context, req ctrl.Re
 			if !createUsergroup {
 				log.Info("DSCI disabled usergroup creation")
 			} else {
-				err := r.createUserGroup(ctx, instance, "rhods-admins")
-				if err != nil {
-					return reconcile.Result{}, err
-				}
-			}
-			if instance.Spec.Monitoring.ManagementState == operatorv1.Managed {
-				log.Info("Monitoring enabled, won't apply changes", "cluster", "Self-Managed RHODS Mode")
-				err = r.configureCommonMonitoring(ctx, instance)
-				if err != nil {
+				if err := r.createUserGroup(ctx, instance, "rhods-admins"); err != nil {
 					return reconcile.Result{}, err
 				}
 			}
 		case cluster.ManagedRhoai:
 			osdConfigsPath := filepath.Join(deploy.DefaultManifestPath, "osd-configs")
-			err = deploy.DeployManifestsFromPath(ctx, r.Client, instance, osdConfigsPath, r.ApplicationsNamespace, "osd", true)
-			if err != nil {
+			if err = deploy.DeployManifestsFromPath(ctx, r.Client, instance, osdConfigsPath, r.ApplicationsNamespace, "osd", true); err != nil {
 				log.Error(err, "Failed to apply osd specific configs from manifests", "Manifests path", osdConfigsPath)
 				r.Recorder.Eventf(instance, corev1.EventTypeWarning, "DSCInitializationReconcileError", "Failed to apply "+osdConfigsPath)
 
 				return reconcile.Result{}, err
 			}
 			if instance.Spec.Monitoring.ManagementState == operatorv1.Managed {
 				log.Info("Monitoring enabled in initialization stage", "cluster", "Managed Service Mode")
-				err := r.configureManagedMonitoring(ctx, instance, "init")
-				if err != nil {
+				if err := r.configureManagedMonitoring(ctx, instance, "init"); err != nil {
 					return reconcile.Result{}, err
 				}
-				err = r.configureCommonMonitoring(ctx, instance)
-				if err != nil {
+				if err = r.configureCommonMonitoring(ctx, instance); err != nil {
 					return reconcile.Result{}, err
 				}
 			}

controllers/dscinitialization/dscinitialization_test.go

@@ -143,7 +143,7 @@ var _ = Describe("DataScienceCluster initialization", func() {
 				WithPolling(interval).
 				Should(BeFalse())
 		})
-		It("Should create default monitoring namespace if monitoring enabled", func(ctx context.Context) {
+		It("Should not create default monitoring namespace even monitoring enabled for non-managed cluster", func(ctx context.Context) {
 			// when
 			desiredDsci := createDSCI(operatorv1.Managed, operatorv1.Managed, monitoringNamespace2)
 			Expect(k8sClient.Create(ctx, desiredDsci)).Should(Succeed())
@@ -159,8 +159,7 @@ var _ = Describe("DataScienceCluster initialization", func() {
 				WithContext(ctx).
 				WithTimeout(timeout).
 				WithPolling(interval).
-				Should(BeTrue())
-			Expect(foundMonitoringNamespace.Name).Should(Equal(monitoringNamespace2))
+				Should(BeFalse())
 		})
 	})
 

controllers/dscinitialization/utils.go

@@ -81,8 +81,8 @@ func (r *DSCInitializationReconciler) createOdhNamespace(ctx context.Context, ds
 			return err
 		}
 	}
-	// Create Monitoring Namespace if it is enabled and not exists
-	if dscInit.Spec.Monitoring.ManagementState == operatorv1.Managed {
+	// Create Monitoring Namespace if it is enabled and not exists and only for Managed cluster
+	if dscInit.Spec.Monitoring.ManagementState == operatorv1.Managed && platform == cluster.ManagedRhoai {
 		foundMonitoringNamespace := &corev1.Namespace{}
 		monitoringName := dscInit.Spec.Monitoring.Namespace
 		err := r.Get(ctx, client.ObjectKey{Name: monitoringName}, foundMonitoringNamespace)
@@ -205,18 +205,20 @@ func (r *DSCInitializationReconciler) reconcileDefaultNetworkPolicy(ctx context.
 			log.Error(err, "error to set networkpolicy in operator namespace", "path", networkpolicyPath)
 			return err
 		}
-		// Deploy networkpolicy for monitoring namespace
-		err = deploy.DeployManifestsFromPath(ctx, r.Client, dscInit, networkpolicyPath+"/monitoring", dscInit.Spec.Monitoring.Namespace, "networkpolicy", true)
-		if err != nil {
-			log.Error(err, "error to set networkpolicy in monitroing namespace", "path", networkpolicyPath)
-			return err
-		}
 		// Deploy networkpolicy for applications namespace
 		err = deploy.DeployManifestsFromPath(ctx, r.Client, dscInit, networkpolicyPath+"/applications", dscInit.Spec.ApplicationsNamespace, "networkpolicy", true)
 		if err != nil {
 			log.Error(err, "error to set networkpolicy in applications namespace", "path", networkpolicyPath)
 			return err
 		}
+		if platform == cluster.ManagedRhoai {
+			// Deploy networkpolicy for monitoring namespace
+			err = deploy.DeployManifestsFromPath(ctx, r.Client, dscInit, networkpolicyPath+"/monitoring", dscInit.Spec.Monitoring.Namespace, "networkpolicy", true)
+			if err != nil {
+				log.Error(err, "error to set networkpolicy in monitroing namespace", "path", networkpolicyPath)
+				return err
+			}
+		}
 	} else { // Expected namespace for the given name in ODH
 		desiredNetworkPolicy := &networkingv1.NetworkPolicy{
 			TypeMeta: metav1.TypeMeta{

pkg/upgrade/upgrade.go

@@ -218,7 +218,7 @@ func CleanupExistingResource(ctx context.Context,
 	oldReleaseVersion cluster.Release,
 ) error {
 	var multiErr *multierror.Error
-	// Special Handling of cleanup of deprecated model monitoring stack
+	// Special Handling of cleanup of deprecated model monitoring stack on managed
 	if platform == cluster.ManagedRhoai {
 		deprecatedDeployments := []string{"rhods-prometheus-operator"}
 		multiErr = multierror.Append(multiErr, deleteDeprecatedResources(ctx, cli, dscMonitoringNamespace, deprecatedDeployments, &appsv1.DeploymentList{}))
@@ -247,9 +247,20 @@ func CleanupExistingResource(ctx context.Context,
 		deprecatedServicemonitors := []string{"modelmesh-federated-metrics"}
 		multiErr = multierror.Append(multiErr, deleteDeprecatedServiceMonitors(ctx, cli, dscMonitoringNamespace, deprecatedServicemonitors))
 	}
+	// Special Handling of cleanup of deprecated SRE monitoring stack on self-managed
+	if platform == cluster.SelfManagedRhoai {
+		deprecatedOperatorSM := []string{"rhods-monitor-federation"}
+		multiErr = multierror.Append(multiErr, deleteDeprecatedServiceMonitors(ctx, cli, dscMonitoringNamespace, deprecatedOperatorSM))
+
+		deprecatedRolebindings := []string{"rhods-prometheus-cluster-monitoring-viewer-binding", "redhat-ods-monitoring"}
+		multiErr = multierror.Append(multiErr, deleteDeprecatedResources(ctx, cli, dscMonitoringNamespace, deprecatedRolebindings, &rbacv1.RoleBindingList{}))
+
+		deprecatedRroles := []string{"redhat-ods-monitoring"}
+		multiErr = multierror.Append(multiErr, deleteDeprecatedResources(ctx, cli, dscMonitoringNamespace, deprecatedRroles, &rbacv1.RoleList{}))
+	}
 	// common logic for both self-managed and managed
-	deprecatedOperatorSM := []string{"rhods-monitor-federation2"}
-	multiErr = multierror.Append(multiErr, deleteDeprecatedServiceMonitors(ctx, cli, dscMonitoringNamespace, deprecatedOperatorSM))
+	deprecatedOperatorSM2 := []string{"rhods-monitor-federation2"}
+	multiErr = multierror.Append(multiErr, deleteDeprecatedServiceMonitors(ctx, cli, dscMonitoringNamespace, deprecatedOperatorSM2))
 
 	// Remove deprecated opendatahub namespace(previously owned by kuberay and Kueue)
 	multiErr = multierror.Append(multiErr, deleteDeprecatedNamespace(ctx, cli, "opendatahub"))