refactor(monitoring): remove monitoring resources when disabled (#193)

GreptimeTeam · Oct 16, 2024 · 4da82a7 · 4da82a7
1 parent a236c37
commit 4da82a7
Show file tree

Hide file tree

Showing 7 changed files with 58 additions and 8 deletions.
diff --git a/controllers/greptimedbcluster/controller.go b/controllers/greptimedbcluster/controller.go
@@ -33,6 +33,8 @@ import (
 
 	"github.com/GreptimeTeam/greptimedb-operator/apis/v1alpha1"
 	"github.com/GreptimeTeam/greptimedb-operator/cmd/operator/app/options"
+	"github.com/GreptimeTeam/greptimedb-operator/controllers/common"
+	"github.com/GreptimeTeam/greptimedb-operator/controllers/constant"
 	"github.com/GreptimeTeam/greptimedb-operator/controllers/greptimedbcluster/deployers"
 	"github.com/GreptimeTeam/greptimedb-operator/pkg/deployer"
 )
@@ -171,6 +173,13 @@ func (r *Reconciler) Reconcile(ctx context.Context, req ctrl.Request) (ctrl.Resu
 		return ctrl.Result{}, err
 	}
 
+	// FIXME(zyy17): The following code should be elegant to move to the deployers.
+	if !cluster.Spec.Monitoring.IsEnabled() {
+		if err := r.removeMonitoringDB(ctx, cluster); err != nil {
+			return ctrl.Result{}, err
+		}
+	}
+
 	return r.sync(ctx, cluster)
 }
 
@@ -298,3 +307,25 @@ func (r *Reconciler) setObservedGeneration(cluster *v1alpha1.GreptimeDBCluster)
 		cluster.Status.ObservedGeneration = &generation
 	}
 }
+
+func (r *Reconciler) removeMonitoringDB(ctx context.Context, cluster *v1alpha1.GreptimeDBCluster) error {
+	objects := map[string]client.Object{
+		common.MonitoringServiceName(cluster.Name): new(v1alpha1.GreptimeDBStandalone),
+		constant.DefaultVectorConfigName:           new(corev1.ConfigMap),
+	}
+
+	for name, obj := range objects {
+		if err := r.Get(ctx, client.ObjectKey{Namespace: cluster.Namespace, Name: name}, obj); err != nil {
+			if k8serrors.IsNotFound(err) {
+				continue
+			}
+			return err
+		}
+
+		if err := r.Delete(ctx, obj); err != nil {
+			return err
+		}
+	}
+
+	return nil
+}
diff --git a/controllers/greptimedbcluster/deployers/datanode.go b/controllers/greptimedbcluster/deployers/datanode.go
@@ -459,7 +459,7 @@ func (b *datanodeBuilder) generatePodTemplateSpec() corev1.PodTemplateSpec {
 		b.AddLogsVolume(podTemplateSpec, logging.GetLogsDir())
 	}
 
-	if b.Cluster.GetMonitoring() != nil && b.Cluster.GetMonitoring().GetVector() != nil {
+	if b.Cluster.GetMonitoring().IsEnabled() && b.Cluster.GetMonitoring().GetVector() != nil {
 		b.AddVectorConfigVolume(podTemplateSpec)
 		b.AddVectorSidecar(podTemplateSpec, v1alpha1.DatanodeComponentKind)
 	}

diff --git a/controllers/greptimedbcluster/deployers/flownode.go b/controllers/greptimedbcluster/deployers/flownode.go
@@ -260,7 +260,7 @@ func (b *flownodeBuilder) generatePodTemplateSpec() corev1.PodTemplateSpec {
 		b.AddLogsVolume(podTemplateSpec, logging.GetLogsDir())
 	}
 
-	if b.Cluster.GetMonitoring() != nil && b.Cluster.GetMonitoring().GetVector() != nil {
+	if b.Cluster.GetMonitoring().IsEnabled() && b.Cluster.GetMonitoring().GetVector() != nil {
 		b.AddVectorConfigVolume(podTemplateSpec)
 		b.AddVectorSidecar(podTemplateSpec, v1alpha1.FlownodeComponentKind)
 	}

diff --git a/controllers/greptimedbcluster/deployers/frontend.go b/controllers/greptimedbcluster/deployers/frontend.go
@@ -279,7 +279,7 @@ func (b *frontendBuilder) generatePodTemplateSpec() *corev1.PodTemplateSpec {
 		b.AddLogsVolume(podTemplateSpec, logging.GetLogsDir())
 	}
 
-	if b.Cluster.GetMonitoring() != nil && b.Cluster.GetMonitoring().GetVector() != nil {
+	if b.Cluster.GetMonitoring().IsEnabled() && b.Cluster.GetMonitoring().GetVector() != nil {
 		b.AddVectorConfigVolume(podTemplateSpec)
 		b.AddVectorSidecar(podTemplateSpec, v1alpha1.FrontendComponentKind)
 	}

diff --git a/controllers/greptimedbcluster/deployers/meta.go b/controllers/greptimedbcluster/deployers/meta.go
@@ -329,7 +329,7 @@ func (b *metaBuilder) generatePodTemplateSpec() *corev1.PodTemplateSpec {
 		b.AddLogsVolume(podTemplateSpec, logging.GetLogsDir())
 	}
 
-	if b.Cluster.GetMonitoring() != nil && b.Cluster.GetMonitoring().GetVector() != nil {
+	if b.Cluster.GetMonitoring().IsEnabled() && b.Cluster.GetMonitoring().GetVector() != nil {
 		b.AddVectorConfigVolume(podTemplateSpec)
 		b.AddVectorSidecar(podTemplateSpec, v1alpha1.MetaComponentKind)
 	}

diff --git a/controllers/greptimedbcluster/deployers/monitoring.go b/controllers/greptimedbcluster/deployers/monitoring.go
@@ -74,7 +74,7 @@ func (d *MonitoringDeployer) CheckAndUpdateStatus(ctx context.Context, crdObject
 		return false, err
 	}
 
-	if cluster.GetMonitoring() == nil || cluster.GetMonitoring().GetStandalone() == nil {
+	if !cluster.GetMonitoring().IsEnabled() || cluster.GetMonitoring().GetStandalone() == nil {
 		return true, nil
 	}
 
@@ -92,7 +92,7 @@ func (d *MonitoringDeployer) CheckAndUpdateStatus(ctx context.Context, crdObject
 		return false, nil
 	}
 
-	if standalone.Status.StandalonePhase == v1alpha1.PhaseRunning {
+	if cluster.GetMonitoring().IsEnabled() && standalone.Status.StandalonePhase == v1alpha1.PhaseRunning {
 		if err := d.createPipeline(cluster); err != nil {
 			klog.Errorf("failed to create pipeline for standalone, err: '%v'", err)
 			return false, err
@@ -196,7 +196,7 @@ type monitoringBuilder struct {
 }
 
 func (b *monitoringBuilder) BuildGreptimeDBStandalone() deployer.Builder {
-	if b.Cluster.GetMonitoring() == nil || b.Cluster.GetMonitoring().GetStandalone() == nil {
+	if !b.Cluster.GetMonitoring().IsEnabled() || b.Cluster.GetMonitoring().GetStandalone() == nil {
 		return b
 	}
 
@@ -222,7 +222,7 @@ func (b *monitoringBuilder) BuildGreptimeDBStandalone() deployer.Builder {
 }
 
 func (b *monitoringBuilder) BuildConfigMap() deployer.Builder {
-	if b.Cluster.GetMonitoring() == nil || b.Cluster.GetMonitoring().GetVector() == nil {
+	if !b.Cluster.GetMonitoring().IsEnabled() || b.Cluster.GetMonitoring().GetVector() == nil {
 		return b
 	}
 

diff --git a/tests/e2e/greptimedbcluster/test_cluster_enable_monitoring.go b/tests/e2e/greptimedbcluster/test_cluster_enable_monitoring.go
@@ -94,6 +94,25 @@ func TestClusterEnableMonitoring(ctx context.Context, h *helper.Helper) {
 	err = testMonitoringStandalone(ctx, monitoringAddr)
 	Expect(err).NotTo(HaveOccurred(), "failed to test monitoring")
 
+	// Disable monitoring.
+	By("Disable monitoring")
+	testCluster.Spec.Monitoring.Enabled = false
+	err = h.Update(ctx, testCluster)
+	Expect(err).NotTo(HaveOccurred(), "failed to update cluster")
+	By("Check the status of testCluster")
+	Eventually(func() error {
+		clusterPhase, err := h.GetPhase(ctx, testCluster.Namespace, testCluster.Name, new(greptimev1alpha1.GreptimeDBCluster))
+		if err != nil {
+			return err
+		}
+
+		if clusterPhase != greptimev1alpha1.PhaseRunning {
+			return fmt.Errorf("cluster is not running")
+		}
+
+		return nil
+	}, helper.DefaultTimeout, time.Second).ShouldNot(HaveOccurred())
+
 	By("Kill the port forwarding process")
 	h.KillPortForwardProcess()