Merge pull request #2710 from openshift-cherrypick-robot/cherry-pick-2684-to-release-4.20

[release-4.20] OCPBUGS-62979: Remove AlertManager endpoints when disabled

Author: openshift-merge-bot[bot]

pkg/manifests/manifests.go

@@ -1448,7 +1448,7 @@ func (f *Factory) PrometheusK8s(grpcTLS *v1.Secret, telemetrySecret *v1.Secret)
 		p.Spec.Thanos.Image = &f.config.Images.Thanos
 	}
 
-	setupAlerting(p, platformAlertmanagerService, f.namespace)
+	setupAlerting(p, platformAlertmanagerService, f.namespace, f.config.ClusterMonitoringConfiguration.AlertmanagerMainConfig.IsEnabled())
 	f.adjustGoGCRelatedConfig(p)
 
 	for i, container := range p.Spec.Containers {
@@ -1514,7 +1514,12 @@ func (f *Factory) adjustGoGCRelatedConfig(p *monv1.Prometheus) {
 	}
 }
 
-func setupAlerting(p *monv1.Prometheus, svcName, svcNamespace string) {
+func setupAlerting(p *monv1.Prometheus, svcName, svcNamespace string, enabled bool) {
+	if !enabled {
+		p.Spec.Alerting.Alertmanagers = []monv1.AlertmanagerEndpoints{}
+		return
+	}
+
 	eps := p.Spec.Alerting.Alertmanagers[0]
 
 	eps.Name = svcName
@@ -1805,9 +1810,9 @@ func (f *Factory) PrometheusUserWorkload(grpcTLS *v1.Secret) (*monv1.Prometheus,
 	f.adjustGoGCRelatedConfig(p)
 
 	if f.config.UserWorkloadConfiguration.Alertmanager.Enabled {
-		setupAlerting(p, userWorkloadAlertmanagerService, f.namespaceUserWorkload)
+		setupAlerting(p, userWorkloadAlertmanagerService, f.namespaceUserWorkload, true)
 	} else {
-		setupAlerting(p, platformAlertmanagerService, f.namespace)
+		setupAlerting(p, platformAlertmanagerService, f.namespace, f.config.ClusterMonitoringConfiguration.AlertmanagerMainConfig.IsEnabled())
 	}
 
 	alertManagerConfigs := f.config.AdditionalAlertmanagerConfigsForPrometheusUserWorkload()

test/e2e/alertmanager_test.go

@@ -702,6 +702,7 @@ func TestAlertmanagerDisabling(t *testing.T) {
 		{name: "assert old service monitor does not exists", assertion: f.AssertServiceMonitorDoesNotExist("alertmanager", f.Ns)},
 		{name: "alertmanager public URL is unset", assertion: f.AssertValueInConfigMapEquals(
 			"monitoring-shared-config", "openshift-config-managed", "alertmanagerPublicURL", "")},
+		{name: "assert prometheus alertmanager endpoints empty", assertion: f.AssertPrometheusAlertmanagerEndpointsEmpty("prometheus-k8s", f.Ns)},
 		{name: "assert operator not degraded", assertion: f.AssertOperatorCondition(statusv1.OperatorDegraded, statusv1.ConditionFalse)},
 	}
 	t.Run("disable alertmanager", func(t *testing.T) {

test/e2e/framework/assertions.go

@@ -311,6 +311,24 @@ func (f *Framework) AssertPrometheusExists(name, namespace string) func(t *testi
 	}
 }
 
+func (f *Framework) AssertPrometheusAlertmanagerEndpointsEmpty(name, namespace string) func(t *testing.T) {
+	return func(t *testing.T) {
+		err := Poll(time.Second, 5*time.Minute, func() error {
+			const prometheusConfigSecretName = "prometheus-k8s"
+			prometheusConfig := f.PrometheusConfigFromSecret(t, namespace, prometheusConfigSecretName)
+
+			if len(prometheusConfig.AlertingConfig.AlertmanagerConfigs) != 0 {
+				return fmt.Errorf("expected no alertmanager endpoints in runtime config, got %d", len(prometheusConfig.AlertingConfig.AlertmanagerConfigs))
+			}
+
+			return nil
+		})
+		if err != nil {
+			t.Fatal(err)
+		}
+	}
+}
+
 type PodAssertion func(pod v1.Pod) error
 
 // AssertPodConfiguration for each pod in the namespace that matches the label selector