chore: add RBAC for endpointslices to Prometheus SA

This change also adds e2e tests for the new EndpointSlice discovery
role.

Signed-off-by: Simon Pasquier <spasquie@redhat.com>

Documentation/rbac.md

@@ -169,6 +169,11 @@ rules:
   resources:
   - configmaps
   verbs: ["get"]
+- apiGroups:
+  - discovery.k8s.io
+  resources:
+  - endpointslices
+  verbs: ["get", "list", "watch"]
 - apiGroups:
   - networking.k8s.io
   resources:

Documentation/user-guides/getting-started.md

@@ -166,6 +166,11 @@ rules:
   resources:
   - configmaps
   verbs: ["get"]
+- apiGroups:
+  - discovery.k8s.io
+  resources:
+  - endpointslices
+  verbs: ["get", "list", "watch"]
 - apiGroups:
   - networking.k8s.io
   resources:

example/rbac/prometheus/prometheus-cluster-role.yaml

@@ -15,6 +15,11 @@ rules:
   resources:
   - configmaps
   verbs: ["get"]
+- apiGroups:
+  - discovery.k8s.io
+  resources:
+  - endpointslices
+  verbs: ["get", "list", "watch"]
 - apiGroups:
   - networking.k8s.io
   resources:

pkg/prometheus/promcfg.go

@@ -70,9 +70,11 @@ type ConfigGenerator struct {
 }
 
 // NewConfigGenerator creates a ConfigGenerator for the provided Prometheus resource.
-func NewConfigGenerator(logger log.Logger,
+func NewConfigGenerator(
+	logger log.Logger,
 	p monitoringv1.PrometheusInterface,
-	endpointSliceSupported bool) (*ConfigGenerator, error) {
+	endpointSliceSupported bool,
+) (*ConfigGenerator, error) {
 	if logger == nil {
 		logger = log.NewNopLogger()
 	}
@@ -96,28 +98,11 @@ func NewConfigGenerator(logger log.Logger,
 		return nil, fmt.Errorf("failed to parse scrape classes: %w", err)
 	}
 
-	endpointSliceConfigured := false // Always assume false to preserve original prometheus-operator behaviour.
-
-	// Check if the user has explicitly set the service discovery role to use.
-	switch serviceDiscoveryRole := ptr.Deref(cpf.ServiceDiscoveryRole, monitoringv1.EndpointsRole); serviceDiscoveryRole {
-	case monitoringv1.EndpointSliceRole:
-		level.Info(logger).Log("msg", "using endpointslice as service discovery role")
-		endpointSliceConfigured = true
-	case monitoringv1.EndpointsRole:
-		level.Info(logger).Log("msg", "using endpoints as service discovery role")
-		endpointSliceConfigured = false
-	default:
-		level.Warn(logger).Log("msg",
-			"unknown service discovery role %q, defaulting to endpoints. Configure serviceDiscoveryRole to 'EndpointSlice' to use endpointslice as service discovery role.",
-			serviceDiscoveryRole)
-		endpointSliceConfigured = false
-	}
-
 	return &ConfigGenerator{
 		logger:                 logger,
 		version:                version,
 		prom:                   p,
-		useEndpointSlice:       endpointSliceConfigured && endpointSliceSupported,
+		useEndpointSlice:       endpointSliceSupported && ptr.Deref(cpf.ServiceDiscoveryRole, monitoringv1.EndpointsRole) == monitoringv1.EndpointSliceRole,
 		scrapeClasses:          scrapeClasses,
 		defaultScrapeClassName: defaultScrapeClassName,
 	}, nil

test/e2e/prometheus_test.go

@@ -2110,41 +2110,58 @@ func testPromWhenDeleteCRDCleanUpViaOwnerRef(t *testing.T) {
 }
 
 func testPromDiscovery(t *testing.T) {
-	t.Parallel()
-	testCtx := framework.NewTestCtx(t)
-	defer testCtx.Cleanup(t)
-	ns := framework.CreateNamespace(context.Background(), t, testCtx)
-	framework.SetupPrometheusRBAC(context.Background(), t, testCtx, ns)
+	for _, tc := range []struct {
+		role *monitoringv1.ServiceDiscoveryRole
+	}{
+		{
+			role: nil,
+		},
+		{
+			role: ptr.To(monitoringv1.EndpointsRole),
+		},
+		{
+			role: ptr.To(monitoringv1.EndpointSliceRole),
+		},
+	} {
+		t.Run(fmt.Sprintf("role=%s", ptr.Deref(tc.role, "<nil>")), func(t *testing.T) {
+			t.Parallel()
+			testCtx := framework.NewTestCtx(t)
+			defer testCtx.Cleanup(t)
+			ns := framework.CreateNamespace(context.Background(), t, testCtx)
+			framework.SetupPrometheusRBAC(context.Background(), t, testCtx, ns)
 
-	prometheusName := "test"
-	group := "servicediscovery-test"
-	svc := framework.MakePrometheusService(prometheusName, group, v1.ServiceTypeClusterIP)
+			prometheusName := "test"
+			group := "servicediscovery-test"
+			svc := framework.MakePrometheusService(prometheusName, group, v1.ServiceTypeClusterIP)
 
-	s := framework.MakeBasicServiceMonitor(group)
-	if _, err := framework.MonClientV1.ServiceMonitors(ns).Create(context.Background(), s, metav1.CreateOptions{}); err != nil {
-		t.Fatal("Creating ServiceMonitor failed: ", err)
-	}
+			s := framework.MakeBasicServiceMonitor(group)
+			if _, err := framework.MonClientV1.ServiceMonitors(ns).Create(context.Background(), s, metav1.CreateOptions{}); err != nil {
+				t.Fatal("Creating ServiceMonitor failed: ", err)
+			}
 
-	p := framework.MakeBasicPrometheus(ns, prometheusName, group, 1)
-	_, err := framework.CreatePrometheusAndWaitUntilReady(context.Background(), ns, p)
-	if err != nil {
-		t.Fatal(err)
-	}
+			p := framework.MakeBasicPrometheus(ns, prometheusName, group, 1)
+			p.Spec.ServiceDiscoveryRole = tc.role
+			_, err := framework.CreatePrometheusAndWaitUntilReady(context.Background(), ns, p)
+			if err != nil {
+				t.Fatal(err)
+			}
 
-	if finalizerFn, err := framework.CreateOrUpdateServiceAndWaitUntilReady(context.Background(), ns, svc); err != nil {
-		t.Fatal(fmt.Errorf("creating prometheus service failed: %w", err))
-	} else {
-		testCtx.AddFinalizerFn(finalizerFn)
-	}
+			if finalizerFn, err := framework.CreateOrUpdateServiceAndWaitUntilReady(context.Background(), ns, svc); err != nil {
+				t.Fatal(fmt.Errorf("creating prometheus service failed: %w", err))
+			} else {
+				testCtx.AddFinalizerFn(finalizerFn)
+			}
 
-	_, err = framework.KubeClient.CoreV1().Secrets(ns).Get(context.Background(), fmt.Sprintf("prometheus-%s", prometheusName), metav1.GetOptions{})
-	if err != nil {
-		t.Fatal("Generated Secret could not be retrieved: ", err)
-	}
+			_, err = framework.KubeClient.CoreV1().Secrets(ns).Get(context.Background(), fmt.Sprintf("prometheus-%s", prometheusName), metav1.GetOptions{})
+			if err != nil {
+				t.Fatal("Generated Secret could not be retrieved: ", err)
+			}
 
-	err = framework.WaitForDiscoveryWorking(context.Background(), ns, svc.Name, prometheusName)
-	if err != nil {
-		t.Fatal(fmt.Errorf("validating Prometheus target discovery failed: %w", err))
+			err = framework.WaitForDiscoveryWorking(context.Background(), ns, svc.Name, prometheusName)
+			if err != nil {
+				t.Fatal(fmt.Errorf("validating Prometheus target discovery failed: %w", err))
+			}
+		})
 	}
 }