[OTAGENT-980] Reference v1.0.5 WorkloadAllowlist exemption when OTel collector is enabled

internal/controller/datadogagent/experimental/autopilot.go

@@ -64,9 +64,12 @@ func IsAutopilotEnabled(obj metav1.Object) bool {
 	return strings.EqualFold(ann[getExperimentalAnnotationKey(ExperimentalAutopilotSubkey)], "true")
 }
 
+// createAllowlistSynchronizer is overridden in tests to avoid hitting kubeconfig loading.
+var createAllowlistSynchronizer = allowlistsynchronizer.CreateAllowlistSynchronizer
+
 func applyExperimentalAutopilotOverrides(dda metav1.Object, manager feature.PodTemplateManagers) {
 	if IsAutopilotEnabled(dda) {
-		allowlistsynchronizer.CreateAllowlistSynchronizer()
+		createAllowlistSynchronizer(hasOtelAgentContainer(manager))
 
 		if manager.PodTemplateSpec().Labels == nil {
 			manager.PodTemplateSpec().Labels = map[string]string{}
@@ -151,3 +154,15 @@ func applyExperimentalAutopilotOverrides(dda metav1.Object, manager feature.PodT
 		}
 	}
 }
+
+// hasOtelAgentContainer reports whether the otel-agent (ddot-collector) container is
+// present in the PodTemplateSpec, which signals that the OTel collector feature is
+// enabled on the DatadogAgent and the v1.0.5 WorkloadAllowlist exemption is required.
+func hasOtelAgentContainer(manager feature.PodTemplateManagers) bool {
+	for _, c := range manager.PodTemplateSpec().Spec.Containers {
+		if c.Name == string(apicommon.OtelAgent) {
+			return true
+		}
+	}
+	return false
+}

internal/controller/datadogagent/experimental/autopilot_test.go

@@ -0,0 +1,124 @@
+// Unless explicitly stated otherwise all files in this repository are licensed
+// under the Apache License Version 2.0.
+// This product includes software developed at Datadog (https://www.datadoghq.com/).
+// Copyright 2016-present Datadog, Inc.
+
+package experimental
+
+import (
+	"testing"
+
+	apicommon "github.com/DataDog/datadog-operator/api/datadoghq/common"
+	"github.com/DataDog/datadog-operator/internal/controller/datadogagent/feature/fake"
+	"github.com/stretchr/testify/assert"
+	corev1 "k8s.io/api/core/v1"
+	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
+)
+
+// TestHasOtelAgentContainer verifies that the otel-agent container detection used to
+// gate the v1.0.5 WorkloadAllowlist exemption inclusion works for both the present
+// and absent cases. See OTAGENT-980.
+func TestHasOtelAgentContainer(t *testing.T) {
+	tests := []struct {
+		name       string
+		containers []corev1.Container
+		expected   bool
+	}{
+		{
+			name: "no otel-agent container",
+			containers: []corev1.Container{
+				{Name: string(apicommon.CoreAgentContainerName)},
+				{Name: string(apicommon.TraceAgentContainerName)},
+			},
+			expected: false,
+		},
+		{
+			name: "otel-agent container present",
+			containers: []corev1.Container{
+				{Name: string(apicommon.CoreAgentContainerName)},
+				{Name: string(apicommon.OtelAgent)},
+			},
+			expected: true,
+		},
+		{
+			name:       "empty containers",
+			containers: nil,
+			expected:   false,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			manager := fake.NewPodTemplateManagers(t, corev1.PodTemplateSpec{
+				Spec: corev1.PodSpec{Containers: tt.containers},
+			})
+			assert.Equal(t, tt.expected, hasOtelAgentContainer(manager))
+		})
+	}
+}
+
+// TestApplyExperimentalAutopilotOverrides_SynchronizerInvocation verifies that the
+// autopilot override branch invokes the synchronizer creator with the correct flag
+// derived from the rendered pod template (OTAGENT-980).
+func TestApplyExperimentalAutopilotOverrides_SynchronizerInvocation(t *testing.T) {
+	// Stub out the synchronizer creator so tests don't hit kubeconfig loading.
+	origCreator := createAllowlistSynchronizer
+	defer func() { createAllowlistSynchronizer = origCreator }()
+
+	autopilotAnnotation := map[string]string{
+		ExperimentalAnnotationPrefix + "/" + ExperimentalAutopilotSubkey: "true",
+	}
+
+	tests := []struct {
+		name                 string
+		annotations          map[string]string
+		containers           []corev1.Container
+		expectInvoked        bool
+		expectOtelFlagPassed bool
+	}{
+		{
+			name:                 "autopilot disabled does not invoke synchronizer",
+			annotations:          nil,
+			containers:           []corev1.Container{{Name: string(apicommon.CoreAgentContainerName)}},
+			expectInvoked:        false,
+			expectOtelFlagPassed: false,
+		},
+		{
+			name:                 "autopilot enabled without otel-agent invokes synchronizer with false",
+			annotations:          autopilotAnnotation,
+			containers:           []corev1.Container{{Name: string(apicommon.CoreAgentContainerName)}},
+			expectInvoked:        true,
+			expectOtelFlagPassed: false,
+		},
+		{
+			name:                 "autopilot enabled with otel-agent invokes synchronizer with true",
+			annotations:          autopilotAnnotation,
+			containers:           []corev1.Container{{Name: string(apicommon.CoreAgentContainerName)}, {Name: string(apicommon.OtelAgent)}},
+			expectInvoked:        true,
+			expectOtelFlagPassed: true,
+		},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			var invoked bool
+			var passedFlag bool
+			createAllowlistSynchronizer = func(otelCollectorEnabled bool) {
+				invoked = true
+				passedFlag = otelCollectorEnabled
+			}
+
+			dda := &metav1.ObjectMeta{Annotations: tt.annotations}
+			manager := fake.NewPodTemplateManagers(t, corev1.PodTemplateSpec{
+				Spec: corev1.PodSpec{Containers: tt.containers},
+			})
+
+			applyExperimentalAutopilotOverrides(dda, manager)
+
+			assert.Equal(t, tt.expectInvoked, invoked, "synchronizer creator invocation")
+			if tt.expectInvoked {
+				assert.Equal(t, tt.expectOtelFlagPassed, passedFlag, "otelCollectorEnabled flag")
+			}
+		})
+	}
+}

pkg/allowlistsynchronizer/allowlistsynchronizer.go

@@ -29,6 +29,16 @@ var (
 
 var logger = logf.Log.WithName("AllowlistSynchronizer")
 
+// Allowlist paths kept in sync with the Datadog partner exemption YAMLs published in
+// the datadog-gke-workload-allowlist repo.
+const (
+	// allowlistPathV101 exempts the base agent / process-agent / trace-agent containers.
+	allowlistPathV101 = "Datadog/datadog/datadog-datadog-daemonset-exemption-v1.0.1.yaml"
+	// allowlistPathV105 exempts the otel-agent (ddot-collector) container when OTel
+	// feature gates are configured. See OTAGENT-980.
+	allowlistPathV105 = "Datadog/datadog/datadog-datadog-daemonset-exemption-v1.0.5.yaml"
+)
+
 type AllowlistSynchronizer struct {
 	metav1.TypeMeta   `json:",inline"`
 	metav1.ObjectMeta `json:"metadata"`
@@ -46,8 +56,21 @@ type AllowlistSynchronizerSpec struct {
 	AllowlistPaths []string `json:"allowlistPaths,omitempty"`
 }
 
-func createAllowlistSynchronizerResource(k8sClient client.Client) error {
-	obj := &AllowlistSynchronizer{
+// ComputeAllowlistPaths returns the set of Datadog WorkloadAllowlist exemption paths
+// that the AllowlistSynchronizer should reference for the given DatadogAgent.
+//
+// v1.0.5 is included when the OTel collector is enabled so that the otel-agent
+// (ddot-collector image) container is permitted by GKE Autopilot Warden. See OTAGENT-980.
+func ComputeAllowlistPaths(otelCollectorEnabled bool) []string {
+	paths := []string{allowlistPathV101}
+	if otelCollectorEnabled {
+		paths = append(paths, allowlistPathV105)
+	}
+	return paths
+}
+
+func newAllowlistSynchronizer(paths []string) *AllowlistSynchronizer {
+	return &AllowlistSynchronizer{
 		TypeMeta: metav1.TypeMeta{
 			APIVersion: "allowlistsynchronizers.auto.gke.io",
 			Kind:       "AllowlistSynchronizer",
@@ -60,19 +83,22 @@ func createAllowlistSynchronizerResource(k8sClient client.Client) error {
 			},
 		},
 		Spec: AllowlistSynchronizerSpec{
-			AllowlistPaths: []string{
-				"Datadog/datadog/datadog-datadog-daemonset-exemption-v1.0.1.yaml",
-			},
+			AllowlistPaths: paths,
 		},
 	}
+}
 
-	return k8sClient.Create(context.TODO(), obj)
+func createAllowlistSynchronizerResource(k8sClient client.Client, paths []string) error {
+	return k8sClient.Create(context.TODO(), newAllowlistSynchronizer(paths))
 }
 
 // CreateAllowlistSynchronizer creates a GKE AllowlistSynchronizer Custom Resource (auto.gke.io/v1) for the Datadog WorkloadAllowlist if it doesn't exist.
 // The AllowlistSynchronizer is needed so that GKE Autopilot can sync the Datadog WorkloadAllowlist to the cluster. See the CRD reference:
 // https://cloud.google.com/kubernetes-engine/docs/reference/crds/allowlistsynchronizer
-func CreateAllowlistSynchronizer() {
+//
+// otelCollectorEnabled controls whether the v1.0.5 exemption path (which permits the
+// otel-agent / ddot-collector container) is included.
+func CreateAllowlistSynchronizer(otelCollectorEnabled bool) {
 	cfg, configErr := config.GetConfig()
 	if configErr != nil {
 		logger.Error(configErr, "failed to load kubeconfig")
@@ -91,6 +117,13 @@ func CreateAllowlistSynchronizer() {
 		return
 	}
 
+	reconcileAllowlistSynchronizer(k8sClient, otelCollectorEnabled)
+}
+
+// reconcileAllowlistSynchronizer creates the AllowlistSynchronizer resource if it does
+// not already exist, logging any failure. Extracted from CreateAllowlistSynchronizer so
+// it can be exercised with a fake client.
+func reconcileAllowlistSynchronizer(k8sClient client.Client, otelCollectorEnabled bool) {
 	existing := &AllowlistSynchronizer{}
 	if existingErr := k8sClient.Get(context.TODO(), client.ObjectKey{Name: "datadog-synchronizer"}, existing); existingErr == nil {
 		return
@@ -99,7 +132,7 @@ func CreateAllowlistSynchronizer() {
 		return
 	}
 
-	if err := createAllowlistSynchronizerResource(k8sClient); err != nil {
+	if err := createAllowlistSynchronizerResource(k8sClient, ComputeAllowlistPaths(otelCollectorEnabled)); err != nil {
 		if apierrors.IsAlreadyExists(err) {
 			return
 		}