Track all NGF pods

Author: sjberman

deploy/helm-chart/templates/rbac.yaml

@@ -66,6 +66,7 @@ rules:
   - replicasets
   verbs:
   - get
+  - list
 - apiGroups:
   - discovery.k8s.io
   resources:

internal/mode/static/telemetry/collector.go

@@ -96,7 +96,7 @@ func (c DataCollectorImpl) Collect(ctx context.Context) (Data, error) {
 		return Data{}, fmt.Errorf("failed to collect NGF resource counts: %w", err)
 	}
 
-	ngfReplicaCount, err := CollectNGFReplicaCount(ctx, c.cfg.K8sClientReader, c.cfg.PodNSName)
+	ngfReplicaCount, err := collectNGFReplicaCount(ctx, c.cfg.K8sClientReader, c.cfg.PodNSName)
 	if err != nil {
 		return Data{}, fmt.Errorf("failed to collect NGF replica count: %w", err)
 	}
@@ -168,8 +168,7 @@ func collectGraphResourceCount(
 	return ngfResourceCounts, nil
 }
 
-// CollectNGFReplicaCount returns the number of NGF replicas.
-func CollectNGFReplicaCount(ctx context.Context, k8sClient client.Reader, podNSName types.NamespacedName) (int, error) {
+func collectNGFReplicaCount(ctx context.Context, k8sClient client.Reader, podNSName types.NamespacedName) (int, error) {
 	var pod v1.Pod
 	if err := k8sClient.Get(
 		ctx,

internal/mode/static/usage/job_worker.go

@@ -2,9 +2,10 @@ package usage
 
 import (
 	"context"
+	"fmt"
 
 	"github.com/go-logr/logr"
-	"k8s.io/apimachinery/pkg/types"
+	appsv1 "k8s.io/api/apps/v1"
 	"sigs.k8s.io/controller-runtime/pkg/client"
 
 	"github.com/nginxinc/nginx-gateway-fabric/internal/mode/static/config"
@@ -23,14 +24,7 @@ func CreateUsageJobWorker(
 			logger.Error(err, "Failed to collect node count")
 		}
 
-		podCount, err := telemetry.CollectNGFReplicaCount(
-			ctx,
-			k8sClient,
-			types.NamespacedName{
-				Namespace: cfg.GatewayPodConfig.Namespace,
-				Name:      cfg.GatewayPodConfig.Name,
-			},
-		)
+		podCount, err := GetTotalNGFPodCount(ctx, k8sClient)
 		if err != nil {
 			logger.Error(err, "Failed to collect replica count")
 		}
@@ -60,3 +54,28 @@ func CreateUsageJobWorker(
 		}
 	}
 }
+
+// GetTotalNGFPodCount returns the total count of NGF Pods in the cluster.
+// Uses the "app.kubernetes.io/name" label with either value of "nginx-gateway" or "nginx-gateway-fabric".
+func GetTotalNGFPodCount(ctx context.Context, k8sClient client.Reader) (int, error) {
+	labelKey := "app.kubernetes.io/name"
+	labelVals := map[string]struct{}{
+		"nginx-gateway-fabric": {},
+		"nginx-gateway":        {},
+	}
+
+	var rsList appsv1.ReplicaSetList
+	if err := k8sClient.List(ctx, &rsList, client.HasLabels{labelKey}); err != nil {
+		return 0, fmt.Errorf("failed to list replicasets: %w", err)
+	}
+
+	var count int
+	for _, rs := range rsList.Items {
+		val := rs.Labels[labelKey]
+		if _, ok := labelVals[val]; ok && rs.Spec.Replicas != nil {
+			count += int(*rs.Spec.Replicas)
+		}
+	}
+
+	return count, nil
+}

internal/mode/static/usage/job_worker_test.go

@@ -25,6 +25,9 @@ func TestCreateUsageJobWorker(t *testing.T) {
 		ObjectMeta: metav1.ObjectMeta{
 			Namespace: "nginx-gateway",
 			Name:      "ngf-replicaset",
+			Labels: map[string]string{
+				"app.kubernetes.io/name": "nginx-gateway",
+			},
 		},
 		Spec: appsv1.ReplicaSetSpec{
 			Replicas: &replicas,
@@ -90,3 +93,53 @@ func TestCreateUsageJobWorker(t *testing.T) {
 	_, data := reporter.ReportArgsForCall(0)
 	g.Expect(data).To(Equal(expData))
 }
+
+func TestGetTotalNGFPodCount(t *testing.T) {
+	g := NewWithT(t)
+
+	rs1Replicas := int32(1)
+	rs1 := &appsv1.ReplicaSet{
+		ObjectMeta: metav1.ObjectMeta{
+			Namespace: "nginx-gateway",
+			Name:      "ngf-replicaset1",
+			Labels: map[string]string{
+				"app.kubernetes.io/name": "nginx-gateway",
+			},
+		},
+		Spec: appsv1.ReplicaSetSpec{
+			Replicas: &rs1Replicas,
+		},
+	}
+
+	rs2Replicas := int32(3)
+	rs2 := &appsv1.ReplicaSet{
+		ObjectMeta: metav1.ObjectMeta{
+			Namespace: "nginx-gateway-2",
+			Name:      "ngf-replicaset2",
+			Labels: map[string]string{
+				"app.kubernetes.io/name": "nginx-gateway-fabric",
+			},
+		},
+		Spec: appsv1.ReplicaSetSpec{
+			Replicas: &rs2Replicas,
+		},
+	}
+
+	rs3Replicas := int32(5)
+	rs3 := &appsv1.ReplicaSet{
+		ObjectMeta: metav1.ObjectMeta{
+			Namespace: "default",
+			Name:      "not-ngf",
+		},
+		Spec: appsv1.ReplicaSetSpec{
+			Replicas: &rs3Replicas,
+		},
+	}
+
+	k8sClient := fake.NewFakeClient(rs1, rs2, rs3)
+
+	expCount := 4
+	count, err := usage.GetTotalNGFPodCount(context.Background(), k8sClient)
+	g.Expect(err).ToNot(HaveOccurred())
+	g.Expect(count).To(Equal(expCount))
+}

site/content/installation/usage-reporting.md

@@ -83,6 +83,13 @@ If using manifests, the following command-line options should be set as necessar
 - `--usage-report-server-url` is the base server URL of the NGINX Instance Manager. This field is required.
 - `--usage-report-cluster-name` is an optional display name in the API for the usage data object.
 
+Your NGINX Gateway Fabric Pods should also have one of the following labels:
+
+- `app.kubernetes.io/name=nginx-gateway`
+- `app.kubernetes.io/name=nginx-gateway-fabric`
+
+{{< note >}}The default installation of NGINX Gateway Fabric already includes at least one of these labels.{{< /note >}}
+
 ## Viewing Usage Data from the NGINX Instance Manager API
 
 NGINX Gateway Fabric sends the number of its instances and nodes in the cluster to NGINX Instance Manager every 24 hours. To view the usage data, query the NGINX Instance Manager API. The usage data is available at the following endpoint (replace `nim.example.com` with your server URL, and set the proper credentials in the `--user` field):