Add collection of deployment replica count

Author: bjee19

deploy/helm-chart/templates/rbac.yaml

@@ -24,6 +24,7 @@ rules:
   # FIXME(bjee19): make nodes permission dependent on telemetry being enabled.
   # https://github.com/nginxinc/nginx-gateway-fabric/issues/1317.
   - nodes
+  - pods
   verbs:
   - list
   - watch
@@ -34,6 +35,13 @@ rules:
   verbs:
   - create
   - patch
+- apiGroups:
+    - "apps"
+  resources:
+    - replicasets
+  verbs:
+    - list
+    - watch
 - apiGroups:
   - discovery.k8s.io
   resources:

deploy/manifests/nginx-gateway.yaml

@@ -35,6 +35,7 @@ rules:
   # FIXME(bjee19): make nodes permission dependent on telemetry being enabled.
   # https://github.com/nginxinc/nginx-gateway-fabric/issues/1317.
   - nodes
+  - pods
   verbs:
   - list
   - watch
@@ -45,6 +46,13 @@ rules:
   verbs:
   - create
   - patch
+- apiGroups:
+    - "apps"
+  resources:
+    - replicasets
+  verbs:
+    - list
+    - watch
 - apiGroups:
   - discovery.k8s.io
   resources:

internal/mode/static/manager.go

@@ -8,6 +8,7 @@ import (
 	"github.com/go-logr/logr"
 	ngxclient "github.com/nginxinc/nginx-plus-go-client/client"
 	"github.com/prometheus/client_golang/prometheus"
+	appsv1 "k8s.io/api/apps/v1"
 	apiv1 "k8s.io/api/core/v1"
 	discoveryV1 "k8s.io/api/discovery/v1"
 	apiext "k8s.io/apiextensions-apiserver/pkg/apis/apiextensions/v1"
@@ -65,6 +66,7 @@ func init() {
 	utilruntime.Must(discoveryV1.AddToScheme(scheme))
 	utilruntime.Must(ngfAPI.AddToScheme(scheme))
 	utilruntime.Must(apiext.AddToScheme(scheme))
+	utilruntime.Must(appsv1.AddToScheme(scheme))
 }
 
 // nolint:gocyclo
@@ -218,6 +220,10 @@ func StartManager(cfg config.Config) error {
 		GraphGetter:         processor,
 		ConfigurationGetter: eventHandler,
 		Version:             cfg.Version,
+		PodNSName: types.NamespacedName{
+			Namespace: cfg.GatewayPodConfig.Namespace,
+			Name:      cfg.LeaderElection.Identity,
+		},
 	})
 	if err = mgr.Add(createTelemetryJob(cfg, dataCollector, nginxChecker.getReadyCh())); err != nil {
 		return fmt.Errorf("cannot register telemetry job: %w", err)

internal/mode/static/telemetry/collector.go

@@ -5,7 +5,9 @@ import (
 	"errors"
 	"fmt"
 
+	appsv1 "k8s.io/api/apps/v1"
 	v1 "k8s.io/api/core/v1"
+	"k8s.io/apimachinery/pkg/types"
 	"sigs.k8s.io/controller-runtime/pkg/client"
 
 	"github.com/nginxinc/nginx-gateway-fabric/internal/mode/static/state/dataplane"
@@ -49,6 +51,7 @@ type Data struct {
 	ProjectMetadata   ProjectMetadata
 	NodeCount         int
 	NGFResourceCounts NGFResourceCounts
+	NGFReplicaCount   int
 }
 
 // DataCollectorConfig holds configuration parameters for DataCollectorImpl.
@@ -61,6 +64,8 @@ type DataCollectorConfig struct {
 	ConfigurationGetter ConfigurationGetter
 	// Version is the NGF version.
 	Version string
+	// PodNSName is the NamespacedName of the NGF Pod.
+	PodNSName types.NamespacedName
 }
 
 // DataCollectorImpl is am implementation of DataCollector.
@@ -89,13 +94,19 @@ 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)
+	if err != nil {
+		return Data{}, fmt.Errorf("failed to collect NGF replica count: %w", err)
+	}
+
 	data := Data{
 		NodeCount:         nodeCount,
 		NGFResourceCounts: graphResourceCount,
 		ProjectMetadata: ProjectMetadata{
 			Name:    "NGF",
 			Version: c.cfg.Version,
 		},
+		NGFReplicaCount: ngfReplicaCount,
 	}
 
 	return data, nil
@@ -147,3 +158,36 @@ func collectGraphResourceCount(
 
 	return ngfResourceCounts, nil
 }
+
+func collectNGFReplicaCount(ctx context.Context, k8sClient client.Reader, podNSName types.NamespacedName) (int, error) {
+	var pod v1.Pod
+	if err := k8sClient.Get(ctx,
+		types.NamespacedName{Namespace: podNSName.Namespace, Name: podNSName.Name},
+		&pod,
+	); err != nil {
+		return 0, err
+	}
+
+	podOwnerRefs := pod.GetOwnerReferences()
+	if podOwnerRefs == nil {
+		return 0, fmt.Errorf("could not get owner reference of NGF Pod")
+	}
+	if len(podOwnerRefs) != 1 {
+		return 0, fmt.Errorf("multiple owner references of NGF Pod")
+	}
+
+	switch kind := podOwnerRefs[0].Kind; kind {
+	case "ReplicaSet":
+		var replicaSet appsv1.ReplicaSet
+		if err := k8sClient.Get(ctx,
+			types.NamespacedName{Namespace: podNSName.Namespace, Name: podOwnerRefs[0].Name},
+			&replicaSet,
+		); err != nil {
+			return 0, err
+		}
+
+		return int(*replicaSet.Spec.Replicas), nil
+	default:
+		return 0, fmt.Errorf("pod owner reference was not ReplicaSet, instead was %s", kind)
+	}
+}

internal/mode/static/telemetry/collector_test.go

@@ -48,6 +48,7 @@ var _ = Describe("Collector", Ordered, func() {
 		version                 string
 		expData                 telemetry.Data
 		ctx                     context.Context
+		podNSName               types.NamespacedName
 	)
 
 	BeforeAll(func() {
@@ -60,11 +61,16 @@ var _ = Describe("Collector", Ordered, func() {
 			ProjectMetadata:   telemetry.ProjectMetadata{Name: "NGF", Version: version},
 			NodeCount:         0,
 			NGFResourceCounts: telemetry.NGFResourceCounts{},
+			NGFReplicaCount:   0,
 		}
 
 		k8sClientReader = &eventsfakes.FakeReader{}
 		fakeGraphGetter = &telemetryfakes.FakeGraphGetter{}
 		fakeConfigurationGetter = &telemetryfakes.FakeConfigurationGetter{}
+		podNSName = types.NamespacedName{
+			Namespace: "nginx-gateway",
+			Name:      "ngf-pod",
+		}
 
 		fakeGraphGetter.GetLatestGraphReturns(&graph.Graph{})
 		fakeConfigurationGetter.GetLatestConfigurationReturns(&dataplane.Configuration{})
@@ -74,6 +80,7 @@ var _ = Describe("Collector", Ordered, func() {
 			GraphGetter:         fakeGraphGetter,
 			ConfigurationGetter: fakeConfigurationGetter,
 			Version:             version,
+			PodNSName:           podNSName,
 		})
 	})