add metrics to track the managed resource count

oliviassss · oliviassss · commit 4e1b9f324bc9 · 2025-01-29T16:02:34.000-08:00
diff --git a/config/webhook/manifests.yaml b/config/webhook/manifests.yaml
@@ -1,4 +1,3 @@
----
 apiVersion: admissionregistration.k8s.io/v1
 kind: MutatingWebhookConfiguration
 metadata:
diff --git a/docs/install/iam_policy.json b/docs/install/iam_policy.json
@@ -42,7 +42,8 @@
                 "elasticloadbalancing:DescribeTags",
                 "elasticloadbalancing:DescribeTrustStores",
                 "elasticloadbalancing:DescribeListenerAttributes",
-                "elasticloadbalancing:DescribeCapacityReservation"
+                "elasticloadbalancing:DescribeCapacityReservation",
+                "tag:GetResources"
             ],
             "Resource": "*"
         },
diff --git a/docs/install/iam_policy_cn.json b/docs/install/iam_policy_cn.json
@@ -42,7 +42,8 @@
                 "elasticloadbalancing:DescribeTags",
                 "elasticloadbalancing:DescribeTrustStores",
                 "elasticloadbalancing:DescribeListenerAttributes",
-                "elasticloadbalancing:DescribeCapacityReservation"
+                "elasticloadbalancing:DescribeCapacityReservation",
+                "tag:GetResources"
             ],
             "Resource": "*"
         },
diff --git a/docs/install/iam_policy_iso.json b/docs/install/iam_policy_iso.json
@@ -39,7 +39,8 @@
                 "elasticloadbalancing:DescribeTargetGroups",
                 "elasticloadbalancing:DescribeTargetGroupAttributes",
                 "elasticloadbalancing:DescribeTargetHealth",
-                "elasticloadbalancing:DescribeTags"
+                "elasticloadbalancing:DescribeTags",
+                "tag:GetResources"
             ],
             "Resource": "*"
         },
diff --git a/docs/install/iam_policy_isob.json b/docs/install/iam_policy_isob.json
@@ -39,7 +39,8 @@
                 "elasticloadbalancing:DescribeTargetGroups",
                 "elasticloadbalancing:DescribeTargetGroupAttributes",
                 "elasticloadbalancing:DescribeTargetHealth",
-                "elasticloadbalancing:DescribeTags"
+                "elasticloadbalancing:DescribeTags",
+                "tag:GetResources"
             ],
             "Resource": "*"
         },
diff --git a/docs/install/iam_policy_isoe.json b/docs/install/iam_policy_isoe.json
@@ -39,7 +39,8 @@
                 "elasticloadbalancing:DescribeTargetGroups",
                 "elasticloadbalancing:DescribeTargetGroupAttributes",
                 "elasticloadbalancing:DescribeTargetHealth",
-                "elasticloadbalancing:DescribeTags"
+                "elasticloadbalancing:DescribeTags",
+                "tag:GetResources"
             ],
             "Resource": "*"
         },
diff --git a/docs/install/iam_policy_isof.json b/docs/install/iam_policy_isof.json
@@ -39,7 +39,8 @@
                 "elasticloadbalancing:DescribeTargetGroups",
                 "elasticloadbalancing:DescribeTargetGroupAttributes",
                 "elasticloadbalancing:DescribeTargetHealth",
-                "elasticloadbalancing:DescribeTags"
+                "elasticloadbalancing:DescribeTags",
+                "tag:GetResources"
             ],
             "Resource": "*"
         },
diff --git a/docs/install/iam_policy_us-gov.json b/docs/install/iam_policy_us-gov.json
@@ -42,7 +42,8 @@
                 "elasticloadbalancing:DescribeTags",
                 "elasticloadbalancing:DescribeTrustStores",
                 "elasticloadbalancing:DescribeListenerAttributes",
-                "elasticloadbalancing:DescribeCapacityReservation"
+                "elasticloadbalancing:DescribeCapacityReservation",
+                "tag:GetResources"
             ],
             "Resource": "*"
         },
diff --git a/main.go b/main.go
@@ -17,26 +17,24 @@ limitations under the License.
 package main
 
 import (
-	"k8s.io/client-go/util/workqueue"
-	"os"
-
-	elbv2deploy "sigs.k8s.io/aws-load-balancer-controller/pkg/deploy/elbv2"
-
 	"github.com/go-logr/logr"
 	"github.com/spf13/pflag"
 	zapraw "go.uber.org/zap"
 	k8sruntime "k8s.io/apimachinery/pkg/runtime"
 	"k8s.io/client-go/kubernetes"
 	clientgoscheme "k8s.io/client-go/kubernetes/scheme"
 	_ "k8s.io/client-go/plugin/pkg/client/auth/gcp"
+	"k8s.io/client-go/util/workqueue"
 	"k8s.io/klog/v2"
+	"os"
 	elbv2api "sigs.k8s.io/aws-load-balancer-controller/apis/elbv2/v1beta1"
 	elbv2controller "sigs.k8s.io/aws-load-balancer-controller/controllers/elbv2"
 	"sigs.k8s.io/aws-load-balancer-controller/controllers/ingress"
 	"sigs.k8s.io/aws-load-balancer-controller/controllers/service"
 	"sigs.k8s.io/aws-load-balancer-controller/pkg/aws"
 	"sigs.k8s.io/aws-load-balancer-controller/pkg/aws/throttle"
 	"sigs.k8s.io/aws-load-balancer-controller/pkg/config"
+	elbv2deploy "sigs.k8s.io/aws-load-balancer-controller/pkg/deploy/elbv2"
 	"sigs.k8s.io/aws-load-balancer-controller/pkg/inject"
 	"sigs.k8s.io/aws-load-balancer-controller/pkg/k8s"
 	awsmetrics "sigs.k8s.io/aws-load-balancer-controller/pkg/metrics/aws"
@@ -84,8 +82,6 @@ func main() {
 	klog.SetLoggerWithOptions(appLogger, klog.ContextualLogger(true))
 
 	var awsMetricsCollector *awsmetrics.Collector
-	lbcMetricsCollector := lbcmetrics.NewCollector(metrics.Registry)
-
 	if metrics.Registry != nil {
 		awsMetricsCollector = awsmetrics.NewCollector(metrics.Registry)
 	}
@@ -107,6 +103,17 @@ func main() {
 		os.Exit(1)
 	}
 
+	// track the k8s resources with finalizers contains "k8s.aws"
+	// track the aws resources with cluster tag "elbv2.k8s.aws/cluster=$ClusterName"
+	lbcMetricsCollector := lbcmetrics.NewCollector(
+		metrics.Registry,
+		mgr.GetClient(),
+		cloud.RGT(),
+		"k8s.aws",
+		"elbv2.k8s.aws/cluster",
+		controllerCFG.ClusterName,
+	)
+
 	clientSet, err := kubernetes.NewForConfig(mgr.GetConfig())
 	if err != nil {
 		setupLog.Error(err, "unable to obtain clientSet")
@@ -210,6 +217,19 @@ func main() {
 		setupLog.Error(err, "problem running manager")
 		os.Exit(1)
 	}
+
+	// update of the managed resource metrics
+	go func() {
+		if err := lbcMetricsCollector.UpdateManagedK8sResourceMetrics(ctx); err != nil {
+			setupLog.Error(err, "failed to update managed Kubernetes resource metrics")
+		}
+		if err := lbcMetricsCollector.UpdateManagedALBMetrics(ctx); err != nil {
+			setupLog.Error(err, "failed to update managed ALB metrics")
+		}
+		if err := lbcMetricsCollector.UpdateManagedNLBMetrics(ctx); err != nil {
+			setupLog.Error(err, "failed to update managed NLB metrics")
+		}
+	}()
 }
 
 // loadControllerConfig loads the controller configuration.
diff --git a/pkg/metrics/aws/instruments.go b/pkg/metrics/aws/instruments.go
@@ -59,8 +59,13 @@ func newInstruments(registerer prometheus.Registerer) *instruments {
 		Name:      metricAPIRequestDurationSeconds,
 		Help:      "Latency of an individual HTTP request to the service endpoint",
 	}, []string{labelService, labelOperation})
-
-	registerer.MustRegister(apiCallsTotal, apiCallDurationSeconds, apiCallRetries, apiRequestsTotal, apiRequestDurationSecond)
+	registerer.MustRegister(
+		apiCallsTotal,
+		apiCallDurationSeconds,
+		apiCallRetries,
+		apiRequestsTotal,
+		apiRequestDurationSecond,
+	)
 	return &instruments{
 		apiCallsTotal:            apiCallsTotal,
 		apiCallDurationSeconds:   apiCallDurationSeconds,
diff --git a/pkg/metrics/lbc/collector.go b/pkg/metrics/lbc/collector.go
@@ -1,33 +1,81 @@
 package lbc
 
 import (
+	"context"
+	awssdk "github.com/aws/aws-sdk-go-v2/aws"
+	rgtsdk "github.com/aws/aws-sdk-go-v2/service/resourcegroupstaggingapi"
+	rgttypes "github.com/aws/aws-sdk-go-v2/service/resourcegroupstaggingapi/types"
 	"github.com/prometheus/client_golang/prometheus"
+	corev1 "k8s.io/api/core/v1"
+	networkingv1 "k8s.io/api/networking/v1"
+	elbv2api "sigs.k8s.io/aws-load-balancer-controller/apis/elbv2/v1beta1"
+	"sigs.k8s.io/aws-load-balancer-controller/pkg/aws/services"
+	"strings"
+
+	"sigs.k8s.io/controller-runtime/pkg/client"
 	"time"
 )
 
+const (
+	networkLoadBalancerStr = "nlb"
+	resourceTypeALB        = "elasticloadbalancing:loadbalancer/app"
+	resourceTypeNLB        = "elasticloadbalancing:loadbalancer/net"
+)
+
 type MetricCollector interface {
 	// ObservePodReadinessGateReady this metric is useful to determine how fast pods are becoming ready in the load balancer.
 	// Due to some architectural constraints, we can only emit this metric for pods that are using readiness gates.
 	ObservePodReadinessGateReady(namespace string, tgbName string, duration time.Duration)
+
+	// UpdateManagedK8sResourceMetrics fetches and updates managed k8s resources metrics.
+	UpdateManagedK8sResourceMetrics(ctx context.Context) error
+
+	// UpdateManagedALBMetrics updates managed ALB count metrics
+	UpdateManagedALBMetrics(ctx context.Context) error
+
+	//UpdateManagedNLBMetrics updates managed NLB count metrics
+	UpdateManagedNLBMetrics(ctx context.Context) error
 }
 
 type collector struct {
-	instruments *instruments
+	instruments      *instruments
+	runtimeClient    client.Client
+	rgt              services.RGT
+	finalizerKeyWord string
+	clusterTagKey    string
+	clusterTagVal    string
 }
 
 type noOpCollector struct{}
 
 func (n *noOpCollector) ObservePodReadinessGateReady(_ string, _ string, _ time.Duration) {
 }
 
-func NewCollector(registerer prometheus.Registerer) MetricCollector {
-	if registerer == nil {
+func (n *noOpCollector) UpdateManagedK8sResourceMetrics(_ context.Context) error {
+	return nil
+}
+
+func (n *noOpCollector) UpdateManagedALBMetrics(_ context.Context) error {
+	return nil
+}
+
+func (n *noOpCollector) UpdateManagedNLBMetrics(_ context.Context) error {
+	return nil
+}
+
+func NewCollector(registerer prometheus.Registerer, runtimeClient client.Client, rgt services.RGT, finalizerKeyWord string, clusterTagKey string, clusterTagVal string) MetricCollector {
+	if registerer == nil || runtimeClient == nil {
 		return &noOpCollector{}
 	}
 
 	instruments := newInstruments(registerer)
 	return &collector{
-		instruments: instruments,
+		instruments:      instruments,
+		runtimeClient:    runtimeClient,
+		rgt:              rgt,
+		finalizerKeyWord: finalizerKeyWord,
+		clusterTagKey:    clusterTagKey,
+		clusterTagVal:    clusterTagVal,
 	}
 }
 
@@ -37,3 +85,99 @@ func (c *collector) ObservePodReadinessGateReady(namespace string, tgbName strin
 		labelName:      tgbName,
 	}).Observe(duration.Seconds())
 }
+
+func (c *collector) UpdateManagedK8sResourceMetrics(ctx context.Context) error {
+	listOpts := &client.ListOptions{
+		Namespace: "",
+	}
+	ingressCount, serviceCount, tgbCount := 0, 0, 0
+	// Fetch ingress count
+	ingressList := &networkingv1.IngressList{}
+	err := c.runtimeClient.List(ctx, ingressList, listOpts)
+	if err != nil {
+		return err
+	}
+	for _, ingress := range ingressList.Items {
+		for _, finalizer := range ingress.Finalizers {
+			if strings.Contains(finalizer, c.finalizerKeyWord) {
+				ingressCount++
+				break
+			}
+		}
+	}
+	c.instruments.managedIngressCount.Set(float64(ingressCount))
+
+	// Fetch service count
+	serviceList := &corev1.ServiceList{}
+	err = c.runtimeClient.List(ctx, serviceList, listOpts)
+	if err != nil {
+		return err
+	}
+	for _, service := range serviceList.Items {
+		hasMatchingFinalizer := false
+		for _, finalizer := range service.Finalizers {
+			if strings.Contains(finalizer, c.finalizerKeyWord) {
+				hasMatchingFinalizer = true
+				break
+			}
+		}
+
+		if hasMatchingFinalizer && service.Spec.LoadBalancerClass != nil && strings.Contains(*service.Spec.LoadBalancerClass, networkLoadBalancerStr) {
+			serviceCount++
+		}
+	}
+	c.instruments.managedServiceCount.Set(float64(serviceCount))
+
+	// Fetch TargetGroupBinding count
+	tgbList := &elbv2api.TargetGroupBindingList{}
+	err = c.runtimeClient.List(ctx, tgbList, listOpts)
+	if err != nil {
+		return err
+	}
+	for _, tgb := range tgbList.Items {
+		for _, finalizer := range tgb.Finalizers {
+			if strings.Contains(finalizer, c.finalizerKeyWord) {
+				tgbCount++
+				break
+			}
+		}
+	}
+	c.instruments.managedTGBCount.Set(float64(tgbCount))
+
+	return nil
+}
+
+func (c *collector) UpdateManagedALBMetrics(ctx context.Context) error {
+	count, err := c.getManagedAWSResourceMetrics(ctx, resourceTypeALB)
+	if err != nil {
+		return err
+	}
+	c.instruments.managedALBCount.Set(float64(count))
+	return nil
+}
+
+func (c *collector) UpdateManagedNLBMetrics(ctx context.Context) error {
+	count, err := c.getManagedAWSResourceMetrics(ctx, resourceTypeNLB)
+	if err != nil {
+		return err
+	}
+	c.instruments.managedNLBCount.Set(float64(count))
+	return nil
+}
+
+func (c *collector) getManagedAWSResourceMetrics(ctx context.Context, resourceType string) (count int, err error) {
+	req := &rgtsdk.GetResourcesInput{
+		ResourceTypeFilters: []string{resourceType},
+		TagFilters: []rgttypes.TagFilter{
+			{
+				Key:    awssdk.String(c.clusterTagKey),
+				Values: []string{c.clusterTagVal},
+			},
+		},
+	}
+	resources, err := c.rgt.GetResourcesAsList(ctx, req)
+	if err != nil {
+		return 0, err
+	}
+	return len(resources), nil
+}
diff --git a/pkg/metrics/lbc/instruments.go b/pkg/metrics/lbc/instruments.go
diff --git a/pkg/metrics/lbc/mockcollector.go b/pkg/metrics/lbc/mockcollector.go

Original file line number	Diff line number	Diff line change
`@@ -1,4 +1,3 @@`
`1`		`----`
`2`	`1`	`apiVersion: admissionregistration.k8s.io/v1`
`3`	`2`	`kind: MutatingWebhookConfiguration`
`4`	`3`	`metadata:`