Fix metrics issues

manager/cmd/manager/main.go

@@ -37,7 +37,7 @@ import (
 	managerconfig "github.com/stolostron/multicluster-global-hub/manager/pkg/config"
 	"github.com/stolostron/multicluster-global-hub/manager/pkg/cronjob"
 	"github.com/stolostron/multicluster-global-hub/manager/pkg/eventcollector"
-	globalhubmetrics "github.com/stolostron/multicluster-global-hub/manager/pkg/metrics"
+	"github.com/stolostron/multicluster-global-hub/manager/pkg/monitoring"
 	"github.com/stolostron/multicluster-global-hub/manager/pkg/nonk8sapi"
 	managerscheme "github.com/stolostron/multicluster-global-hub/manager/pkg/scheme"
 	"github.com/stolostron/multicluster-global-hub/manager/pkg/specsyncer"
@@ -73,6 +73,7 @@ var (
 
 func init() {
 	managerscheme.AddToScheme(scheme)
+	monitoring.RegisterMetrics()
 }
 
 func parseFlags() *managerconfig.ManagerConfig {
@@ -272,10 +273,6 @@ func createManager(ctx context.Context, restConfig *rest.Config, managerConfig *
 		return nil, fmt.Errorf("failed to add event collector: %w", err)
 	}
 
-	if err := mgr.Add(globalhubmetrics.NewGlobalHubMetrics()); err != nil {
-		return nil, fmt.Errorf("failed to add metrics to manager: %w", err)
-	}
-
 	return mgr, nil
 }
 

manager/pkg/cronjob/scheduler.go

@@ -11,6 +11,7 @@ import (
 
 	"github.com/stolostron/multicluster-global-hub/manager/pkg/config"
 	"github.com/stolostron/multicluster-global-hub/manager/pkg/cronjob/task"
+	"github.com/stolostron/multicluster-global-hub/manager/pkg/monitoring"
 )
 
 const (
@@ -73,6 +74,9 @@ func AddSchedulerToManager(ctx context.Context, mgr ctrl.Manager,
 
 func (s *GlobalHubJobScheduler) Start(ctx context.Context) error {
 	s.log.Info("start job scheduler")
+	// Set the status of the job to 0 (success) when the job is started.
+	monitoring.GlobalHubCronJobGaugeVec.WithLabelValues(task.RetentionTaskName).Set(0)
+	monitoring.GlobalHubCronJobGaugeVec.WithLabelValues(task.LocalComplianceTaskName).Set(0)
 	s.scheduler.StartAsync()
 	if err := s.execJobs(ctx); err != nil {
 		return err

manager/pkg/cronjob/task/data_retention.go

@@ -9,7 +9,7 @@ import (
 	"github.com/go-co-op/gocron"
 	ctrl "sigs.k8s.io/controller-runtime"
 
-	"github.com/stolostron/multicluster-global-hub/manager/pkg/metrics"
+	"github.com/stolostron/multicluster-global-hub/manager/pkg/monitoring"
 	"github.com/stolostron/multicluster-global-hub/pkg/database"
 	"github.com/stolostron/multicluster-global-hub/pkg/database/models"
 )
@@ -42,19 +42,15 @@ var (
 	retentionLog = ctrl.Log.WithName(RetentionTaskName)
 )
 
-func init() {
-	metrics.GlobalHubCronJobGaugeVec.WithLabelValues(RetentionTaskName).Set(0)
-}
-
 func DataRetention(ctx context.Context, retentionMonth int, job gocron.Job) {
 	currentTime := time.Now()
 
 	var err error
 	defer func() {
 		if err != nil {
-			metrics.GlobalHubCronJobGaugeVec.WithLabelValues(RetentionTaskName).Set(1)
+			monitoring.GlobalHubCronJobGaugeVec.WithLabelValues(RetentionTaskName).Set(1)
 		} else {
-			metrics.GlobalHubCronJobGaugeVec.WithLabelValues(RetentionTaskName).Set(0)
+			monitoring.GlobalHubCronJobGaugeVec.WithLabelValues(RetentionTaskName).Set(0)
 		}
 	}()
 

manager/pkg/cronjob/task/local_compliance_history.go

@@ -11,7 +11,7 @@ import (
 	"k8s.io/apimachinery/pkg/util/wait"
 	ctrl "sigs.k8s.io/controller-runtime"
 
-	"github.com/stolostron/multicluster-global-hub/manager/pkg/metrics"
+	"github.com/stolostron/multicluster-global-hub/manager/pkg/monitoring"
 	"github.com/stolostron/multicluster-global-hub/pkg/database"
 	"github.com/stolostron/multicluster-global-hub/pkg/database/models"
 )
@@ -35,10 +35,6 @@ var (
 	// sizes and measure the performance of the queries.
 )
 
-func init() {
-	metrics.GlobalHubCronJobGaugeVec.WithLabelValues(LocalComplianceTaskName).Set(0)
-}
-
 func SyncLocalCompliance(ctx context.Context, enableSimulation bool, job gocron.Job) {
 	startTime = time.Now()
 
@@ -57,9 +53,9 @@ func SyncLocalCompliance(ctx context.Context, enableSimulation bool, job gocron.
 	var err error
 	defer func() {
 		if err != nil {
-			metrics.GlobalHubCronJobGaugeVec.WithLabelValues(LocalComplianceTaskName).Set(1)
+			monitoring.GlobalHubCronJobGaugeVec.WithLabelValues(LocalComplianceTaskName).Set(1)
 		} else {
-			metrics.GlobalHubCronJobGaugeVec.WithLabelValues(LocalComplianceTaskName).Set(0)
+			monitoring.GlobalHubCronJobGaugeVec.WithLabelValues(LocalComplianceTaskName).Set(0)
 		}
 	}()
 

manager/pkg/monitoring/metrics.go

@@ -0,0 +1,21 @@
+package monitoring
+
+import (
+	"github.com/prometheus/client_golang/prometheus"
+	"sigs.k8s.io/controller-runtime/pkg/metrics"
+)
+
+var GlobalHubCronJobGaugeVec = prometheus.NewGaugeVec(
+	prometheus.GaugeOpts{
+		Name: "multicluster_global_hub_jobs_status",
+		Help: "The status of the job. 0 == success, 1 == failure.",
+	},
+	[]string{
+		"type", // The name of the cronjob.
+	},
+)
+
+// RegisterMetrics will register metrics with the global prometheus registry
+func RegisterMetrics() {
+	metrics.Registry.MustRegister(GlobalHubCronJobGaugeVec)
+}

operator/pkg/config/multiclusterglobalhub_config.go

@@ -76,9 +76,10 @@ var (
 		GrafanaImageKey:          "quay.io/stolostron/grafana:globalhub-1.0",
 		PostgresImageKey:         "quay.io/stolostron/postgresql-13:1-101",
 	}
-	statisticLogInterval = "1m"
-	imagePullSecretName  = ""
-	transporter          transport.Transporter
+	statisticLogInterval  = "1m"
+	metricsScrapeInterval = "12h"
+	imagePullSecretName   = ""
+	transporter           transport.Transporter
 )
 
 func SetMGHNamespacedName(namespacedName types.NamespacedName) {
@@ -223,6 +224,14 @@ func GetStatisticLogInterval() string {
 	return statisticLogInterval
 }
 
+func GetMetricsScrapeInterval(mgh *globalhubv1alpha4.MulticlusterGlobalHub) string {
+	interval := getAnnotation(mgh, operatorconstants.AnnotationMetricsScrapeInterval)
+	if interval == "" {
+		interval = metricsScrapeInterval
+	}
+	return interval
+}
+
 func GetPostgresStorageSize(mgh *globalhubv1alpha4.MulticlusterGlobalHub) string {
 	if mgh.Spec.DataLayer.Postgres.StorageSize != "" {
 		return mgh.Spec.DataLayer.Postgres.StorageSize

operator/pkg/constants/constants.go

@@ -56,6 +56,8 @@ const (
 	MGHOperandImagePrefix = "RELATED_IMAGE_"
 	// AnnotationStatisticInterval to log the interval of statistic log
 	AnnotationStatisticInterval = "mgh-statistic-interval"
+	// AnnotationMetricsScrapeInterval to set the scrape interval for metrics
+	AnnotationMetricsScrapeInterval = "mgh-metrics-scrape-interval"
 	// AnnotationONMulticlusterHub indicates the addons are running on a hub cluster
 	AnnotationONMulticlusterHub = "addon.open-cluster-management.io/on-multicluster-hub"
 	// AnnotationPolicyONMulticlusterHub indicates the policy spec sync is running on a hub cluster

operator/pkg/controllers/hubofhubs/globalhub_metrics.go

@@ -11,6 +11,7 @@ import (
 	"sigs.k8s.io/controller-runtime/pkg/client"
 
 	globalhubv1alpha4 "github.com/stolostron/multicluster-global-hub/operator/apis/v1alpha4"
+	"github.com/stolostron/multicluster-global-hub/operator/pkg/config"
 	operatorconstants "github.com/stolostron/multicluster-global-hub/operator/pkg/constants"
 	"github.com/stolostron/multicluster-global-hub/pkg/constants"
 	"github.com/stolostron/multicluster-global-hub/pkg/utils"
@@ -67,7 +68,7 @@ func (r *MulticlusterGlobalHubReconciler) reconcileMetrics(ctx context.Context,
 				{
 					Port:     "metrics",
 					Path:     "/metrics",
-					Interval: promv1.Duration("12h"),
+					Interval: promv1.Duration(config.GetMetricsScrapeInterval(mgh)),
 				},
 			},
 		},