fixing nits and bug for watch metrics

Signed-off-by: Jaydip Gabani <[email protected]>

pkg/controller/constraint/stats_reporter.go

@@ -31,7 +31,7 @@ func init() {
 	}
 }
 
-func (r *reporter) observeConstraints(ctx context.Context, observer metric.Observer) error {
+func (r *reporter) observeConstraints(_ context.Context, observer metric.Observer) error {
 	r.mux.RLock()
 	defer r.mux.RUnlock()
 	for t, v := range r.constraintsReport {
@@ -40,7 +40,7 @@ func (r *reporter) observeConstraints(ctx context.Context, observer metric.Obser
 	return nil
 }
 
-func (r *reporter) reportConstraints(ctx context.Context, t tags, v int64) error {
+func (r *reporter) reportConstraints(_ context.Context, t tags, v int64) error {
 	r.mux.Lock()
 	defer r.mux.Unlock()
 	if r.constraintsReport == nil {

pkg/controller/expansion/stats_reporter.go

@@ -64,7 +64,7 @@ func (r *etRegistry) remove(key types.NamespacedName) {
 	r.dirty = true
 }
 
-func (r *etRegistry) report(ctx context.Context) {
+func (r *etRegistry) report(_ context.Context) {
 	if !r.dirty {
 		return
 	}

pkg/metrics/exporters/opentelemetry/opentelemetry.go

@@ -14,7 +14,6 @@ import (
 
 const (
 	Name                          = "opentelemetry"
-	metricPrefix                  = "gatekeeper"
 	defaultMetricsCollectInterval = 10 * time.Second
 	defaultMetricsTimeout         = 30 * time.Second
 )

pkg/metrics/exporters/prometheus/prometheus_exporter.go

@@ -68,7 +68,7 @@ func newPromSrv(port int) *http.Server {
 	server := &http.Server{
 		Addr:              fmt.Sprintf(":%v", port),
 		Handler:           sm,
-		ReadHeaderTimeout: 30 * time.Second,
+		ReadHeaderTimeout: readHeaderTimeout,
 	}
 	return server
 }

pkg/syncutil/stats_reporter.go

@@ -2,7 +2,6 @@ package syncutil
 
 import (
 	"context"
-	"errors"
 	"strings"
 	"sync"
 	"time"
@@ -27,25 +26,25 @@ const (
 )
 
 var (
-	syncM         metric.Int64ObservableGauge
 	syncDurationM metric.Float64Histogram
-	lastRunSyncM  metric.Float64ObservableGauge
 	meter         metric.Meter
+	r             *Reporter
 )
 
 func init() {
 	var err error
 	meter = otel.GetMeterProvider().Meter("gatekeeper")
+	r = &Reporter{now: now}
 
-	syncM, err = meter.Int64ObservableGauge(syncMetricName, metric.WithDescription("Total number of resources of each kind being cached"))
+	_, err = meter.Int64ObservableGauge(syncMetricName, metric.WithDescription("Total number of resources of each kind being cached"), metric.WithInt64Callback(r.observeSync))
 	if err != nil {
 		panic(err)
 	}
 	syncDurationM, err = meter.Float64Histogram(syncDurationMetricName, metric.WithDescription("Latency of sync operation in seconds"), metric.WithUnit("s"))
 	if err != nil {
 		panic(err)
 	}
-	lastRunSyncM, err = meter.Float64ObservableGauge(lastRunTimeMetricName, metric.WithDescription("Timestamp of last sync operation"), metric.WithUnit("s"))
+	_, err = meter.Float64ObservableGauge(lastRunTimeMetricName, metric.WithDescription("Timestamp of last sync operation"), metric.WithUnit("s"), metric.WithFloat64Callback(r.observeLastSync))
 	if err != nil {
 		panic(err)
 	}
@@ -180,14 +179,7 @@ type Reporter struct {
 
 // NewStatsReporter creates a reporter for sync metrics.
 func NewStatsReporter() (*Reporter, error) {
-	r := &Reporter{now: now}
-	return r, r.registerCallback()
-}
-
-func (r *Reporter) registerCallback() error {
-	_, err1 := meter.RegisterCallback(r.observeSync, syncM)
-	_, err2 := meter.RegisterCallback(r.observeLastSync, lastRunSyncM)
-	return errors.Join(err1, err2)
+	return r, nil
 }
 
 func (r *Reporter) ReportSyncDuration(d time.Duration) error {
@@ -213,18 +205,18 @@ func (r *Reporter) ReportSync(t Tags, v int64) error {
 	return nil
 }
 
-func (r *Reporter) observeLastSync(_ context.Context, observer metric.Observer) error {
+func (r *Reporter) observeLastSync(_ context.Context, observer metric.Float64Observer) error {
 	r.mu.RLock()
 	defer r.mu.RUnlock()
-	observer.ObserveFloat64(lastRunSyncM, r.lastRun)
+	observer.Observe(r.lastRun)
 	return nil
 }
 
-func (r *Reporter) observeSync(_ context.Context, observer metric.Observer) error {
+func (r *Reporter) observeSync(_ context.Context, observer metric.Int64Observer) error {
 	r.mu.RLock()
 	defer r.mu.RUnlock()
 	for t, v := range r.syncReport {
-		observer.ObserveInt64(syncM, v, metric.WithAttributes(attribute.String(kindKey, t.Kind), attribute.String(statusKey, string(t.Status))))
+		observer.Observe(v, metric.WithAttributes(attribute.String(kindKey, t.Kind), attribute.String(statusKey, string(t.Status))))
 	}
 	return nil
 }

pkg/syncutil/stats_reporter_test.go

@@ -9,6 +9,7 @@ import (
 	testmetric "github.com/open-policy-agent/gatekeeper/v3/test/metrics"
 	"github.com/stretchr/testify/assert"
 	"go.opentelemetry.io/otel/attribute"
+	"go.opentelemetry.io/otel/metric"
 	sdkmetric "go.opentelemetry.io/otel/sdk/metric"
 	"go.opentelemetry.io/otel/sdk/metric/metricdata"
 	"go.opentelemetry.io/otel/sdk/metric/metricdata/metricdatatest"
@@ -20,14 +21,13 @@ func initializeTestInstruments(t *testing.T) (rdr *sdkmetric.PeriodicReader, r *
 	mp := sdkmetric.NewMeterProvider(sdkmetric.WithReader(rdr))
 	meter = mp.Meter("test")
 
-	syncM, err = meter.Int64ObservableGauge(syncMetricName)
+	r, err = NewStatsReporter()
 	assert.NoError(t, err)
-	syncDurationM, err = meter.Float64Histogram(syncDurationMetricName)
+	_, err = meter.Int64ObservableGauge(syncMetricName, metric.WithInt64Callback(r.observeSync))
 	assert.NoError(t, err)
-	lastRunSyncM, err = meter.Float64ObservableGauge(lastRunTimeMetricName)
+	syncDurationM, err = meter.Float64Histogram(syncDurationMetricName)
 	assert.NoError(t, err)
-
-	r, err = NewStatsReporter()
+	_, err = meter.Float64ObservableGauge(lastRunTimeMetricName, metric.WithFloat64Callback(r.observeLastSync))
 	assert.NoError(t, err)
 
 	return rdr, r

pkg/watch/stats_reporter.go

@@ -2,7 +2,6 @@ package watch
 
 import (
 	"context"
-	"errors"
 	"sync"
 
 	"go.opentelemetry.io/otel"
@@ -15,25 +14,28 @@ const (
 )
 
 var (
-	meter           metric.Meter
-	gvkCountM       metric.Int64ObservableGauge
-	gvkIntentCountM metric.Int64ObservableGauge
+	meter metric.Meter
+	r     *reporter
 )
 
 func init() {
 	var err error
 	meterProvider := otel.GetMeterProvider()
 	meter = meterProvider.Meter("gatekeeper")
-	gvkCountM, err = meter.Int64ObservableGauge(
+	r = &reporter{}
+	_, err = meter.Int64ObservableGauge(
 		gvkCountMetricName,
 		metric.WithDescription("The total number of Group/Version/Kinds currently watched by the watch manager"),
+		metric.WithInt64Callback(r.observeGvkCount),
 	)
 	if err != nil {
 		panic(err)
 	}
-	gvkIntentCountM, err = meter.Int64ObservableGauge(
+	_, err = meter.Int64ObservableGauge(
 		gvkIntentCountMetricName,
-		metric.WithDescription("The total number of Group/Version/Kinds that the watch manager has instructions to watch. This could differ from the actual count due to resources being pending, non-existent, or a failure of the watch manager to restart"))
+		metric.WithDescription("The total number of Group/Version/Kinds that the watch manager has instructions to watch. This could differ from the actual count due to resources being pending, non-existent, or a failure of the watch manager to restart"),
+		metric.WithInt64Callback(r.observeGvkIntentCount),
+	)
 	if err != nil {
 		panic(err)
 	}
@@ -53,17 +55,9 @@ func (r *reporter) reportGvkIntentCount(count int64) error {
 	return nil
 }
 
-func (r *reporter) registerCallback() error {
-	_, err1 := meter.RegisterCallback(r.observeGvkIntentCount, gvkIntentCountM)
-	_, err2 := meter.RegisterCallback(r.observeGvkCount, gvkCountM)
-
-	return errors.Join(err1, err2)
-}
-
 // newStatsReporter creates a reporter for watch metrics.
 func newStatsReporter() (*reporter, error) {
-	r := &reporter{}
-	return r, r.registerCallback()
+	return r, nil
 }
 
 type reporter struct {
@@ -72,14 +66,17 @@ type reporter struct {
 	intentCount int64
 }
 
-func (r *reporter) observeGvkCount(ctx context.Context, observer metric.Observer) error {
-	log.Info("reporting gvk count")
-	observer.ObserveInt64(gvkCountM, r.gvkCount)
+func (r *reporter) observeGvkCount(_ context.Context, observer metric.Int64Observer) error {
+	r.mu.RLock()
+	defer r.mu.RUnlock()
+	observer.Observe(r.gvkCount)
 	return nil
 }
 
 // count returns total gvk count across all registrars.
-func (r *reporter) observeGvkIntentCount(_ context.Context, observer metric.Observer) error {
-	observer.ObserveInt64(gvkIntentCountM, r.intentCount)
+func (r *reporter) observeGvkIntentCount(_ context.Context, observer metric.Int64Observer) error {
+	r.mu.RLock()
+	defer r.mu.RUnlock()
+	observer.Observe(r.intentCount)
 	return nil
 }

pkg/watch/stats_reporter_test.go

@@ -6,6 +6,7 @@ import (
 
 	testmetric "github.com/open-policy-agent/gatekeeper/v3/test/metrics"
 	"github.com/stretchr/testify/assert"
+	"go.opentelemetry.io/otel/metric"
 	sdkmetric "go.opentelemetry.io/otel/sdk/metric"
 	"go.opentelemetry.io/otel/sdk/metric/metricdata"
 	"go.opentelemetry.io/otel/sdk/metric/metricdata/metricdatatest"
@@ -17,12 +18,12 @@ func initializeTestInstruments(t *testing.T) (rdr *sdkmetric.PeriodicReader, r *
 	mp := sdkmetric.NewMeterProvider(sdkmetric.WithReader(rdr))
 	meter = mp.Meter("test")
 
-	// Ensure the pipeline has a callback setup
-	gvkIntentCountM, err = meter.Int64ObservableGauge(gvkIntentCountMetricName)
+	r, err = newStatsReporter()
 	assert.NoError(t, err)
-	gvkCountM, err = meter.Int64ObservableGauge(gvkCountMetricName)
+	// Ensure the pipeline has a callback setup
+	_, err = meter.Int64ObservableGauge(gvkIntentCountMetricName, metric.WithInt64Callback(r.observeGvkIntentCount))
 	assert.NoError(t, err)
-	r, err = newStatsReporter()
+	_, err = meter.Int64ObservableGauge(gvkCountMetricName, metric.WithInt64Callback(r.observeGvkCount))
 	assert.NoError(t, err)
 	return rdr, r
 }