metric: migrate all histograms to use prometheus-backed version

In a previous change, a new prometheus-backed histogram library was
introdced to help standardize histogram buckets across the codebase.
This change migrates all existing histograms to use the new library.

related: cockroachdb#85990

Release justification: low risk, high benefit

Release note (ops change): This change introduces a new histogram
implementation that will reduce the total number of buckets and
standardize them across all usage. This should help increase the
usability of histograms when exxported to a UI (i.e. Grafana) and reduce
the storage overhead.

After applying this patch it is expected to see fewer buckets in
prometheus/grafana, but still  have similar values for histogram
percentiles due to the use of interpolated values by Prometheus.

pkg/ccl/changefeedccl/changefeed_processors.go

@@ -817,7 +817,7 @@ func (j *jobState) checkpointCompleted(ctx context.Context, checkpointDuration t
 
 	j.metrics.CheckpointHistNanos.RecordValue(checkpointDuration.Nanoseconds())
 	j.lastProgressUpdate = j.ts.Now()
-	j.checkpointDuration = time.Duration(j.metrics.CheckpointHistNanos.Snapshot().Mean())
+	j.checkpointDuration = time.Duration(j.metrics.CheckpointHistNanos.Mean())
 	j.progressUpdatesSkipped = false
 }
 

pkg/ccl/changefeedccl/metrics.go

@@ -450,20 +450,15 @@ func newAggregateMetrics(histogramWindow time.Duration) *AggMetrics {
 	a := &AggMetrics{
 		ErrorRetries:    b.Counter(metaChangefeedErrorRetries),
 		EmittedMessages: b.Counter(metaChangefeedEmittedMessages),
-		MessageSize: b.Histogram(metaMessageSize,
-			histogramWindow, 10<<20 /* 10MB max message size */, 1),
-		EmittedBytes: b.Counter(metaChangefeedEmittedBytes),
-		FlushedBytes: b.Counter(metaChangefeedFlushedBytes),
-		Flushes:      b.Counter(metaChangefeedFlushes),
-
-		BatchHistNanos: b.Histogram(metaChangefeedBatchHistNanos,
-			histogramWindow, changefeedBatchHistMaxLatency.Nanoseconds(), 1),
-		FlushHistNanos: b.Histogram(metaChangefeedFlushHistNanos,
-			histogramWindow, changefeedFlushHistMaxLatency.Nanoseconds(), 2),
-		CommitLatency: b.Histogram(metaCommitLatency,
-			histogramWindow, commitLatencyMaxValue.Nanoseconds(), 1),
-		AdmitLatency: b.Histogram(metaAdmitLatency, histogramWindow,
-			admitLatencyMaxValue.Nanoseconds(), 1),
+		MessageSize:     b.Histogram(metaMessageSize, histogramWindow, metric.DataSize16MBBuckets),
+		EmittedBytes:    b.Counter(metaChangefeedEmittedBytes),
+		FlushedBytes:    b.Counter(metaChangefeedFlushedBytes),
+		Flushes:         b.Counter(metaChangefeedFlushes),
+
+		BatchHistNanos:            b.Histogram(metaChangefeedBatchHistNanos, histogramWindow, metric.BatchProcessLatencyBuckets),
+		FlushHistNanos:            b.Histogram(metaChangefeedFlushHistNanos, histogramWindow, metric.BatchProcessLatencyBuckets),
+		CommitLatency:             b.Histogram(metaCommitLatency, histogramWindow, metric.BatchProcessLatencyBuckets),
+		AdmitLatency:              b.Histogram(metaAdmitLatency, histogramWindow, metric.BatchProcessLatencyBuckets),
 		BackfillCount:             b.Gauge(metaChangefeedBackfillCount),
 		BackfillPendingRanges:     b.Gauge(metaChangefeedBackfillPendingRanges),
 		RunningCount:              b.Gauge(metaChangefeedRunning),
@@ -566,17 +561,16 @@ func (m *Metrics) getSLIMetrics(scope string) (*sliMetrics, error) {
 // MakeMetrics makes the metrics for changefeed monitoring.
 func MakeMetrics(histogramWindow time.Duration) metric.Struct {
 	m := &Metrics{
-		AggMetrics:        newAggregateMetrics(histogramWindow),
-		KVFeedMetrics:     kvevent.MakeMetrics(histogramWindow),
-		SchemaFeedMetrics: schemafeed.MakeMetrics(histogramWindow),
-		ResolvedMessages:  metric.NewCounter(metaChangefeedForwardedResolvedMessages),
-		Failures:          metric.NewCounter(metaChangefeedFailures),
-		QueueTimeNanos:    metric.NewCounter(metaEventQueueTime),
-		CheckpointHistNanos: metric.NewHistogram(metaChangefeedCheckpointHistNanos, histogramWindow,
-			changefeedCheckpointHistMaxLatency.Nanoseconds(), 2),
-		FrontierUpdates: metric.NewCounter(metaChangefeedFrontierUpdates),
-		ThrottleMetrics: cdcutils.MakeMetrics(histogramWindow),
-		ReplanCount:     metric.NewCounter(metaChangefeedReplanCount),
+		AggMetrics:          newAggregateMetrics(histogramWindow),
+		KVFeedMetrics:       kvevent.MakeMetrics(histogramWindow),
+		SchemaFeedMetrics:   schemafeed.MakeMetrics(histogramWindow),
+		ResolvedMessages:    metric.NewCounter(metaChangefeedForwardedResolvedMessages),
+		Failures:            metric.NewCounter(metaChangefeedFailures),
+		QueueTimeNanos:      metric.NewCounter(metaEventQueueTime),
+		CheckpointHistNanos: metric.NewHistogram(metaChangefeedCheckpointHistNanos, histogramWindow, metric.IOLatencyBuckets),
+		FrontierUpdates:     metric.NewCounter(metaChangefeedFrontierUpdates),
+		ThrottleMetrics:     cdcutils.MakeMetrics(histogramWindow),
+		ReplanCount:         metric.NewCounter(metaChangefeedReplanCount),
 	}
 
 	m.mu.resolved = make(map[int]hlc.Timestamp)

pkg/ccl/sqlproxyccl/connector_test.go

@@ -375,7 +375,9 @@ func TestConnector_dialTenantCluster(t *testing.T) {
 		defer cancel()
 
 		c := &connector{
-			DialTenantLatency: metric.NewLatency(metaDialTenantLatency, time.Millisecond),
+			DialTenantLatency: metric.NewHistogram(
+				metaDialTenantLatency, time.Millisecond, metric.IOLatencyBuckets,
+			),
 			DialTenantRetries: metric.NewCounter(metaDialTenantRetries),
 		}
 		c.testingKnobs.lookupAddr = func(ctx context.Context) (string, error) {
@@ -403,8 +405,10 @@ func TestConnector_dialTenantCluster(t *testing.T) {
 
 		var reportFailureFnCount int
 		c := &connector{
-			TenantID:          roachpb.MakeTenantID(42),
-			DialTenantLatency: metric.NewLatency(metaDialTenantLatency, time.Millisecond),
+			TenantID: roachpb.MakeTenantID(42),
+			DialTenantLatency: metric.NewHistogram(
+				metaDialTenantLatency, time.Millisecond, metric.IOLatencyBuckets,
+			),
 			DialTenantRetries: metric.NewCounter(metaDialTenantRetries),
 		}
 		c.DirectoryCache = &testTenantDirectoryCache{

pkg/ccl/sqlproxyccl/metrics.go

@@ -224,32 +224,34 @@ func makeProxyMetrics() metrics {
 		RoutingErrCount:        metric.NewCounter(metaRoutingErrCount),
 		RefusedConnCount:       metric.NewCounter(metaRefusedConnCount),
 		SuccessfulConnCount:    metric.NewCounter(metaSuccessfulConnCount),
-		ConnectionLatency: metric.NewLatency(
+		ConnectionLatency: metric.NewHistogram(
 			metaConnMigrationAttemptedCount,
 			base.DefaultHistogramWindowInterval(),
+			metric.IOLatencyBuckets,
 		),
 		AuthFailedCount:        metric.NewCounter(metaAuthFailedCount),
 		ExpiredClientConnCount: metric.NewCounter(metaExpiredClientConnCount),
 		// Connector metrics.
-		DialTenantLatency: metric.NewLatency(
+		DialTenantLatency: metric.NewHistogram(
 			metaDialTenantLatency,
 			base.DefaultHistogramWindowInterval(),
+			metric.NetworkLatencyBuckets,
 		),
 		DialTenantRetries: metric.NewCounter(metaDialTenantRetries),
 		// Connection migration metrics.
 		ConnMigrationSuccessCount:          metric.NewCounter(metaConnMigrationSuccessCount),
 		ConnMigrationErrorFatalCount:       metric.NewCounter(metaConnMigrationErrorFatalCount),
 		ConnMigrationErrorRecoverableCount: metric.NewCounter(metaConnMigrationErrorRecoverableCount),
 		ConnMigrationAttemptedCount:        metric.NewCounter(metaConnMigrationAttemptedCount),
-		ConnMigrationAttemptedLatency: metric.NewLatency(
+		ConnMigrationAttemptedLatency: metric.NewHistogram(
 			metaConnMigrationAttemptedLatency,
 			base.DefaultHistogramWindowInterval(),
+			metric.NetworkLatencyBuckets,
 		),
 		ConnMigrationTransferResponseMessageSize: metric.NewHistogram(
 			metaConnMigrationTransferResponseMessageSize,
 			base.DefaultHistogramWindowInterval(),
-			maxExpectedTransferResponseMessageSize,
-			1,
+			metric.DataSize16MBBuckets,
 		),
 		QueryCancelReceivedPGWire: metric.NewCounter(metaQueryCancelReceivedPGWire),
 		QueryCancelReceivedHTTP:   metric.NewCounter(metaQueryCancelReceivedHTTP),

pkg/ccl/streamingccl/streamingest/metrics.go

@@ -135,11 +135,11 @@ func MakeMetrics(histogramWindow time.Duration) metric.Struct {
 		ResolvedEvents:     metric.NewCounter(metaStreamingResolvedEventsIngested),
 		JobProgressUpdates: metric.NewCounter(metaJobProgressUpdates),
 		FlushHistNanos: metric.NewHistogram(metaStreamingFlushHistNanos,
-			histogramWindow, streamingFlushHistMaxLatency.Nanoseconds(), 1),
+			histogramWindow, metric.BatchProcessLatencyBuckets),
 		CommitLatency: metric.NewHistogram(metaStreamingCommitLatency,
-			histogramWindow, streamingCommitLatencyMaxValue.Nanoseconds(), 1),
+			histogramWindow, metric.BatchProcessLatencyBuckets),
 		AdmitLatency: metric.NewHistogram(metaStreamingAdmitLatency,
-			histogramWindow, streamingAdmitLatencyMaxValue.Nanoseconds(), 1),
+			histogramWindow, metric.BatchProcessLatencyBuckets),
 		RunningCount:                metric.NewGauge(metaStreamsRunning),
 		EarliestDataCheckpointSpan:  metric.NewGauge(metaEarliestDataCheckpointSpan),
 		LatestDataCheckpointSpan:    metric.NewGauge(metaLatestDataCheckpointSpan),

pkg/kv/bulk/bulk_metrics.go

@@ -43,14 +43,10 @@ var (
 	}
 )
 
-// See pkg/sql/mem_metrics.go
-// log10int64times1000 = log10(math.MaxInt64) * 1000, rounded up somewhat
-const log10int64times1000 = 19 * 1000
-
 // MakeBulkMetrics instantiates the metrics holder for bulk operation monitoring.
 func MakeBulkMetrics(histogramWindow time.Duration) Metrics {
 	return Metrics{
-		MaxBytesHist:  metric.NewHistogram(metaMemMaxBytes, histogramWindow, log10int64times1000, 3),
+		MaxBytesHist:  metric.NewHistogram(metaMemMaxBytes, histogramWindow, metric.MemoryUsage16MBBuckets),
 		CurBytesCount: metric.NewGauge(metaMemCurBytes),
 	}
 }

pkg/kv/kvclient/kvcoord/txn_coord_sender_test.go

@@ -1073,15 +1073,12 @@ func checkTxnMetrics(
 	commits, commits1PC, aborts, restarts int64,
 ) {
 	testutils.SucceedsSoon(t, func() error {
-		return checkTxnMetricsOnce(t, metrics, name, commits, commits1PC, aborts, restarts)
+		return checkTxnMetricsOnce(metrics, name, commits, commits1PC, aborts, restarts)
 	})
 }
 
 func checkTxnMetricsOnce(
-	t *testing.T,
-	metrics kvcoord.TxnMetrics,
-	name string,
-	commits, commits1PC, aborts, restarts int64,
+	metrics kvcoord.TxnMetrics, name string, commits, commits1PC, aborts, restarts int64,
 ) error {
 	testcases := []struct {
 		name string
@@ -1091,6 +1088,7 @@ func checkTxnMetricsOnce(
 		{"commits1PC", metrics.Commits1PC.Count(), commits1PC},
 		{"aborts", metrics.Aborts.Count(), aborts},
 		{"durations", metrics.Durations.TotalCount(), commits + aborts},
+		{"restarts", metrics.Restarts.TotalCount(), restarts},
 	}
 
 	for _, tc := range testcases {
@@ -1099,22 +1097,6 @@ func checkTxnMetricsOnce(
 		}
 	}
 
-	// Handle restarts separately, because that's a histogram. Though the
-	// histogram is approximate, we're recording so few distinct values
-	// that we should be okay.
-	dist := metrics.Restarts.Snapshot().Distribution()
-	var actualRestarts int64
-	for _, b := range dist {
-		if b.From == b.To {
-			actualRestarts += b.From * b.Count
-		} else {
-			t.Fatalf("unexpected value in histogram: %d-%d", b.From, b.To)
-		}
-	}
-	if a, e := actualRestarts, restarts; a != e {
-		return errors.Errorf("%s: actual restarts %d != expected %d", name, a, e)
-	}
-
 	return nil
 }
 
@@ -1323,11 +1305,6 @@ func TestTxnDurations(t *testing.T) {
 	if a, e := hist.TotalCount(), int64(puts); a != e {
 		t.Fatalf("durations %d != expected %d", a, e)
 	}
-
-	// Metrics lose fidelity, so we can't compare incr directly.
-	if min, thresh := hist.Min(), (incr - 10).Nanoseconds(); min < thresh {
-		t.Fatalf("min %d < %d", min, thresh)
-	}
 }
 
 // TestTxnCommitWait tests the commit-wait sleep phase of transactions under

pkg/kv/kvclient/kvcoord/txn_metrics.go

@@ -274,11 +274,11 @@ func MakeTxnMetrics(histogramWindow time.Duration) TxnMetrics {
 		RefreshFailWithCondensedSpans:  metric.NewCounter(metaRefreshFailWithCondensedSpans),
 		RefreshMemoryLimitExceeded:     metric.NewCounter(metaRefreshMemoryLimitExceeded),
 		RefreshAutoRetries:             metric.NewCounter(metaRefreshAutoRetries),
-		Durations:                      metric.NewLatency(metaDurationsHistograms, histogramWindow),
+		Durations:                      metric.NewHistogram(metaDurationsHistograms, histogramWindow, metric.IOLatencyBuckets),
 		TxnsWithCondensedIntents:       metric.NewCounter(metaTxnsWithCondensedIntentSpans),
 		TxnsWithCondensedIntentsGauge:  metric.NewGauge(metaTxnsWithCondensedIntentSpansGauge),
 		TxnsRejectedByLockSpanBudget:   metric.NewCounter(metaTxnsRejectedByLockSpanBudget),
-		Restarts:                       metric.NewHistogram(metaRestartsHistogram, histogramWindow, 100, 3),
+		Restarts:                       metric.NewHistogram(metaRestartsHistogram, histogramWindow, metric.Count1KBuckets),
 		RestartsWriteTooOld:            telemetry.NewCounterWithMetric(metaRestartsWriteTooOld),
 		RestartsWriteTooOldMulti:       telemetry.NewCounterWithMetric(metaRestartsWriteTooOldMulti),
 		RestartsSerializable:           telemetry.NewCounterWithMetric(metaRestartsSerializable),

pkg/kv/kvprober/kvprober.go

@@ -214,14 +214,18 @@ func NewProber(opts Opts) *Prober {
 		writePlanner: newMeta2Planner(opts.DB, opts.Settings, func() time.Duration { return writeInterval.Get(&opts.Settings.SV) }),
 
 		metrics: Metrics{
-			ReadProbeAttempts:  metric.NewCounter(metaReadProbeAttempts),
-			ReadProbeFailures:  metric.NewCounter(metaReadProbeFailures),
-			ReadProbeLatency:   metric.NewLatency(metaReadProbeLatency, opts.HistogramWindowInterval),
+			ReadProbeAttempts: metric.NewCounter(metaReadProbeAttempts),
+			ReadProbeFailures: metric.NewCounter(metaReadProbeFailures),
+			ReadProbeLatency: metric.NewHistogram(
+				metaReadProbeLatency, opts.HistogramWindowInterval, metric.IOLatencyBuckets,
+			),
 			WriteProbeAttempts: metric.NewCounter(metaWriteProbeAttempts),
 			WriteProbeFailures: metric.NewCounter(metaWriteProbeFailures),
-			WriteProbeLatency:  metric.NewLatency(metaWriteProbeLatency, opts.HistogramWindowInterval),
-			ProbePlanAttempts:  metric.NewCounter(metaProbePlanAttempts),
-			ProbePlanFailures:  metric.NewCounter(metaProbePlanFailures),
+			WriteProbeLatency: metric.NewHistogram(
+				metaWriteProbeLatency, opts.HistogramWindowInterval, metric.IOLatencyBuckets,
+			),
+			ProbePlanAttempts: metric.NewCounter(metaProbePlanAttempts),
+			ProbePlanFailures: metric.NewCounter(metaProbePlanFailures),
 		},
 		tracer: opts.Tracer,
 	}

pkg/kv/kvserver/liveness/liveness.go

@@ -309,7 +309,9 @@ func NewNodeLiveness(opts NodeLivenessOptions) *NodeLiveness {
 		HeartbeatSuccesses: metric.NewCounter(metaHeartbeatSuccesses),
 		HeartbeatFailures:  telemetry.NewCounterWithMetric(metaHeartbeatFailures),
 		EpochIncrements:    telemetry.NewCounterWithMetric(metaEpochIncrements),
-		HeartbeatLatency:   metric.NewLatency(metaHeartbeatLatency, opts.HistogramWindowInterval),
+		HeartbeatLatency: metric.NewHistogram(
+			metaHeartbeatLatency, opts.HistogramWindowInterval, metric.IOLatencyBuckets,
+		),
 	}
 	nl.mu.nodes = make(map[roachpb.NodeID]Record)
 	nl.heartbeatToken <- struct{}{}

pkg/kv/kvserver/metrics.go

@@ -2245,18 +2245,27 @@ func newStoreMetrics(histogramWindow time.Duration) *StoreMetrics {
 		// Raft processing metrics.
 		RaftTicks: metric.NewCounter(metaRaftTicks),
 		RaftQuotaPoolPercentUsed: metric.NewHistogram(
-			// NB: this results in 64 buckets (i.e. 64 timeseries in prometheus).
-			metaRaftQuotaPoolPercentUsed, histogramWindow, 100 /* maxVal */, 1, /* sigFigs */
+			metaRaftQuotaPoolPercentUsed, histogramWindow, metric.Percent100Buckets,
 		),
-		RaftWorkingDurationNanos:  metric.NewCounter(metaRaftWorkingDurationNanos),
-		RaftTickingDurationNanos:  metric.NewCounter(metaRaftTickingDurationNanos),
-		RaftCommandsApplied:       metric.NewCounter(metaRaftCommandsApplied),
-		RaftLogCommitLatency:      metric.NewLatency(metaRaftLogCommitLatency, histogramWindow),
-		RaftCommandCommitLatency:  metric.NewLatency(metaRaftCommandCommitLatency, histogramWindow),
-		RaftHandleReadyLatency:    metric.NewLatency(metaRaftHandleReadyLatency, histogramWindow),
-		RaftApplyCommittedLatency: metric.NewLatency(metaRaftApplyCommittedLatency, histogramWindow),
-		RaftSchedulerLatency:      metric.NewLatency(metaRaftSchedulerLatency, histogramWindow),
-		RaftTimeoutCampaign:       metric.NewCounter(metaRaftTimeoutCampaign),
+		RaftWorkingDurationNanos: metric.NewCounter(metaRaftWorkingDurationNanos),
+		RaftTickingDurationNanos: metric.NewCounter(metaRaftTickingDurationNanos),
+		RaftCommandsApplied:      metric.NewCounter(metaRaftCommandsApplied),
+		RaftLogCommitLatency: metric.NewHistogram(
+			metaRaftLogCommitLatency, histogramWindow, metric.IOLatencyBuckets,
+		),
+		RaftCommandCommitLatency: metric.NewHistogram(
+			metaRaftCommandCommitLatency, histogramWindow, metric.IOLatencyBuckets,
+		),
+		RaftHandleReadyLatency: metric.NewHistogram(
+			metaRaftHandleReadyLatency, histogramWindow, metric.IOLatencyBuckets,
+		),
+		RaftApplyCommittedLatency: metric.NewHistogram(
+			metaRaftApplyCommittedLatency, histogramWindow, metric.IOLatencyBuckets,
+		),
+		RaftSchedulerLatency: metric.NewHistogram(
+			metaRaftSchedulerLatency, histogramWindow, metric.IOLatencyBuckets,
+		),
+		RaftTimeoutCampaign: metric.NewCounter(metaRaftTimeoutCampaign),
 
 		// Raft message metrics.
 		RaftRcvdMessages: [...]*metric.Counter{
@@ -2394,8 +2403,12 @@ func newStoreMetrics(histogramWindow time.Duration) *StoreMetrics {
 		ReplicaCircuitBreakerCumTripped: metric.NewCounter(metaReplicaCircuitBreakerCumTripped),
 
 		// Replica batch evaluation.
-		ReplicaReadBatchEvaluationLatency:  metric.NewLatency(metaReplicaReadBatchEvaluationLatency, histogramWindow),
-		ReplicaWriteBatchEvaluationLatency: metric.NewLatency(metaReplicaWriteBatchEvaluationLatency, histogramWindow),
+		ReplicaReadBatchEvaluationLatency: metric.NewHistogram(
+			metaReplicaReadBatchEvaluationLatency, histogramWindow, metric.IOLatencyBuckets,
+		),
+		ReplicaWriteBatchEvaluationLatency: metric.NewHistogram(
+			metaReplicaWriteBatchEvaluationLatency, histogramWindow, metric.IOLatencyBuckets,
+		),
 	}
 
 	{

pkg/kv/kvserver/txnwait/metrics.go

@@ -74,8 +74,7 @@ func NewMetrics(histogramWindowInterval time.Duration) *Metrics {
 				Unit:        metric.Unit_NANOSECONDS,
 			},
 			histogramWindowInterval,
-			time.Hour.Nanoseconds(),
-			1,
+			metric.LongRunning60mLatencyBuckets,
 		),
 
 		QueryWaitTime: metric.NewHistogram(
@@ -86,8 +85,7 @@ func NewMetrics(histogramWindowInterval time.Duration) *Metrics {
 				Unit:        metric.Unit_NANOSECONDS,
 			},
 			histogramWindowInterval,
-			time.Hour.Nanoseconds(),
-			1,
+			metric.LongRunning60mLatencyBuckets,
 		),
 
 		DeadlocksTotal: metric.NewCounter(

pkg/rpc/clock_offset.go

@@ -122,7 +122,9 @@ func newRemoteClockMonitor(
 	r.metrics = RemoteClockMetrics{
 		ClockOffsetMeanNanos:   metric.NewGauge(metaClockOffsetMeanNanos),
 		ClockOffsetStdDevNanos: metric.NewGauge(metaClockOffsetStdDevNanos),
-		LatencyHistogramNanos:  metric.NewLatency(metaLatencyHistogramNanos, histogramWindowInterval),
+		LatencyHistogramNanos: metric.NewHistogram(
+			metaLatencyHistogramNanos, histogramWindowInterval, metric.IOLatencyBuckets,
+		),
 	}
 	return &r
 }

pkg/server/node.go

@@ -173,7 +173,9 @@ type nodeMetrics struct {
 
 func makeNodeMetrics(reg *metric.Registry, histogramWindow time.Duration) nodeMetrics {
 	nm := nodeMetrics{
-		Latency:    metric.NewLatency(metaExecLatency, histogramWindow),
+		Latency: metric.NewHistogram(
+			metaExecLatency, histogramWindow, metric.IOLatencyBuckets,
+		),
 		Success:    metric.NewCounter(metaExecSuccess),
 		Err:        metric.NewCounter(metaExecError),
 		DiskStalls: metric.NewCounter(metaDiskStalls),

pkg/server/status/BUILD.bazel

@@ -147,7 +147,6 @@ go_test(
         "//pkg/util/system",
         "//pkg/util/timeutil",
         "@com_github_kr_pretty//:pretty",
-        "@com_github_prometheus_client_golang//prometheus",
         "@com_github_shirou_gopsutil_v3//net",
     ],
 )