b.ReportMetric(float64(latencies[len(latencies)*90/100]), "p90-ns")
			b.ReportMetric(float64(latencies[len(latencies)*99/100]), "p99-ns")
		}
		return func(b *testing.B) {
			b.Run("idle", runOne)

			b.Run("loaded", func(b *testing.B) {
				stop := applyGCLoad(b)
				runOne(b)
				// Make sure to stop the timer before we wait! The load created above

	b.ReportMetric(0, "ns/op")
	b.ReportMetric(0, "B/op")
	b.ReportMetric(0, "allocs/op")

	// Sort latencies then report percentiles.
	slices.Sort(latencies)
	b.ReportMetric(float64(latencies[len(latencies)*50/100]), "p50-ns")
	b.ReportMetric(float64(latencies[len(latencies)*90/100]), "p90-ns")
	b.ReportMetric(float64(latencies[len(latencies)*99/100]), "p99-ns")
}

func TestUserForcedGC(t *testing.T) {

	b.ReportMetric(0, "allocs/op")

	// Sort latencies then report percentiles.
	sort.Slice(latencies, func(i, j int) bool {
		return latencies[i] < latencies[j]
	})
	b.ReportMetric(float64(latencies[len(latencies)*50/100]), "p50-ns")
	b.ReportMetric(float64(latencies[len(latencies)*90/100]), "p90-ns")
	b.ReportMetric(float64(latencies[len(latencies)*99/100]), "p99-ns")
}

var readMetricsSink [1024]interface{}

query results are printed to the console, organized by workload name.

All metrics returned are from server-side reports. This means that latencies
and error rates are from the perspective of the service itself and not of an
individual client (or aggregate set of clients). Rates and latencies are
calculated over a time interval of 1 minute.
`,
		Example: `  # Retrieve workload metrics for productpage-v1 workload

		in a runnable state before actually running. Bucket counts
		increase monotonically.

	/sched/pauses/stopping/gc:seconds
		Distribution of individual GC-related stop-the-world stopping
		latencies. This is the time it takes from deciding to stop the
		world until all Ps are stopped. This is a subset of the total
		GC-related stop-the-world time (/sched/pauses/total/gc:seconds).

	m.Set(driveAvailabilityErrorsTotal, float64(disk.Metrics.TotalErrorsAvailability), labels...)
	m.Set(driveWaitingIO, float64(disk.Metrics.TotalWaiting), labels...)

	// Append the api label for the drive API latencies.
	labels = append(labels, "api", "")
	lastIdx := len(labels) - 1
	for apiName, latency := range disk.Metrics.LastMinute {
		labels[lastIdx] = "storage." + apiName

			Help: "Etcd request latency in seconds for each operation and object type.",
			// Etcd request latency in seconds for each operation and object type.
			// This metric is used for verifying etcd api call latencies SLO
			// keep consistent with apiserver metric 'requestLatencies' in
			// staging/src/k8s.io/apiserver/pkg/endpoints/metrics/metrics.go
			Buckets: []float64{0.005, 0.025, 0.05, 0.1, 0.2, 0.4, 0.6, 0.8, 1.0, 1.25, 1.5, 2, 3,

	// latencies, defined as the time taken by stopTheWorldWithSema to get
	// all Ps to stop. stwStoppingTimeGC covers all GC-related STWs,
	// stwStoppingTimeOther covers the others.
	stwStoppingTimeGC    timeHistogram
	stwStoppingTimeOther timeHistogram

	// stwTotalTimeGC/Other are distributions of stop-the-world total
	// latencies, defined as the total time from stopTheWorldWithSema to

	// configuration. This is an additional safety check inaddition to adding
	// cachesSynced logic to readiness probe to handle cases where kube-proxy
	// ip tables update latencies.
	// See https://github.com/istio/istio/issues/25495.
	if !s.IsServerReady() {
		return status.Error(codes.Unavailable, "server is not ready to serve discovery information")
	}

	ctx := stream.Context()

	// configuration. This is an additional safety check inaddition to adding
	// cachesSynced logic to readiness probe to handle cases where kube-proxy
	// ip tables update latencies.
	// See https://github.com/istio/istio/issues/25495.
	if !s.IsServerReady() {
		return errors.New("server is not ready to serve discovery information")
	}

	ctx := stream.Context()
	peerAddr := "0.0.0.0"