go-ethereum/metrics/runtimehistogram_test.go
Felix Lange c539bda166
metrics: improve reading Go runtime metrics (#25886)
This changes how we read performance metrics from the Go runtime. Instead
of using runtime.ReadMemStats, we now rely on the API provided by package
runtime/metrics.

runtime/metrics provides more accurate information. For example, the new
interface has better reporting of memory use. In my testing, the reported
value of held memory more accurately reflects the usage reported by the OS.

The semantics of metrics system/memory/allocs and system/memory/frees have
changed to report amounts in bytes. ReadMemStats only reported the count of
allocations in number-of-objects. This is imprecise: 'tiny objects' are not
counted because the runtime allocates them in batches; and certain
improvements in allocation behavior, such as struct size optimizations,
will be less visible when the number of allocs doesn't change.

Changing allocation reports to be in bytes makes it appear in graphs that
lots more is being allocated. I don't think that's a problem because this
metric is primarily interesting for geth developers.

The metric system/memory/pauses has been changed to report statistical
values from the histogram provided by the runtime. Its name in influxdb has
changed from geth.system/memory/pauses.meter to
geth.system/memory/pauses.histogram.

We also have a new histogram metric, system/cpu/schedlatency, reporting the
Go scheduler latency.
2022-11-11 13:16:13 +01:00

134 lines
3.3 KiB
Go

package metrics
import (
"fmt"
"math"
"reflect"
"runtime/metrics"
"testing"
)
var _ Histogram = (*runtimeHistogram)(nil)
type runtimeHistogramTest struct {
h metrics.Float64Histogram
Count int64
Min int64
Max int64
Sum int64
Mean float64
Variance float64
StdDev float64
Percentiles []float64 // .5 .8 .9 .99 .995
}
// This test checks the results of statistical functions implemented
// by runtimeHistogramSnapshot.
func TestRuntimeHistogramStats(t *testing.T) {
tests := []runtimeHistogramTest{
0: {
h: metrics.Float64Histogram{
Counts: []uint64{},
Buckets: []float64{},
},
Count: 0,
Max: 0,
Min: 0,
Sum: 0,
Mean: 0,
Variance: 0,
StdDev: 0,
Percentiles: []float64{0, 0, 0, 0, 0},
},
1: {
// This checks the case where the highest bucket is +Inf.
h: metrics.Float64Histogram{
Counts: []uint64{0, 1, 2},
Buckets: []float64{0, 0.5, 1, math.Inf(1)},
},
Count: 3,
Max: 1,
Min: 0,
Sum: 3,
Mean: 0.9166666,
Percentiles: []float64{1, 1, 1, 1, 1},
Variance: 0.020833,
StdDev: 0.144433,
},
2: {
h: metrics.Float64Histogram{
Counts: []uint64{8, 6, 3, 1},
Buckets: []float64{12, 16, 18, 24, 25},
},
Count: 18,
Max: 25,
Min: 12,
Sum: 270,
Mean: 16.75,
Variance: 10.3015,
StdDev: 3.2096,
Percentiles: []float64{16, 18, 18, 24, 24},
},
}
for i, test := range tests {
t.Run(fmt.Sprint(i), func(t *testing.T) {
s := runtimeHistogramSnapshot(test.h)
if v := s.Count(); v != test.Count {
t.Errorf("Count() = %v, want %v", v, test.Count)
}
if v := s.Min(); v != test.Min {
t.Errorf("Min() = %v, want %v", v, test.Min)
}
if v := s.Max(); v != test.Max {
t.Errorf("Max() = %v, want %v", v, test.Max)
}
if v := s.Sum(); v != test.Sum {
t.Errorf("Sum() = %v, want %v", v, test.Sum)
}
if v := s.Mean(); !approxEqual(v, test.Mean, 0.0001) {
t.Errorf("Mean() = %v, want %v", v, test.Mean)
}
if v := s.Variance(); !approxEqual(v, test.Variance, 0.0001) {
t.Errorf("Variance() = %v, want %v", v, test.Variance)
}
if v := s.StdDev(); !approxEqual(v, test.StdDev, 0.0001) {
t.Errorf("StdDev() = %v, want %v", v, test.StdDev)
}
ps := []float64{.5, .8, .9, .99, .995}
if v := s.Percentiles(ps); !reflect.DeepEqual(v, test.Percentiles) {
t.Errorf("Percentiles(%v) = %v, want %v", ps, v, test.Percentiles)
}
})
}
}
func approxEqual(x, y, ε float64) bool {
if math.IsInf(x, -1) && math.IsInf(y, -1) {
return true
}
if math.IsInf(x, 1) && math.IsInf(y, 1) {
return true
}
if math.IsNaN(x) && math.IsNaN(y) {
return true
}
return math.Abs(x-y) < ε
}
// This test verifies that requesting Percentiles in unsorted order
// returns them in the requested order.
func TestRuntimeHistogramStatsPercentileOrder(t *testing.T) {
p := runtimeHistogramSnapshot{
Counts: []uint64{1, 1, 1, 1, 1, 1, 1, 1, 1, 1},
Buckets: []float64{0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10},
}
result := p.Percentiles([]float64{1, 0.2, 0.5, 0.1, 0.2})
expected := []float64{10, 2, 5, 1, 2}
if !reflect.DeepEqual(result, expected) {
t.Fatal("wrong result:", result)
}
}