go-pulse/metrics/sample.go
Martin Holst Swende 8b6cf128af
metrics: refactor metrics (#28035)
This change includes a lot of things, listed below. 

### Split up interfaces, write vs read

The interfaces have been split up into one write-interface and one read-interface, with `Snapshot` being the gateway from write to read. This simplifies the semantics _a lot_. 

Example of splitting up an interface into one readonly 'snapshot' part, and one updatable writeonly part: 

```golang
type MeterSnapshot interface {
	Count() int64
	Rate1() float64
	Rate5() float64
	Rate15() float64
	RateMean() float64
}

// Meters count events to produce exponentially-weighted moving average rates
// at one-, five-, and fifteen-minutes and a mean rate.
type Meter interface {
	Mark(int64)
	Snapshot() MeterSnapshot
	Stop()
}
```

### A note about concurrency

This PR makes the concurrency model clearer. We have actual meters and snapshot of meters. The `meter` is the thing which can be accessed from the registry, and updates can be made to it. 

- For all `meters`, (`Gauge`, `Timer` etc), it is assumed that they are accessed by different threads, making updates. Therefore, all `meters` update-methods (`Inc`, `Add`, `Update`, `Clear` etc) need to be concurrency-safe. 
- All `meters` have a `Snapshot()` method. This method is _usually_ called from one thread, a backend-exporter. But it's fully possible to have several exporters simultaneously: therefore this method should also be concurrency-safe. 

TLDR: `meter`s are accessible via registry, all their methods must be concurrency-safe. 

For all `Snapshot`s, it is assumed that an individual exporter-thread has obtained a `meter` from the registry, and called the `Snapshot` method to obtain a readonly snapshot. This snapshot is _not_ guaranteed to be concurrency-safe. There's no need for a snapshot to be concurrency-safe, since exporters should not share snapshots. 

Note, though: that by happenstance a lot of the snapshots _are_ concurrency-safe, being unmutable minimal representations of a value. Only the more complex ones are _not_ threadsafe, those that lazily calculate things like `Variance()`, `Mean()`.

Example of how a background exporter typically works, obtaining the snapshot and sequentially accessing the non-threadsafe methods in it: 
```golang
		ms := metric.Snapshot()
                ...
		fields := map[string]interface{}{
			"count":    ms.Count(),
			"max":      ms.Max(),
			"mean":     ms.Mean(),
			"min":      ms.Min(),
			"stddev":   ms.StdDev(),
			"variance": ms.Variance(),
```

TLDR: `snapshots` are not guaranteed to be concurrency-safe (but often are).

### Sample changes

I also changed the `Sample` type: previously, it iterated the samples fully every time `Mean()`,`Sum()`, `Min()` or `Max()` was invoked. Since we now have readonly base data, we can just iterate it once, in the constructor, and set all four values at once. 

The same thing has been done for runtimehistogram. 

### ResettingTimer API

Back when ResettingTImer was implemented, as part of https://github.com/ethereum/go-ethereum/pull/15910, Anton implemented a `Percentiles` on the new type. However, the method did not conform to the other existing types which also had a `Percentiles`. 

1. The existing ones, on input, took `0.5` to mean `50%`. Anton used `50` to mean `50%`. 
2. The existing ones returned `float64` outputs, thus interpolating between values. A value-set of `0, 10`, at `50%` would return `5`, whereas Anton's would return either `0` or `10`. 

This PR removes the 'new' version, and uses only the 'legacy' percentiles, also for the ResettingTimer type. 

The resetting timer snapshot was also defined so that it would expose the internal values. This has been removed, and getters for `Max, Min, Mean` have been added instead. 

### Unexport types

A lot of types were exported, but do not need to be. This PR unexports quite a lot of them.
2023-09-13 13:13:47 -04:00

447 lines
10 KiB
Go

package metrics
import (
"math"
"math/rand"
"sync"
"time"
"golang.org/x/exp/slices"
)
const rescaleThreshold = time.Hour
type SampleSnapshot interface {
Count() int64
Max() int64
Mean() float64
Min() int64
Percentile(float64) float64
Percentiles([]float64) []float64
Size() int
StdDev() float64
Sum() int64
Variance() float64
}
// Samples maintain a statistically-significant selection of values from
// a stream.
type Sample interface {
Snapshot() SampleSnapshot
Clear()
Update(int64)
}
// ExpDecaySample is an exponentially-decaying sample using a forward-decaying
// priority reservoir. See Cormode et al's "Forward Decay: A Practical Time
// Decay Model for Streaming Systems".
//
// <http://dimacs.rutgers.edu/~graham/pubs/papers/fwddecay.pdf>
type ExpDecaySample struct {
alpha float64
count int64
mutex sync.Mutex
reservoirSize int
t0, t1 time.Time
values *expDecaySampleHeap
rand *rand.Rand
}
// NewExpDecaySample constructs a new exponentially-decaying sample with the
// given reservoir size and alpha.
func NewExpDecaySample(reservoirSize int, alpha float64) Sample {
if !Enabled {
return NilSample{}
}
s := &ExpDecaySample{
alpha: alpha,
reservoirSize: reservoirSize,
t0: time.Now(),
values: newExpDecaySampleHeap(reservoirSize),
}
s.t1 = s.t0.Add(rescaleThreshold)
return s
}
// SetRand sets the random source (useful in tests)
func (s *ExpDecaySample) SetRand(prng *rand.Rand) Sample {
s.rand = prng
return s
}
// Clear clears all samples.
func (s *ExpDecaySample) Clear() {
s.mutex.Lock()
defer s.mutex.Unlock()
s.count = 0
s.t0 = time.Now()
s.t1 = s.t0.Add(rescaleThreshold)
s.values.Clear()
}
// Snapshot returns a read-only copy of the sample.
func (s *ExpDecaySample) Snapshot() SampleSnapshot {
s.mutex.Lock()
defer s.mutex.Unlock()
var (
samples = s.values.Values()
values = make([]int64, len(samples))
max int64 = math.MinInt64
min int64 = math.MaxInt64
sum int64
)
for i, item := range samples {
v := item.v
values[i] = v
sum += v
if v > max {
max = v
}
if v < min {
min = v
}
}
return newSampleSnapshotPrecalculated(s.count, values, min, max, sum)
}
// Update samples a new value.
func (s *ExpDecaySample) Update(v int64) {
s.update(time.Now(), v)
}
// update samples a new value at a particular timestamp. This is a method all
// its own to facilitate testing.
func (s *ExpDecaySample) update(t time.Time, v int64) {
s.mutex.Lock()
defer s.mutex.Unlock()
s.count++
if s.values.Size() == s.reservoirSize {
s.values.Pop()
}
var f64 float64
if s.rand != nil {
f64 = s.rand.Float64()
} else {
f64 = rand.Float64()
}
s.values.Push(expDecaySample{
k: math.Exp(t.Sub(s.t0).Seconds()*s.alpha) / f64,
v: v,
})
if t.After(s.t1) {
values := s.values.Values()
t0 := s.t0
s.values.Clear()
s.t0 = t
s.t1 = s.t0.Add(rescaleThreshold)
for _, v := range values {
v.k = v.k * math.Exp(-s.alpha*s.t0.Sub(t0).Seconds())
s.values.Push(v)
}
}
}
// NilSample is a no-op Sample.
type NilSample struct{}
func (NilSample) Clear() {}
func (NilSample) Snapshot() SampleSnapshot { return (*emptySnapshot)(nil) }
func (NilSample) Update(v int64) {}
// SamplePercentiles returns an arbitrary percentile of the slice of int64.
func SamplePercentile(values []int64, p float64) float64 {
return CalculatePercentiles(values, []float64{p})[0]
}
// CalculatePercentiles returns a slice of arbitrary percentiles of the slice of
// int64. This method returns interpolated results, so e.g if there are only two
// values, [0, 10], a 50% percentile will land between them.
//
// Note: As a side-effect, this method will also sort the slice of values.
// Note2: The input format for percentiles is NOT percent! To express 50%, use 0.5, not 50.
func CalculatePercentiles(values []int64, ps []float64) []float64 {
scores := make([]float64, len(ps))
size := len(values)
if size == 0 {
return scores
}
slices.Sort(values)
for i, p := range ps {
pos := p * float64(size+1)
if pos < 1.0 {
scores[i] = float64(values[0])
} else if pos >= float64(size) {
scores[i] = float64(values[size-1])
} else {
lower := float64(values[int(pos)-1])
upper := float64(values[int(pos)])
scores[i] = lower + (pos-math.Floor(pos))*(upper-lower)
}
}
return scores
}
// sampleSnapshot is a read-only copy of another Sample.
type sampleSnapshot struct {
count int64
values []int64
max int64
min int64
mean float64
sum int64
variance float64
}
// newSampleSnapshotPrecalculated creates a read-only sampleSnapShot, using
// precalculated sums to avoid iterating the values
func newSampleSnapshotPrecalculated(count int64, values []int64, min, max, sum int64) *sampleSnapshot {
if len(values) == 0 {
return &sampleSnapshot{
count: count,
values: values,
}
}
return &sampleSnapshot{
count: count,
values: values,
max: max,
min: min,
mean: float64(sum) / float64(len(values)),
sum: sum,
}
}
// newSampleSnapshot creates a read-only sampleSnapShot, and calculates some
// numbers.
func newSampleSnapshot(count int64, values []int64) *sampleSnapshot {
var (
max int64 = math.MinInt64
min int64 = math.MaxInt64
sum int64
)
for _, v := range values {
sum += v
if v > max {
max = v
}
if v < min {
min = v
}
}
return newSampleSnapshotPrecalculated(count, values, min, max, sum)
}
// Count returns the count of inputs at the time the snapshot was taken.
func (s *sampleSnapshot) Count() int64 { return s.count }
// Max returns the maximal value at the time the snapshot was taken.
func (s *sampleSnapshot) Max() int64 { return s.max }
// Mean returns the mean value at the time the snapshot was taken.
func (s *sampleSnapshot) Mean() float64 { return s.mean }
// Min returns the minimal value at the time the snapshot was taken.
func (s *sampleSnapshot) Min() int64 { return s.min }
// Percentile returns an arbitrary percentile of values at the time the
// snapshot was taken.
func (s *sampleSnapshot) Percentile(p float64) float64 {
return SamplePercentile(s.values, p)
}
// Percentiles returns a slice of arbitrary percentiles of values at the time
// the snapshot was taken.
func (s *sampleSnapshot) Percentiles(ps []float64) []float64 {
return CalculatePercentiles(s.values, ps)
}
// Size returns the size of the sample at the time the snapshot was taken.
func (s *sampleSnapshot) Size() int { return len(s.values) }
// Snapshot returns the snapshot.
func (s *sampleSnapshot) Snapshot() SampleSnapshot { return s }
// StdDev returns the standard deviation of values at the time the snapshot was
// taken.
func (s *sampleSnapshot) StdDev() float64 {
if s.variance == 0.0 {
s.variance = SampleVariance(s.mean, s.values)
}
return math.Sqrt(s.variance)
}
// Sum returns the sum of values at the time the snapshot was taken.
func (s *sampleSnapshot) Sum() int64 { return s.sum }
// Values returns a copy of the values in the sample.
func (s *sampleSnapshot) Values() []int64 {
values := make([]int64, len(s.values))
copy(values, s.values)
return values
}
// Variance returns the variance of values at the time the snapshot was taken.
func (s *sampleSnapshot) Variance() float64 {
if s.variance == 0.0 {
s.variance = SampleVariance(s.mean, s.values)
}
return s.variance
}
// SampleVariance returns the variance of the slice of int64.
func SampleVariance(mean float64, values []int64) float64 {
if len(values) == 0 {
return 0.0
}
var sum float64
for _, v := range values {
d := float64(v) - mean
sum += d * d
}
return sum / float64(len(values))
}
// A uniform sample using Vitter's Algorithm R.
//
// <http://www.cs.umd.edu/~samir/498/vitter.pdf>
type UniformSample struct {
count int64
mutex sync.Mutex
reservoirSize int
values []int64
rand *rand.Rand
}
// NewUniformSample constructs a new uniform sample with the given reservoir
// size.
func NewUniformSample(reservoirSize int) Sample {
if !Enabled {
return NilSample{}
}
return &UniformSample{
reservoirSize: reservoirSize,
values: make([]int64, 0, reservoirSize),
}
}
// SetRand sets the random source (useful in tests)
func (s *UniformSample) SetRand(prng *rand.Rand) Sample {
s.rand = prng
return s
}
// Clear clears all samples.
func (s *UniformSample) Clear() {
s.mutex.Lock()
defer s.mutex.Unlock()
s.count = 0
s.values = make([]int64, 0, s.reservoirSize)
}
// Snapshot returns a read-only copy of the sample.
func (s *UniformSample) Snapshot() SampleSnapshot {
s.mutex.Lock()
values := make([]int64, len(s.values))
copy(values, s.values)
count := s.count
s.mutex.Unlock()
return newSampleSnapshot(count, values)
}
// Update samples a new value.
func (s *UniformSample) Update(v int64) {
s.mutex.Lock()
defer s.mutex.Unlock()
s.count++
if len(s.values) < s.reservoirSize {
s.values = append(s.values, v)
} else {
var r int64
if s.rand != nil {
r = s.rand.Int63n(s.count)
} else {
r = rand.Int63n(s.count)
}
if r < int64(len(s.values)) {
s.values[int(r)] = v
}
}
}
// expDecaySample represents an individual sample in a heap.
type expDecaySample struct {
k float64
v int64
}
func newExpDecaySampleHeap(reservoirSize int) *expDecaySampleHeap {
return &expDecaySampleHeap{make([]expDecaySample, 0, reservoirSize)}
}
// expDecaySampleHeap is a min-heap of expDecaySamples.
// The internal implementation is copied from the standard library's container/heap
type expDecaySampleHeap struct {
s []expDecaySample
}
func (h *expDecaySampleHeap) Clear() {
h.s = h.s[:0]
}
func (h *expDecaySampleHeap) Push(s expDecaySample) {
n := len(h.s)
h.s = h.s[0 : n+1]
h.s[n] = s
h.up(n)
}
func (h *expDecaySampleHeap) Pop() expDecaySample {
n := len(h.s) - 1
h.s[0], h.s[n] = h.s[n], h.s[0]
h.down(0, n)
n = len(h.s)
s := h.s[n-1]
h.s = h.s[0 : n-1]
return s
}
func (h *expDecaySampleHeap) Size() int {
return len(h.s)
}
func (h *expDecaySampleHeap) Values() []expDecaySample {
return h.s
}
func (h *expDecaySampleHeap) up(j int) {
for {
i := (j - 1) / 2 // parent
if i == j || !(h.s[j].k < h.s[i].k) {
break
}
h.s[i], h.s[j] = h.s[j], h.s[i]
j = i
}
}
func (h *expDecaySampleHeap) down(i, n int) {
for {
j1 := 2*i + 1
if j1 >= n || j1 < 0 { // j1 < 0 after int overflow
break
}
j := j1 // left child
if j2 := j1 + 1; j2 < n && !(h.s[j1].k < h.s[j2].k) {
j = j2 // = 2*i + 2 // right child
}
if !(h.s[j].k < h.s[i].k) {
break
}
h.s[i], h.s[j] = h.s[j], h.s[i]
i = j
}
}