forked from rcrowley/go-metrics
-
Notifications
You must be signed in to change notification settings - Fork 0
/
sample.go
191 lines (174 loc) · 4.45 KB
/
sample.go
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
package metrics
import (
"math"
"rand"
"time"
)
const rescaleThreshold = 1e9 * 60 * 60
// Samples maintain a statistically-significant selection of values from
// a stream.
//
// This is an interface so as to encourage other structs to implement
// the Sample API as appropriate.
type Sample interface {
Clear()
Size() int
Update(int64)
Values() []int64
}
// 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://www.research.att.com/people/Cormode_Graham/library/publications/CormodeShkapenyukSrivastavaXu09.pdf>
type ExpDecaySample struct {
reservoirSize int
alpha float64
in chan int64
out chan []int64
reset chan bool
}
// Create a new exponentially-decaying sample with the given reservoir size
// and alpha.
func NewExpDecaySample(reservoirSize int, alpha float64) *ExpDecaySample {
s := &ExpDecaySample{
reservoirSize,
alpha,
make(chan int64),
make(chan []int64),
make(chan bool),
}
go s.arbiter()
return s
}
// Clear all samples.
func (s *ExpDecaySample) Clear() {
s.reset <- true
}
// Return the size of the sample, which is at most the reservoir size.
func (s *ExpDecaySample) Size() int {
return len(<-s.out)
}
// Update the sample with a new value.
func (s *ExpDecaySample) Update(v int64) {
s.in <- v
}
// Return all the values in the sample.
func (s *ExpDecaySample) Values() []int64 {
return <-s.out
}
// Receive inputs and send outputs. Count and save each input value,
// rescaling the sample if enough time has elapsed since the last rescaling.
// Send a copy of the values as output.
func (s *ExpDecaySample) arbiter() {
count := 0
values := make(map[float64]int64)
tsStart := time.Seconds()
tsNext := time.Nanoseconds() + rescaleThreshold
var valuesCopy []int64
for {
select {
case v := <-s.in:
ts := time.Seconds()
k := math.Exp(float64(ts - tsStart) * s.alpha) / rand.Float64()
count++
values[k] = v
if count > s.reservoirSize {
min := math.MaxFloat64
for k, _ := range values {
if k < min { min = k }
}
values[min] = 0, false
valuesCopy = make([]int64, s.reservoirSize)
} else {
valuesCopy = make([]int64, count)
}
tsNano := time.Nanoseconds()
if tsNano > tsNext {
tsOldStart := tsStart
tsStart = time.Seconds()
tsNext = tsNano + rescaleThreshold
oldValues := values
values = make(map[float64]int64, len(oldValues))
for k, v := range oldValues {
values[k * math.Exp(-s.alpha * float64(
tsStart - tsOldStart))] = v
}
}
i := 0
for _, v := range values {
valuesCopy[i] = v
i++
}
case s.out <- valuesCopy: // TODO Might need to make another copy here.
case <-s.reset:
count = 0
values = make(map[float64]int64)
valuesCopy = make([]int64, 0)
tsStart = time.Seconds()
tsNext = tsStart + rescaleThreshold
}
}
}
// A uniform sample using Vitter's Algorithm R.
//
// <http://www.cs.umd.edu/~samir/498/vitter.pdf>
type UniformSample struct {
reservoirSize int
in chan int64
out chan []int64
reset chan bool
}
// Create a new uniform sample with the given reservoir size.
func NewUniformSample(reservoirSize int) *UniformSample {
s := &UniformSample{
reservoirSize,
make(chan int64),
make(chan []int64),
make(chan bool),
}
go s.arbiter()
return s
}
// Clear all samples.
func (s *UniformSample) Clear() {
s.reset <- true
}
// Return the size of the sample, which is at most the reservoir size.
func (s *UniformSample) Size() int {
return len(<-s.out)
}
// Update the sample with a new value.
func (s *UniformSample) Update(v int64) {
s.in <- v
}
// Return all the values in the sample.
func (s *UniformSample) Values() []int64 {
return <-s.out
}
// Receive inputs and send outputs. Count and save each input value at a
// random index. Send a copy of the values as output.
func (s *UniformSample) arbiter() {
count := 0
values := make([]int64, s.reservoirSize)
var valuesCopy []int64
for {
select {
case v := <-s.in:
count++
if count < s.reservoirSize {
values[count - 1] = v
valuesCopy = make([]int64, count)
} else {
values[rand.Intn(s.reservoirSize)] = v
valuesCopy = make([]int64, len(values))
}
for i := 0; i < len(valuesCopy); i++ { valuesCopy[i] = values[i] }
case s.out <- valuesCopy: // TODO Might need to make another copy here.
case <-s.reset:
count = 0
values = make([]int64, s.reservoirSize)
valuesCopy = make([]int64, 0)
}
}
}