forked from codemartial/smartcb
-
Notifications
You must be signed in to change notification settings - Fork 2
/
smartcb.go
216 lines (183 loc) · 5.78 KB
/
smartcb.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
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
package smartcb
import (
"math"
"sync"
"time"
"github.com/rubyist/circuitbreaker"
)
// Policies for configuring the circuit breaker's decision making.
//
// MaxFail is the only parameter that might need adjustment.
// Do not tweak the other parameters unless you are a statistician.
// If you must, experiment with changing one parameter at a time.
// All parameters are required to be > 0
//
type Policies struct {
// Absolute highest failure rate above which the breaker must open
// Default is 0.05 (5%).
MaxFail float64
// Number of "decision windows" used for learning
LearningWindowX float64
// Number of "decision windows" after which learning is restarted.
//
// This setting must be greater than LearningWindowX otherwise the breaker
// would be in a perpetual learning state
ReLearningWindowX float64
// Smoothing factor for error rate learning. Higher numbers reduce jitter
// but cause more lag
EWMADecayFactor float64
// Number of trials in a decision window.
SamplesPerWindow int64
}
var defaults = Policies{
MaxFail: 0.05,
LearningWindowX: 10.0,
ReLearningWindowX: 100.0,
EWMADecayFactor: 10.0,
SamplesPerWindow: 1000,
}
// Circuit Breaker's Learning State
type State int
const (
// Circuit Breaker has learned
Learned State = iota
// Circuit Breaker is learning
Learning
)
const minFail = 0.001
func (s State) String() string {
switch s {
case Learning:
return "Learning"
case Learned:
fallthrough
default:
return "Learned"
}
}
// A Smart TripFunction Generator
//
// All circuit breakers obtained out of a generator
// share their learning state, but the circuit breaker state
// (error rates, event counts, etc.) is not shared
//
type SmartTripper struct {
decisionWindow time.Duration
policies Policies
state State
rate float64
initTime time.Time
mu sync.Mutex
}
// Returns Policies initialised to default values
//
func NewPolicies() Policies {
return defaults
}
// Create a SmartTripper based on the nominal QPS for your task
//
// "Nominal QPS" is the basis on which the SmartTripper configures its
// responsiveness settings. A suitable value for this parameter would be
// your median QPS. If your QPS varies a lot during operation, choosing this
// value closer to max QPS will make the circuit breaker more prone to tripping
// during low traffic periods and choosing a value closer to min QPS will make it
// slow to respond during high traffic periods.
//
// NOTE: Provide QPS value applicable for one instance of the circuit breaker,
// not the overall QPS across multiple instances.
//
func NewSmartTripper(QPS int, p Policies) *SmartTripper {
if QPS <= 0 {
panic("smartcb.NewSmartTripper: QPS should be >= 1")
}
decisionWindow := time.Millisecond * time.Duration(float64(p.SamplesPerWindow)*1000.0/float64(QPS))
return &SmartTripper{decisionWindow: decisionWindow, policies: p, rate: minFail}
}
// Returns the Learning/Learned state of the Smart Tripper
//
// State change only happens when an error is triggered
// Therefore timing alone can not be relied upon to detect state changes
func (t *SmartTripper) State() State {
t.mu.Lock()
defer t.mu.Unlock()
return t.state
}
// Returns the error rate that has been learned by the Smart Tripper
//
func (t *SmartTripper) LearnedRate() float64 {
t.mu.Lock()
defer t.mu.Unlock()
return t.rate
}
func (t *SmartTripper) tripFunc() circuit.TripFunc {
learningCycles := t.policies.LearningWindowX
relearningCycles := t.policies.ReLearningWindowX
maxFail := t.policies.MaxFail
initLearning := func(cb *circuit.Breaker) {
t.initTime = time.Now()
t.state = Learning
}
recordError := func(cbr, samples float64) bool {
weightage := samples / float64(t.policies.SamplesPerWindow)
t.rate += (cbr - t.rate) * weightage / (t.policies.EWMADecayFactor + weightage)
// Enforce learned error rate limits
if t.rate < minFail {
t.rate = minFail
}
if t.rate > maxFail {
t.rate = maxFail
}
return false
}
// Use Adjusted Wald Method to estimate whether we are confident enough to trip based on the no. of samples
shouldPerhapsTrip := func(target, actual float64, sampleSize int64) bool {
ss := float64(sampleSize)
ssig := float64(t.policies.SamplesPerWindow)
if ss > ssig {
ss = ssig
}
pf := (ssig - ss) / (ssig - 1)
fearFactor := math.Sqrt(pf*actual*(1-actual)/ss) * 2.58 // 2.58 = z-Critical at 99% confidence
return actual-fearFactor > target
}
tripper := func(cb *circuit.Breaker) bool {
t.mu.Lock()
defer t.mu.Unlock()
tElapsed := time.Since(t.initTime)
// Initiate Learning Phase
if t.initTime == (time.Time{}) || tElapsed > t.decisionWindow*time.Duration(relearningCycles) {
initLearning(cb)
tElapsed = time.Since(t.initTime)
}
cycles := float64(tElapsed) / float64(t.decisionWindow)
// Terminate Learning Phase
if t.state == Learning && cycles > learningCycles {
t.state = Learned
}
samples := cb.Failures() + cb.Successes()
errorRate := cb.ErrorRate()
if samples < t.policies.SamplesPerWindow/10 { // Not enough data to decide
return false
}
if t.state == Learning {
tripRate := math.Sqrt(maxFail/t.rate) * t.rate
// Either trip or learn the error rate
return shouldPerhapsTrip(tripRate, errorRate, samples) || recordError(errorRate, float64(samples))
}
return shouldPerhapsTrip(math.Sqrt(maxFail/t.rate)*t.rate,
errorRate,
samples)
}
return tripper
}
// Create a new circuit.Breaker using the dynamically self-configuring SmartTripper
//
// It returns a circuit.Breaker from github.com/rubyist/circuitbreaker
// Please see its documentation to understand how to use the breaker
func NewSmartCircuitBreaker(t *SmartTripper) *circuit.Breaker {
options := &circuit.Options{
WindowTime: t.decisionWindow,
}
options.ShouldTrip = t.tripFunc()
return circuit.NewBreakerWithOptions(options)
}