-
Notifications
You must be signed in to change notification settings - Fork 1
/
workerpool.go
713 lines (601 loc) · 15.4 KB
/
workerpool.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
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
package workerpool
import (
"container/list"
"runtime"
"sync"
"time"
log "github.com/sirupsen/logrus"
)
var (
// DefaultSizesPercentil defines regular increase ten by ten from 1 to 100.
DefaultSizesPercentil = []int{1, 10, 20, 30, 40, 50, 60, 70, 80, 90, 100}
// AllSizesPercentil defines all percentils from 1 to 100. it allows WorkerPool to find the perfect sizing fo optimal velocity. Only worth for long running operation
AllSizesPercentil = []int{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}
// LogSizesPercentil defines a logarithmic distribution from 1 to 100. Perfect for job targeting client sensible to load.
LogSizesPercentil = []int{1, 15, 23, 30, 34, 38, 42, 45, 47, 50, 52, 54, 57, 58, 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}
// AllInSizesPercentil allows only 100% of workerpool, meaning WorkerPool will always use MaxWorker goroutines
AllInSizesPercentil = []int{100}
)
// JobFnc defines the job function executed by WorkerPool workers
//
// Note: if both response and error are nil, response will not be stacked
type JobFnc func(payload interface{}) (response interface{}, err error)
// RunnableFnc is called when quorum is enabled, before accepting Tasks
type RunnableFnc func(tasks []Payload) bool
// OptFunc defines functionnal parameter to New() function
type OptFunc func(w *WorkerPool)
// Response from workers. Can be read from ReturnChannel
type Response struct {
Body interface{}
Err error
}
// Payload contains actual payload push via Feed()
// and metadata like try number
type Payload struct {
Body interface{}
Try int
ResponseChan chan Response
}
type Status string
const (
Running Status = "running"
Paused Status = "paused"
Stopped Status = "stopped"
)
// WorkerPool is an auto-scaling generic worker pool.
//
// Features include:
// - Automatic scaling in response to effective velocity (op/s)
// - Convenient worker response reading
// - Automatic scale down uppon overload
// - Easy goroutine cleanup
type WorkerPool struct {
// MaxDuration sets the maximum allowed duration to execute a job
MaxDuration time.Duration
// MaxWorker defines the maximum goroutine executing Job concurrently
MaxWorker int
// Job to be executed by WorkerPool
Job JobFnc
// ReturnChannel contains JobFnc returned values
ReturnChannel chan Response
// SizePercentil sets the possible percentage of MaxWorker usable by WorkerPool
SizePercentil []int
// EvaluationTime represents the duration of velocity collection
EvaluationTime int
// Retry will fead again payload N times on failure
// returning error only on last fail
Retry int
// KillRoutineOnError will exit goroutine if job end with error
KillRoutineOnError bool
// KillRoutineOnMaxDuration will exit goroutine if job end after max duration
KillRoutineOnMaxDuration bool
// MaxQueue limits the number of items in the queue.
// 0 is unlimited, using list.List
MaxQueue int
// Runnable called when a quorum is formed, before accepting Tasks
Runnable RunnableFnc
status Status
stopped bool
jobmu sync.Mutex
jobq *list.List
jobch chan *Payload
active int
wanted int
velocity map[int]float64
avg_sum map[int]float64
ops map[int]int
sizeindex int
responses *list.List
respmu sync.RWMutex
sync.WaitGroup
mu sync.RWMutex
quorum *Quorum
}
// Velocity information for each sizing
// rps describe the number of operation per second
// avg describe the average duration of operation
type Velocity struct {
Ops float64
Avg float64
}
// WithRetry instructs new WorkerPool to
// retry payload n times on error.
// Note: error is only sent to ReturnChannel on last retry
// Default is 0
func WithRetry(n int) OptFunc {
fn := func(wp *WorkerPool) {
wp.Retry = n
}
return fn
}
// WithMaxWorker instructs new WorkerPool to
// limit the maximum number of goroutine to max.
// Default is 500
func WithMaxWorker(max int) OptFunc {
fn := func(wp *WorkerPool) {
wp.MaxWorker = max
}
return fn
}
// WithSizePercentil creates new WorkerPool
// with given SizePercentil. Default is DefaultSizesPercentil
func WithSizePercentil(s []int) OptFunc {
fn := func(wp *WorkerPool) {
wp.SizePercentil = s
}
return fn
}
// Set evaluation time in second
func WithEvaluationTime(s int) OptFunc {
fn := func(wp *WorkerPool) {
wp.EvaluationTime = s
}
return fn
}
func WithMaxDuration(d time.Duration) OptFunc {
fn := func(wp *WorkerPool) {
wp.MaxDuration = d
}
return fn
}
// WithMaxQueue limits queue size, sometimes you may risk OOM kill...
func WithMaxQueue(max int) OptFunc {
fn := func(wp *WorkerPool) {
wp.MaxQueue = max
}
return fn
}
// WithKillRoutineOnError will exit routine if job ends on error,
// forcing slow down before next evaluation
func WithKillRoutineOnError(b bool) OptFunc {
fn := func(wp *WorkerPool) {
wp.KillRoutineOnError = b
}
return fn
}
// WithKillRoutineOnMaxDuration will exit routine if job ends after max duration,
// forcing slow down before next evaluation
func WithKillRoutineOnMaxDuration(b bool) OptFunc {
fn := func(wp *WorkerPool) {
wp.KillRoutineOnMaxDuration = b
}
return fn
}
func New(jobfnc JobFnc, opts ...OptFunc) (*WorkerPool, error) {
wp := &WorkerPool{
Job: jobfnc,
MaxDuration: 3 * time.Second,
MaxWorker: runtime.NumCPU(),
SizePercentil: DefaultSizesPercentil,
EvaluationTime: 5,
ops: make(map[int]int),
velocity: make(map[int]float64),
avg_sum: make(map[int]float64),
responses: list.New(),
jobq: list.New(),
status: Running,
}
// apply options
for _, fn := range opts {
fn(wp)
}
if wp.quorum != nil {
err := wp.quorum.Start(wp.MaxWorker, wp)
if err != nil {
return nil, err
}
}
wp.ReturnChannel = make(chan Response)
if wp.MaxQueue > 0 && wp.quorum == nil {
wp.jobch = make(chan *Payload, wp.MaxQueue)
} else {
wp.jobch = make(chan *Payload, wp.MaxWorker)
}
// spawn the first size worker
wp.setsize(0)
// start velocity routine
go wp.velocityRoutine()
go wp.responseRoutine()
if wp.MaxQueue == 0 || wp.quorum != nil {
go wp.feedRoutine()
go wp.shareRoutine()
}
return wp, nil
}
// Responses returns Response channel containing all Job returned values
func (wp *WorkerPool) Responses() chan Response {
return wp.ReturnChannel
}
// VelocityValues returns map of recorded velocity for each used velocity percentil
func (wp *WorkerPool) VelocityValues() map[int]Velocity {
c := make(map[int]Velocity)
wp.mu.RLock()
for k, v := range wp.velocity {
c[k] = Velocity{
Ops: v,
Avg: wp.avg_sum[k] / float64(wp.ops[k]),
}
}
wp.mu.RUnlock()
return c
}
func (wp *WorkerPool) CurrentVelocityValues() (percentil int, velocity float64, average float64) {
i := wp.index()
wp.mu.RLock()
percentil = wp.SizePercentil[i]
velocity = wp.velocity[wp.SizePercentil[i]]
average = wp.avg_sum[wp.SizePercentil[i]] / float64(wp.ops[wp.SizePercentil[i]])
wp.mu.RUnlock()
return percentil, velocity, average
}
func (wp *WorkerPool) GlobalVelocityValues() (map[int]Velocity, map[string]*Instance) {
// Return same format even if quorum is not enabled
if wp.quorum == nil {
all := make(map[string]*Instance)
local := wp.VelocityValues()
i := &Instance{Name: "local", Velocity: local}
all["local"] = i
return local, all
}
return wp.quorum.GlobalVelocityValues()
}
// Stop WorkerPool. All worker goroutine will exit, but stacked responses can still be consumed
func (wp *WorkerPool) Stop() {
wp.mu.Lock()
defer wp.mu.Unlock()
wp.stopped = true
wp.status = Stopped
if wp.quorum != nil {
wp.quorum.Stop()
}
}
// Pause all workers without killing them
func (wp *WorkerPool) Pause() {
wp.mu.Lock()
wp.status = Paused
}
// Resume all workers
func (wp *WorkerPool) Resume() {
wp.status = Running
wp.mu.Unlock()
}
// Status return WorkerPool current status
func (wp *WorkerPool) Status() Status {
wp.mu.RLock()
st := wp.status
wp.mu.RUnlock()
return st
}
// Active returns the number of active goroutines
func (wp *WorkerPool) Active() int {
wp.mu.RLock()
a := wp.active
wp.mu.RUnlock()
return a
}
// InQueue returns number of tasks in queue if:
// - using unlimited queue (MaxQueue = 0)
// - using quorum
//
// Otherwise returns 0
func (wp *WorkerPool) InQueue() int {
if wp.jobq == nil {
return 0
}
wp.jobmu.Lock()
n := wp.jobq.Len()
wp.jobmu.Unlock()
return n
}
// Exec job with direct response and blocking until response is received
// Usefull for ratelimiting execution
func (wp *WorkerPool) Exec(body interface{}) (response interface{}, err error) {
ch := make(chan Response)
wp.Add(1)
payload := &Payload{
Body: body,
Try: 0,
ResponseChan: ch,
}
wp.jobch <- payload
resp := <-ch
return resp.Body, resp.Err
}
// Feed payload to worker
func (wp *WorkerPool) Feed(body interface{}) {
if wp.MaxQueue > 0 && wp.quorum == nil {
wp.feedSync(body)
} else {
wp.feedAsync(body)
}
}
func (wp *WorkerPool) feedAsync(body interface{}) {
wp.Add(1)
wp.jobmu.Lock()
defer wp.jobmu.Unlock()
wp.jobq.PushBack(body)
}
func (wp *WorkerPool) feedSync(body interface{}) {
wp.Add(1)
payload := &Payload{
Body: body,
Try: 0,
}
wp.jobch <- payload
}
func (wp *WorkerPool) canAcceptTasks(tasks []Payload) bool {
wp.jobmu.Lock()
n := wp.jobq.Len()
wp.jobmu.Unlock()
if n < wp.maxWorker() {
return true
}
if wp.Runnable != nil {
return wp.Runnable(tasks)
}
return false
}
func (wp *WorkerPool) shareRoutine() {
if wp.quorum == nil {
return
}
for {
if wp.isStopped() {
return
}
// nothing to do, yay zZzZzZ
time.Sleep(10 * time.Millisecond)
// can we share the load ?
wp.jobmu.Lock()
n := wp.jobq.Len()
qsize := wp.quorum.InstancesConnected()
mw := wp.maxWorker()
if wp.quorum != nil && n > 3*mw && qsize > 1 && wp.quorum.SharedTasks() <= (qsize*10) {
// If we have a quorum
// and local queue greater than 3 times the max number of worker
// qsize if more than 1
// and there is less than 3 tasks bundle shared per instance connected
// then yes
// share MAX_WORKER last in the queue
var tasks []Payload
for i := 0; i < mw; i++ {
e := wp.jobq.Back()
body := e.Value
p := Payload{
Body: body,
Try: 0,
}
wp.jobq.Remove(e)
tasks = append(tasks, p)
}
wp.quorum.ShareTasks(tasks, wp)
}
wp.jobmu.Unlock()
}
}
func (wp *WorkerPool) feedRoutine() {
for {
if wp.isStopped() {
close(wp.jobch)
return
}
wp.jobmu.Lock()
n := wp.jobq.Len()
wp.jobmu.Unlock()
// nothing to do, yay zZzZzZ
if n == 0 {
time.Sleep(10 * time.Millisecond)
continue
}
// Feed jobch
wp.jobmu.Lock()
e := wp.jobq.Front()
body := e.Value
wp.jobmu.Unlock()
payload := &Payload{
Body: body,
Try: 0,
}
wp.jobch <- payload
wp.jobmu.Lock()
wp.jobq.Remove(e)
wp.jobmu.Unlock()
}
}
// AvailableResponses returns the current number of stacked responses. Consume ReturnChannel to read them
func (wp *WorkerPool) AvailableResponses() int {
wp.respmu.RLock()
n := wp.responses.Len()
wp.respmu.RUnlock()
return n
}
func (wp *WorkerPool) index() int {
wp.mu.RLock()
si := wp.sizeindex
wp.mu.RUnlock()
return si
}
func (wp *WorkerPool) evaluate(d time.Duration, err error) bool {
if wp.KillRoutineOnError && err != nil {
return wp.exit()
}
if wp.KillRoutineOnMaxDuration && d > wp.MaxDuration*time.Second {
return wp.exit()
}
wp.mu.RLock()
active := wp.active
wanted := wp.wanted
wp.mu.RUnlock()
if active > wanted {
return wp.exit()
}
return false
}
func (wp *WorkerPool) exit() bool {
wp.mu.Lock()
defer wp.mu.Unlock()
// leave at least 1 worker
if wp.active == 1 {
return false
}
wp.active--
return true
}
func (wp *WorkerPool) spawn() {
wp.active++
go wp.worker()
}
func (wp *WorkerPool) worker() {
var err error
var body interface{}
for {
p, ok := <-wp.jobch
if !ok {
return
}
begin := time.Now()
body, err = wp.Job(p.Body)
t := time.Since(begin)
wp.tick(t)
if body != nil || p.ResponseChan != nil || (err != nil && p.Try == wp.Retry) {
r := Response{
Body: body,
Err: err,
}
if p.ResponseChan != nil {
p.ResponseChan <- r
} else {
wp.pushResponse(r)
}
}
// maybe retry
if err != nil && p.Try < wp.Retry {
p.Try++
wp.retry(p)
} else {
wp.Done()
}
if shouldExit := wp.evaluate(t, err); shouldExit {
return
}
}
}
func (wp *WorkerPool) isStopped() bool {
wp.mu.RLock()
stopped := wp.stopped
wp.mu.RUnlock()
return stopped
}
func (wp *WorkerPool) tick(d time.Duration) {
wp.mu.Lock()
wp.ops[wp.SizePercentil[wp.sizeindex]]++
wp.avg_sum[wp.SizePercentil[wp.sizeindex]] += float64(d.Milliseconds())
wp.mu.Unlock()
}
func (wp *WorkerPool) velocityRoutine() {
for {
if wp.isStopped() {
return
}
wp.mu.Lock()
wp.ops[wp.SizePercentil[wp.sizeindex]] = 0
wp.avg_sum[wp.SizePercentil[wp.sizeindex]] = 0
wp.mu.Unlock()
time.Sleep(time.Duration(wp.EvaluationTime) * time.Second)
i := wp.index()
if wp.quorum != nil {
mw := wp.quorum.MaxInstanceWorker()
wp.mu.Lock()
wp.MaxWorker = mw
wp.mu.Unlock()
wp.setsize(i)
}
wp.mu.Lock()
wp.velocity[wp.SizePercentil[i]] = float64(wp.ops[wp.SizePercentil[i]]) / float64(wp.EvaluationTime)
wp.mu.Unlock()
wp.mu.RLock()
// always increase from first value
if i == 0 && i < len(wp.SizePercentil)-1 {
wp.mu.RUnlock()
wp.setsize(i + 1)
continue
}
// if velocity is 0, decrease
if i > 0 && wp.velocity[wp.SizePercentil[i]] == 0 {
wp.mu.RUnlock()
wp.setsize(i - 1)
continue
}
// if velocity increased, then increase worker pool size
if i < len(wp.SizePercentil)-1 && wp.velocity[wp.SizePercentil[i]] > wp.velocity[wp.SizePercentil[i-1]] {
wp.mu.RUnlock()
wp.setsize(i + 1)
continue
}
// if velocity decreased then decrease worker pool size
if wp.sizeindex > 0 && wp.velocity[wp.SizePercentil[i]] < wp.velocity[wp.SizePercentil[i-1]] {
wp.mu.RUnlock()
wp.setsize(i - 1)
continue
}
wp.mu.RUnlock()
}
}
func (wp *WorkerPool) setsize(i int) {
wp.mu.Lock()
wp.sizeindex = i
wp.wanted = int(wp.MaxWorker * wp.SizePercentil[i] / 100)
if wp.wanted == 0 {
wp.wanted++
}
log.Debugf("MaxWorker: %d, SizePercentil: %d, Wanted: %d, Active: %d\n", wp.MaxWorker, wp.SizePercentil[i], wp.wanted, wp.active)
active := wp.active
wanted := wp.wanted
defer wp.mu.Unlock()
if active < wanted {
for i := active; i < wanted; i++ {
wp.spawn()
}
}
}
func (wp *WorkerPool) pushResponse(r Response) {
wp.respmu.Lock()
wp.responses.PushBack(r)
wp.respmu.Unlock()
}
func (wp *WorkerPool) responseRoutine() {
var n int
for {
n = wp.AvailableResponses()
if wp.isStopped() && n == 0 {
close(wp.ReturnChannel)
return
}
if n == 0 {
time.Sleep(1 * time.Millisecond)
continue
}
wp.respmu.RLock()
e := wp.responses.Front()
wp.respmu.RUnlock()
r, ok := e.Value.(Response)
if !ok {
continue
}
wp.ReturnChannel <- r
wp.respmu.Lock()
wp.responses.Remove(e)
wp.respmu.Unlock()
}
}
func (wp *WorkerPool) retry(payload *Payload) {
go func() {
wp.jobch <- payload
}()
}
func (wp *WorkerPool) maxWorker() int {
wp.mu.RLock()
mw := wp.MaxWorker
wp.mu.RUnlock()
return mw
}