forked from elastic/beats
-
Notifications
You must be signed in to change notification settings - Fork 0
/
acker.go
435 lines (351 loc) · 9.22 KB
/
acker.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
package pipeline
import (
"sync"
"time"
"github.com/elastic/beats/libbeat/beat"
"github.com/elastic/beats/libbeat/common/atomic"
)
// acker is used to account for published and non-published events to be ACKed
// to the beats client.
// All pipeline and client ACK handling support is provided by acker instances.
type acker interface {
close()
addEvent(event beat.Event, published bool) bool
ackEvents(int)
}
// emptyACK ignores any ACK signals and events.
type emptyACK struct{}
var nilACKer acker = (*emptyACK)(nil)
func (*emptyACK) close() {}
func (*emptyACK) addEvent(_ beat.Event, _ bool) bool { return true }
func (*emptyACK) ackEvents(_ int) {}
type ackerFn struct {
Close func()
AddEvent func(beat.Event, bool) bool
AckEvents func(int)
}
func (a *ackerFn) close() { a.Close() }
func (a *ackerFn) addEvent(e beat.Event, b bool) bool { return a.AddEvent(e, b) }
func (a *ackerFn) ackEvents(n int) { a.AckEvents(n) }
// countACK is used when broker ACK events can be simply forwarded to the
// producers ACKCount callback.
// The countACK is only applicable if no processors are configured.
// ACKs for closed clients will be ignored.
type countACK struct {
pipeline *Pipeline
fn func(total, acked int)
}
func newCountACK(fn func(total, acked int)) *countACK {
a := &countACK{fn: fn}
return a
}
func (a *countACK) close() {}
func (a *countACK) addEvent(_ beat.Event, _ bool) bool { return true }
func (a *countACK) ackEvents(n int) {
if a.pipeline.ackActive.Load() {
a.fn(n, n)
}
}
// gapCountACK returns event ACKs to the producer, taking account for dropped events.
// Events being dropped by processors will always be ACKed with the last batch ACKed
// by the broker. This way clients waiting for ACKs can expect all processed
// events being always ACKed.
type gapCountACK struct {
pipeline *Pipeline
fn func(total int, acked int)
done chan struct{}
drop chan struct{}
acks chan int
events atomic.Uint32
lst gapList
}
type gapList struct {
sync.Mutex
head, tail *gapInfo
}
type gapInfo struct {
sync.Mutex
next *gapInfo
send, dropped int
}
func newGapCountACK(pipeline *Pipeline, fn func(total, acked int)) *gapCountACK {
a := &gapCountACK{}
a.init(pipeline, fn)
return a
}
func (a *gapCountACK) init(pipeline *Pipeline, fn func(int, int)) {
*a = gapCountACK{
pipeline: pipeline,
fn: fn,
done: make(chan struct{}),
drop: make(chan struct{}),
acks: make(chan int, 1),
}
init := &gapInfo{}
a.lst.head = init
a.lst.tail = init
go a.ackLoop()
}
func (a *gapCountACK) ackLoop() {
// close channels, as no more events should be ACKed:
// - once pipeline is closed
// - all events of the closed client have been acked/processed by the pipeline
acks, drop := a.acks, a.drop
closing := false
for {
select {
case <-a.done:
closing = true
a.done = nil
case <-a.pipeline.ackDone:
return
case n := <-acks:
empty := a.handleACK(n)
if empty && closing && a.events.Load() == 0 {
// stop worker, iff all events accounted for have been ACKed
return
}
case <-drop:
// TODO: accumulate multiple drop events + flush count with timer
a.fn(1, 0)
}
}
}
func (a *gapCountACK) handleACK(n int) bool {
// collect items and compute total count from gapList
var (
total = 0
acked = n
emptyLst bool
)
for n > 0 {
if emptyLst {
panic("too many events acked")
}
a.lst.Lock()
current := a.lst.head
current.Lock()
if n >= current.send {
nxt := current.next
emptyLst = nxt == nil
if !emptyLst {
// advance list all event in current entry have been send and list as
// more then 1 gapInfo entry. If only 1 entry is present, list item will be
// reset and reused
a.lst.head = nxt
}
}
// hand over lock list-entry, so ACK handler and producer can operate
// on potentially different list ends
a.lst.Unlock()
if n < current.send {
current.send -= n
total += n
n = 0
} else {
total += current.send + current.dropped
n -= current.send
current.dropped = 0
current.send = 0
}
current.Unlock()
}
a.events.Sub(uint32(total))
a.fn(total, acked)
return emptyLst
}
func (a *gapCountACK) close() {
// close client only, pipeline itself still can handle pending ACKs
close(a.done)
}
func (a *gapCountACK) addEvent(_ beat.Event, published bool) bool {
// if gapList is empty and event is being dropped, forward drop event to ack
// loop worker:
a.events.Inc()
if !published {
a.addDropEvent()
} else {
a.addPublishedEvent()
}
return true
}
func (a *gapCountACK) addDropEvent() {
a.lst.Lock()
current := a.lst.tail
current.Lock()
if current.send == 0 && current.next == nil {
// send can only be 0 if no no events/gaps present yet
if a.lst.head != a.lst.tail {
panic("gap list expected to be empty")
}
current.Unlock()
a.lst.Unlock()
a.drop <- struct{}{}
} else {
a.lst.Unlock()
current.dropped++
current.Unlock()
}
}
func (a *gapCountACK) addPublishedEvent() {
// event is publisher -> add a new gap list entry if gap is present in current
// gapInfo
a.lst.Lock()
current := a.lst.tail
current.Lock()
if current.dropped > 0 {
tmp := &gapInfo{}
a.lst.tail.next = tmp
a.lst.tail = tmp
current.Unlock()
tmp.Lock()
current = tmp
}
a.lst.Unlock()
current.send++
current.Unlock()
}
func (a *gapCountACK) ackEvents(n int) {
select {
case <-a.pipeline.ackDone: // pipeline is closing down -> ignore event
a.acks = nil
case a.acks <- n: // send ack event to worker
}
}
// boundGapCountACK guards a gapCountACK instance by bounding the maximum number of
// active events.
// As beats might accumulate state while waiting for ACK, the boundGapCountACK blocks
// if too many events have been filtered out by processors.
type boundGapCountACK struct {
active bool
fn func(total, acked int)
acker gapCountACK
sema *sema
}
func newBoundGapCountACK(
pipeline *Pipeline,
sema *sema,
fn func(total, acked int),
) *boundGapCountACK {
a := &boundGapCountACK{active: true, sema: sema, fn: fn}
a.acker.init(pipeline, a.onACK)
return a
}
func (a *boundGapCountACK) close() {
a.acker.close()
}
func (a *boundGapCountACK) addEvent(event beat.Event, published bool) bool {
a.sema.inc()
return a.acker.addEvent(event, published)
}
func (a *boundGapCountACK) ackEvents(n int) { a.acker.ackEvents(n) }
func (a *boundGapCountACK) onACK(total, acked int) {
a.sema.release(total)
a.fn(total, acked)
}
// eventDataACK reports all dropped and ACKed events private fields.
// An instance of eventDataACK requires a counting ACKer (boundGapCountACK or countACK),
// for accounting for potentially dropped events.
type eventDataACK struct {
mutex sync.Mutex
acker acker
pipeline *Pipeline
// TODO: replace with more efficient dynamic sized ring-buffer?
data []interface{}
fn func(data []interface{}, acked int)
}
func newEventACK(
pipeline *Pipeline,
canDrop bool,
sema *sema,
fn func([]interface{}, int),
) *eventDataACK {
a := &eventDataACK{pipeline: pipeline, fn: fn}
a.acker = makeCountACK(pipeline, canDrop, sema, a.onACK)
return a
}
func makeCountACK(pipeline *Pipeline, canDrop bool, sema *sema, fn func(int, int)) acker {
if canDrop {
return newBoundGapCountACK(pipeline, sema, fn)
}
return newCountACK(fn)
}
func (a *eventDataACK) close() {
a.acker.close()
}
func (a *eventDataACK) addEvent(event beat.Event, published bool) bool {
a.mutex.Lock()
active := a.pipeline.ackActive.Load()
if active {
a.data = append(a.data, event.Private)
}
a.mutex.Unlock()
if active {
return a.acker.addEvent(event, published)
}
return false
}
func (a *eventDataACK) ackEvents(n int) { a.acker.ackEvents(n) }
func (a *eventDataACK) onACK(total, acked int) {
n := total
a.mutex.Lock()
data := a.data[:n]
a.data = a.data[n:]
a.mutex.Unlock()
if len(data) > 0 && a.pipeline.ackActive.Load() {
a.fn(data, acked)
}
}
// waitACK keeps track of events being produced and ACKs for events.
// On close waitACK will wait for pending events to be ACKed by the broker.
// The acker continues the closing operation if all events have been published
// or the maximum configured sleep time has been reached.
type waitACK struct {
acker acker
signal chan struct{}
waitClose time.Duration
active atomic.Bool
// number of active events
events atomic.Uint64
}
func newWaitACK(acker acker, timeout time.Duration) *waitACK {
return &waitACK{
acker: acker,
signal: make(chan struct{}, 1),
waitClose: timeout,
active: atomic.MakeBool(true),
}
}
func (a *waitACK) close() {
// TODO: wait for events
a.active.Store(false)
if a.events.Load() > 0 {
select {
case <-a.signal:
case <-time.After(a.waitClose):
}
}
// close the underlying acker upon exit
a.acker.close()
}
func (a *waitACK) addEvent(event beat.Event, published bool) bool {
if published {
a.events.Inc()
}
return a.acker.addEvent(event, published)
}
func (a *waitACK) ackEvents(n int) {
// return ACK signal to upper layers
a.acker.ackEvents(n)
a.releaseEvents(n)
}
func (a *waitACK) releaseEvents(n int) {
value := a.events.Sub(uint64(n))
if value != 0 {
return
}
// send done signal, if close is waiting
if !a.active.Load() {
a.signal <- struct{}{}
}
}