-
Notifications
You must be signed in to change notification settings - Fork 442
/
worker.go
274 lines (238 loc) · 7.57 KB
/
worker.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
package collector
import (
"fmt"
"sort"
"sync/atomic"
"time"
conf "github.com/alibaba/MongoShake/v2/collector/configure"
utils "github.com/alibaba/MongoShake/v2/common"
"github.com/alibaba/MongoShake/v2/oplog"
"github.com/alibaba/MongoShake/v2/tunnel"
nimo "github.com/gugemichael/nimo4go"
LOG "github.com/vinllen/log4go"
)
const MaxUnAckListLength = 128 * 256
type Worker struct {
// parent syncer
syncer *OplogSyncer
// worker sequence id
id uint32
// job queue from OplogTailer
queue chan []*oplog.GenericOplog
// worker tunnel controller (include tunnel and modules)
writeController *WriteController
// buffered oplogs . include unack buffer
// and send buffer
listUnACK []*oplog.GenericOplog
// ack offset (used for checkpoint)
ack, unack int64
count uint64
// if all oplogs are acked for righ now
allAcked bool
// retransmit on tunnel controller tunnel required
retransmit bool
// event listener
eventListener *TransferEventListener
}
type TransferEventListener struct {
whenTransferBatchSuccess func(worker *Worker, buffer []*oplog.GenericOplog)
whenTransferRetry func(worker *Worker, buffer []*oplog.GenericOplog)
}
func NewWorker(syncer *OplogSyncer, id uint32) *Worker {
return &Worker{
syncer: syncer,
id: id,
queue: make(chan []*oplog.GenericOplog, conf.Options.IncrSyncWorkerBatchQueueSize),
}
}
func (worker *Worker) String() string {
return fmt.Sprintf("Collector-worker-%d", worker.id)
}
func (worker *Worker) Init() bool {
worker.RestAPI()
worker.writeController = NewWriteController(worker)
if worker.writeController == nil {
return false
}
return true
}
func (worker *Worker) SetInitSyncFinishTs(fullSyncFinishPosition int64) {
worker.writeController.SetInitSyncFinishTs(fullSyncFinishPosition)
}
func (worker *Worker) IsAllAcked() bool {
return worker.allAcked
}
func (worker *Worker) AllAcked(allAcked bool) {
worker.allAcked = allAcked
}
func (worker *Worker) Offer(batch []*oplog.GenericOplog) {
if batch != nil {
atomic.StoreInt64(&worker.unack, utils.TimeStampToInt64(batch[len(batch)-1].Parsed.Timestamp))
}
worker.queue <- batch
}
func (worker *Worker) shouldDelay() bool {
// unack buffer is too big. There should be a mass of accumulated oplogs
// have already sent but not be ack yet. No more oplogs pushed !
if len(worker.listUnACK) > MaxUnAckListLength {
// try to transfer remained oplogs to free some space
return true
}
return false
}
func (worker *Worker) shouldStall() bool {
// suspend while system pause is set. This is always operated by outside
// manual system. It needs stopping here util turned off
return utils.IncrSentinelOptions.Pause
}
func (worker *Worker) findFirstAvailableBatch() []*oplog.GenericOplog {
var batch []*oplog.GenericOplog
for {
select {
case batch = <-worker.queue:
case <-time.After(DDLCheckpointInterval * time.Millisecond): // timeout, add probe message here
return nil
}
// if we got empty (nil) batch. check the queue still has more
// oplogs. That will merge continuous nil batch to single one
if batch != nil || len(worker.queue) == 0 {
return batch
}
}
}
func (worker *Worker) StartWorker() {
LOG.Info("%s start working with jobs batch queue. buffer capacity %d",
worker, cap(worker.queue))
var batch []*oplog.GenericOplog
for {
switch {
case worker.shouldDelay():
// we guess there were lots of oplogs have been pended in jobs queue.
// we need wait for a few moment
worker.probe()
worker.purgeACK()
fallthrough
case worker.shouldStall():
utils.DelayFor(10)
default:
if batch = worker.findFirstAvailableBatch(); batch == nil {
worker.probe()
} else {
worker.transfer(batch)
worker.syncer.replMetric.AddConsume(uint64(len(batch)))
}
utils.DEBUG_LOG("%s poll queued batch oplogs. total[%d]", worker, len(batch))
}
}
}
/*
*
* [ Before transfer ]
*
* batch |9,10,11|
* listSent |1,2,3,4,5,6,7,8|
*
* [ After transfer ]
*
* batch | (empty) |
* listSent |1,2,3,4,5,6,7,8,9,10,11|
*
* [ Purge listWaitACK (ack == 7) ]
*
* listUnACK |8,9,10,11|
*
*/
func (worker *Worker) transfer(batch []*oplog.GenericOplog) {
nimo.AssertTrue(batch != nil, "batch oplogs should not empty")
var logs []*oplog.GenericOplog
var tag uint32
done := false
// transfer util current batch is sent(done == true)
for !done {
if worker.retransmit {
// TODO: send all unack logs ? it's possible very big
logs = worker.listUnACK
tag = tunnel.MsgRetransmission
worker.syncer.replMetric.AddRetransmission(1)
} else {
logs = batch
tag = tunnel.MsgNormal
}
replyAndAcked := worker.writeController.Send(logs, tag)
LOG.Info("%s transfer retransmit:%t send [%d] logs. reply_acked [%v], list_unack [%d] ",
worker, worker.retransmit, len(logs), utils.ExtractTimestampForLog(replyAndAcked), len(worker.listUnACK))
switch {
case replyAndAcked >= 0:
if !worker.retransmit {
worker.count += uint64(len(logs))
worker.syncer.replMetric.SetLSNACK(replyAndAcked)
worker.syncer.replMetric.AddApply(uint64(len(logs)))
worker.retain(batch)
// update ack
atomic.StoreInt64(&worker.ack, replyAndAcked)
done = true
}
// remove all unack values which are smaller than worker.ack
worker.purgeACK()
// reset
worker.retransmit = false
// notify success listener
worker.syncer.replMetric.ReplStatus.Clear(utils.TunnelSendBad)
case replyAndAcked == tunnel.ReplyRetransmission:
LOG.Info("%s received ReplyRetransmission reply, now status %t", worker, worker.retransmit)
// next step. keep trying with retransmission util we received
// a non-retransmission message
worker.retransmit = true
default:
LOG.Warn("%s transfer oplogs failed with reply value %d", worker, replyAndAcked)
// we treat batched logs fail as just one time failed. and
// notify failed retry listener
worker.syncer.replMetric.AddFailed(1)
//worker.retransmit = true
worker.syncer.replMetric.ReplStatus.Update(utils.TunnelSendBad)
}
}
}
func (worker *Worker) probe() {
if replyAcked := worker.writeController.Send([]*oplog.GenericOplog{}, tunnel.MsgProbe); replyAcked > 0 {
// only change ack offset on reply is OK
worker.syncer.replMetric.SetLSNACK(replyAcked)
atomic.StoreInt64(&worker.ack, replyAcked)
}
}
func (worker *Worker) retain(batch []*oplog.GenericOplog) {
worker.listUnACK = append(worker.listUnACK, batch...)
LOG.Debug("%s copy batch oplogs [%d] to listUnACK count. UnACK remained [%d]",
worker, len(batch), len(worker.listUnACK))
}
func (worker *Worker) purgeACK() {
bigger := sort.Search(len(worker.listUnACK), func(i int) bool {
return utils.TimeStampToInt64(worker.listUnACK[i].Parsed.Timestamp) > worker.ack
})
if bigger != 0 {
LOG.Debug("%s purge unacked [lsn_ack:%d]. keep slice position from %d util %d",
worker, worker.ack, bigger, len(worker.listUnACK))
worker.listUnACK = worker.listUnACK[bigger:]
worker.syncer.replMetric.AddSuccess(uint64(bigger))
}
}
func (worker *Worker) RestAPI() {
type WorkerInfo struct {
Id uint32 `json:"worker_id"`
JobsQueued int `json:"jobs_in_queue"`
JobsUnACKBuffer int `json:"jobs_unack_buffer"`
LastUnACK string `json:"last_unack"`
LastACK string `json:"last_ack"`
COUNT uint64 `json:"count"`
}
utils.IncrSyncHttpApi.RegisterAPI("/worker", nimo.HttpGet, func([]byte) interface{} {
return &WorkerInfo{
Id: worker.id,
JobsQueued: len(worker.queue),
JobsUnACKBuffer: len(worker.listUnACK),
LastUnACK: utils.Int64ToString(worker.unack),
LastACK: utils.Int64ToString(worker.ack),
COUNT: worker.count,
}
})
}