-
Notifications
You must be signed in to change notification settings - Fork 5.8k
/
write.go
348 lines (312 loc) · 8.22 KB
/
write.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
// Copyright 2019-present PingCAP, Inc.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// See the License for the specific language governing permissions and
// limitations under the License.
package tikv
import (
"bytes"
"math"
"sync"
"sync/atomic"
"github.com/cznic/mathutil"
"github.com/pingcap/badger"
"github.com/pingcap/badger/y"
"github.com/pingcap/kvproto/pkg/kvrpcpb"
"github.com/pingcap/tidb/store/mockstore/unistore/lockstore"
"github.com/pingcap/tidb/store/mockstore/unistore/tikv/dbreader"
"github.com/pingcap/tidb/store/mockstore/unistore/tikv/mvcc"
)
const (
batchChanSize = 1024
)
type writeDBBatch struct {
entries []*badger.Entry
err error
wg sync.WaitGroup
}
func newWriteDBBatch() *writeDBBatch {
return &writeDBBatch{}
}
func (batch *writeDBBatch) set(key y.Key, val []byte, userMeta []byte) {
batch.entries = append(batch.entries, &badger.Entry{
Key: key,
Value: val,
UserMeta: userMeta,
})
}
// delete is a badger level operation, only used in DeleteRange, so we don't need to set UserMeta.
// Then we can tell the entry is delete if UserMeta is nil.
func (batch *writeDBBatch) delete(key y.Key) {
batch.entries = append(batch.entries, &badger.Entry{
Key: key,
})
}
type writeLockBatch struct {
entries []*badger.Entry
err error
wg sync.WaitGroup
}
func (batch *writeLockBatch) set(key, val []byte) {
batch.entries = append(batch.entries, &badger.Entry{
Key: y.KeyWithTs(key, 0),
Value: val,
UserMeta: mvcc.LockUserMetaNone,
})
}
func (batch *writeLockBatch) delete(key []byte) {
batch.entries = append(batch.entries, &badger.Entry{
Key: y.KeyWithTs(key, 0),
UserMeta: mvcc.LockUserMetaDelete,
})
}
type writeDBWorker struct {
batchCh chan *writeDBBatch
writer *dbWriter
}
func (w writeDBWorker) run() {
defer w.writer.wg.Done()
var batches []*writeDBBatch
for {
for i := range batches {
batches[i] = nil
}
batches = batches[:0]
select {
case <-w.writer.closeCh:
return
case batch := <-w.batchCh:
batches = append(batches, batch)
}
chLen := len(w.batchCh)
for i := 0; i < chLen; i++ {
batches = append(batches, <-w.batchCh)
}
if len(batches) > 0 {
w.updateBatchGroup(batches)
}
}
}
func (w writeDBWorker) updateBatchGroup(batchGroup []*writeDBBatch) {
e := w.writer.bundle.DB.Update(func(txn *badger.Txn) error {
for _, batch := range batchGroup {
for _, entry := range batch.entries {
var err error
if len(entry.UserMeta) == 0 {
entry.SetDelete()
}
err = txn.SetEntry(entry)
if err != nil {
return err
}
}
}
return nil
})
for _, batch := range batchGroup {
batch.err = e
batch.wg.Done()
}
}
type writeLockWorker struct {
batchCh chan *writeLockBatch
writer *dbWriter
}
func (w writeLockWorker) run() {
defer w.writer.wg.Done()
ls := w.writer.bundle.LockStore
var batches []*writeLockBatch
for {
for i := range batches {
batches[i] = nil
}
batches = batches[:0]
select {
case <-w.writer.closeCh:
return
case batch := <-w.batchCh:
batches = append(batches, batch)
}
chLen := len(w.batchCh)
for i := 0; i < chLen; i++ {
batches = append(batches, <-w.batchCh)
}
hint := new(lockstore.Hint)
var delCnt, insertCnt int
for _, batch := range batches {
for _, entry := range batch.entries {
switch entry.UserMeta[0] {
case mvcc.LockUserMetaDeleteByte:
delCnt++
// Ignore if the key doesn't exist
ls.DeleteWithHint(entry.Key.UserKey, hint)
default:
insertCnt++
ls.PutWithHint(entry.Key.UserKey, entry.Value, hint)
}
}
batch.wg.Done()
}
}
}
type dbWriter struct {
bundle *mvcc.DBBundle
dbCh chan<- *writeDBBatch
lockCh chan<- *writeLockBatch
wg sync.WaitGroup
closeCh chan struct{}
latestTS uint64
}
// NewDBWriter returns a new DBWriter.
func NewDBWriter(bundle *mvcc.DBBundle) mvcc.DBWriter {
return &dbWriter{
bundle: bundle,
closeCh: make(chan struct{}),
}
}
func (writer *dbWriter) Open() {
writer.wg.Add(2)
dbCh := make(chan *writeDBBatch, batchChanSize)
writer.dbCh = dbCh
go writeDBWorker{
batchCh: dbCh,
writer: writer,
}.run()
lockCh := make(chan *writeLockBatch, batchChanSize)
writer.lockCh = lockCh
go writeLockWorker{
batchCh: lockCh,
writer: writer,
}.run()
}
func (writer *dbWriter) Close() {
close(writer.closeCh)
writer.wg.Wait()
}
func (writer *dbWriter) Write(batch mvcc.WriteBatch) error {
wb := batch.(*writeBatch)
if len(wb.dbBatch.entries) > 0 {
wb.dbBatch.wg.Add(1)
writer.dbCh <- &wb.dbBatch
wb.dbBatch.wg.Wait()
err := wb.dbBatch.err
if err != nil {
return err
}
}
if len(wb.lockBatch.entries) > 0 {
// We must delete lock after commit succeed, or there will be inconsistency.
wb.lockBatch.wg.Add(1)
writer.lockCh <- &wb.lockBatch
wb.lockBatch.wg.Wait()
return wb.lockBatch.err
}
return nil
}
type writeBatch struct {
startTS uint64
commitTS uint64
dbBatch writeDBBatch
lockBatch writeLockBatch
}
func (wb *writeBatch) Prewrite(key []byte, lock *mvcc.Lock) {
wb.lockBatch.set(key, lock.MarshalBinary())
}
func (wb *writeBatch) Commit(key []byte, lock *mvcc.Lock) {
userMeta := mvcc.NewDBUserMeta(wb.startTS, wb.commitTS)
k := y.KeyWithTs(key, wb.commitTS)
if lock.Op != uint8(kvrpcpb.Op_Lock) {
wb.dbBatch.set(k, lock.Value, userMeta)
} else if bytes.Equal(key, lock.Primary) {
opLockKey := y.KeyWithTs(mvcc.EncodeExtraTxnStatusKey(key, wb.startTS), wb.startTS)
wb.dbBatch.set(opLockKey, nil, userMeta)
}
wb.lockBatch.delete(key)
}
func (wb *writeBatch) Rollback(key []byte, deleteLock bool) {
rollbackKey := y.KeyWithTs(mvcc.EncodeExtraTxnStatusKey(key, wb.startTS), wb.startTS)
userMeta := mvcc.NewDBUserMeta(wb.startTS, 0)
wb.dbBatch.set(rollbackKey, nil, userMeta)
if deleteLock {
wb.lockBatch.delete(key)
}
}
func (wb *writeBatch) PessimisticLock(key []byte, lock *mvcc.Lock) {
wb.lockBatch.set(key, lock.MarshalBinary())
}
func (wb *writeBatch) PessimisticRollback(key []byte) {
wb.lockBatch.delete(key)
}
func (writer *dbWriter) NewWriteBatch(startTS, commitTS uint64, ctx *kvrpcpb.Context) mvcc.WriteBatch {
if commitTS > 0 {
writer.updateLatestTS(commitTS)
} else {
writer.updateLatestTS(startTS)
}
return &writeBatch{
startTS: startTS,
commitTS: commitTS,
}
}
func (writer *dbWriter) getLatestTS() uint64 {
return atomic.LoadUint64(&writer.latestTS)
}
func (writer *dbWriter) updateLatestTS(ts uint64) {
latestTS := writer.getLatestTS()
if ts != math.MaxUint64 && ts > latestTS {
atomic.CompareAndSwapUint64(&writer.latestTS, latestTS, ts)
}
}
const delRangeBatchSize = 4096
func (writer *dbWriter) DeleteRange(startKey, endKey []byte, latchHandle mvcc.LatchHandle) error {
keys := make([]y.Key, 0, delRangeBatchSize)
txn := writer.bundle.DB.NewTransaction(false)
defer txn.Discard()
reader := dbreader.NewDBReader(startKey, endKey, txn)
keys = writer.collectRangeKeys(reader.GetIter(), startKey, endKey, keys)
reader.Close()
return writer.deleteKeysInBatch(latchHandle, keys, delRangeBatchSize)
}
func (writer *dbWriter) collectRangeKeys(it *badger.Iterator, startKey, endKey []byte, keys []y.Key) []y.Key {
if len(endKey) == 0 {
panic("invalid end key")
}
for it.Seek(startKey); it.Valid(); it.Next() {
item := it.Item()
key := item.KeyCopy(nil)
if exceedEndKey(key, endKey) {
break
}
keys = append(keys, y.KeyWithTs(key, item.Version()))
}
return keys
}
func (writer *dbWriter) deleteKeysInBatch(latchHandle mvcc.LatchHandle, keys []y.Key, batchSize int) error {
for len(keys) > 0 {
batchSize := mathutil.Min(len(keys), batchSize)
batchKeys := keys[:batchSize]
keys = keys[batchSize:]
hashVals := userKeysToHashVals(batchKeys...)
dbBatch := newWriteDBBatch()
for _, key := range batchKeys {
key.Version++
dbBatch.delete(key)
}
latchHandle.AcquireLatches(hashVals)
dbBatch.wg.Add(1)
writer.dbCh <- dbBatch
dbBatch.wg.Wait()
latchHandle.ReleaseLatches(hashVals)
if dbBatch.err != nil {
return dbBatch.err
}
}
return nil
}