forked from dgraph-io/badger
-
Notifications
You must be signed in to change notification settings - Fork 0
/
compaction.go
246 lines (215 loc) · 6.3 KB
/
compaction.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
/*
* Copyright 2017 Dgraph Labs, Inc. and Contributors
*
* 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,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package badger
import (
"bytes"
"fmt"
"log"
"math"
"sync"
"github.com/dgraph-io/badger/v3/table"
"github.com/dgraph-io/badger/v3/y"
)
type keyRange struct {
left []byte
right []byte
inf bool
}
func (r keyRange) isEmpty() bool {
return len(r.left) == 0 && len(r.right) == 0 && !r.inf
}
var infRange = keyRange{inf: true}
func (r keyRange) String() string {
return fmt.Sprintf("[left=%x, right=%x, inf=%v]", r.left, r.right, r.inf)
}
func (r keyRange) equals(dst keyRange) bool {
return bytes.Equal(r.left, dst.left) &&
bytes.Equal(r.right, dst.right) &&
r.inf == dst.inf
}
func (r *keyRange) extend(kr keyRange) {
// TODO(ibrahim): Is this needed?
if kr.isEmpty() {
return
}
if r.isEmpty() {
*r = kr
}
if len(r.left) == 0 || y.CompareKeys(kr.left, r.left) < 0 {
r.left = kr.left
}
if len(r.right) == 0 || y.CompareKeys(kr.right, r.right) > 0 {
r.right = kr.right
}
if kr.inf {
r.inf = true
}
}
func (r keyRange) overlapsWith(dst keyRange) bool {
// Empty keyRange always overlaps.
if r.isEmpty() {
return true
}
// TODO(ibrahim): Do you need this?
// Empty dst doesn't overlap with anything.
if dst.isEmpty() {
return false
}
if r.inf || dst.inf {
return true
}
// If my left is greater than dst right, we have no overlap.
if y.CompareKeys(r.left, dst.right) > 0 {
return false
}
// If my right is less than dst left, we have no overlap.
if y.CompareKeys(r.right, dst.left) < 0 {
return false
}
// We have overlap.
return true
}
// getKeyRange returns the smallest and the biggest in the list of tables.
// TODO(naman): Write a test for this. The smallest and the biggest should
// be the smallest of the leftmost table and the biggest of the right most table.
func getKeyRange(tables ...*table.Table) keyRange {
if len(tables) == 0 {
return keyRange{}
}
smallest := tables[0].Smallest()
biggest := tables[0].Biggest()
for i := 1; i < len(tables); i++ {
if y.CompareKeys(tables[i].Smallest(), smallest) < 0 {
smallest = tables[i].Smallest()
}
if y.CompareKeys(tables[i].Biggest(), biggest) > 0 {
biggest = tables[i].Biggest()
}
}
// We pick all the versions of the smallest and the biggest key. Note that version zero would
// be the rightmost key, considering versions are default sorted in descending order.
return keyRange{
left: y.KeyWithTs(y.ParseKey(smallest), math.MaxUint64),
right: y.KeyWithTs(y.ParseKey(biggest), 0),
}
}
type levelCompactStatus struct {
ranges []keyRange
delSize int64
}
func (lcs *levelCompactStatus) debug() string {
var b bytes.Buffer
for _, r := range lcs.ranges {
b.WriteString(r.String())
}
return b.String()
}
func (lcs *levelCompactStatus) overlapsWith(dst keyRange) bool {
for _, r := range lcs.ranges {
if r.overlapsWith(dst) {
return true
}
}
return false
}
func (lcs *levelCompactStatus) remove(dst keyRange) bool {
final := lcs.ranges[:0]
var found bool
for _, r := range lcs.ranges {
if !r.equals(dst) {
final = append(final, r)
} else {
found = true
}
}
lcs.ranges = final
return found
}
type compactStatus struct {
sync.RWMutex
levels []*levelCompactStatus
tables map[uint64]struct{}
}
func (cs *compactStatus) overlapsWith(level int, this keyRange) bool {
cs.RLock()
defer cs.RUnlock()
thisLevel := cs.levels[level]
return thisLevel.overlapsWith(this)
}
func (cs *compactStatus) delSize(l int) int64 {
cs.RLock()
defer cs.RUnlock()
return cs.levels[l].delSize
}
type thisAndNextLevelRLocked struct{}
// compareAndAdd will check whether we can run this compactDef. That it doesn't overlap with any
// other running compaction. If it can be run, it would store this run in the compactStatus state.
func (cs *compactStatus) compareAndAdd(_ thisAndNextLevelRLocked, cd compactDef) bool {
cs.Lock()
defer cs.Unlock()
tl := cd.thisLevel.level
y.AssertTruef(tl < len(cs.levels), "Got level %d. Max levels: %d", tl, len(cs.levels))
thisLevel := cs.levels[cd.thisLevel.level]
nextLevel := cs.levels[cd.nextLevel.level]
if thisLevel.overlapsWith(cd.thisRange) {
return false
}
if nextLevel.overlapsWith(cd.nextRange) {
return false
}
// Check whether this level really needs compaction or not. Otherwise, we'll end up
// running parallel compactions for the same level.
// Update: We should not be checking size here. Compaction priority already did the size checks.
// Here we should just be executing the wish of others.
thisLevel.ranges = append(thisLevel.ranges, cd.thisRange)
nextLevel.ranges = append(nextLevel.ranges, cd.nextRange)
thisLevel.delSize += cd.thisSize
for _, t := range append(cd.top, cd.bot...) {
cs.tables[t.ID()] = struct{}{}
}
return true
}
func (cs *compactStatus) delete(cd compactDef) {
cs.Lock()
defer cs.Unlock()
tl := cd.thisLevel.level
y.AssertTruef(tl < len(cs.levels), "Got level %d. Max levels: %d", tl, len(cs.levels))
thisLevel := cs.levels[cd.thisLevel.level]
nextLevel := cs.levels[cd.nextLevel.level]
thisLevel.delSize -= cd.thisSize
found := thisLevel.remove(cd.thisRange)
// The following check makes sense only if we're compacting more than one
// table. In case of the max level, we might rewrite a single table to
// remove stale data.
if cd.thisLevel != cd.nextLevel && !cd.nextRange.isEmpty() {
found = nextLevel.remove(cd.nextRange) && found
}
if !found {
this := cd.thisRange
next := cd.nextRange
fmt.Printf("Looking for: %s in this level %d.\n", this, tl)
fmt.Printf("This Level:\n%s\n", thisLevel.debug())
fmt.Println()
fmt.Printf("Looking for: %s in next level %d.\n", next, cd.nextLevel.level)
fmt.Printf("Next Level:\n%s\n", nextLevel.debug())
log.Fatal("keyRange not found")
}
for _, t := range append(cd.top, cd.bot...) {
_, ok := cs.tables[t.ID()]
y.AssertTrue(ok)
delete(cs.tables, t.ID())
}
}