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
|
package needle_map
/* CompactMap is an in-memory map of needle indeces, optimized for memory usage.
*
* It's implemented as a map of sorted indeces segments, which are in turn accessed through binary
* search. This guarantees a best-case scenario (ordered inserts/updates) of O(1) and a worst case
* scenario of O(log n) runtime, with memory usage unaffected by insert ordering.
*
* Note that even at O(log n), the clock time for both reads and writes is very low, so CompactMap
* will seldom bottleneck index operations.
*/
import (
"fmt"
"math"
"slices"
"sort"
"sync"
"github.com/seaweedfs/seaweedfs/weed/storage/types"
)
const (
MaxCompactKey = math.MaxUint16
SegmentChunkSize = 50000 // should be <= MaxCompactKey
)
type CompactKey uint16
type CompactOffset [types.OffsetSize]byte
type CompactNeedleValue struct {
key CompactKey
offset CompactOffset
size types.Size
}
type Chunk uint64
type CompactMapSegment struct {
list []CompactNeedleValue
chunk Chunk
firstKey CompactKey
lastKey CompactKey
}
type CompactMap struct {
sync.RWMutex
segments map[Chunk]*CompactMapSegment
}
func (ck CompactKey) Key(chunk Chunk) types.NeedleId {
return (types.NeedleId(SegmentChunkSize) * types.NeedleId(chunk)) + types.NeedleId(ck)
}
func OffsetToCompact(offset types.Offset) CompactOffset {
var co CompactOffset
types.OffsetToBytes(co[:], offset)
return co
}
func (co CompactOffset) Offset() types.Offset {
return types.BytesToOffset(co[:])
}
func (cnv CompactNeedleValue) NeedleValue(chunk Chunk) NeedleValue {
return NeedleValue{
Key: cnv.key.Key(chunk),
Offset: cnv.offset.Offset(),
Size: cnv.size,
}
}
func newCompactMapSegment(chunk Chunk) *CompactMapSegment {
return &CompactMapSegment{
list: []CompactNeedleValue{},
chunk: chunk,
firstKey: MaxCompactKey,
lastKey: 0,
}
}
func (cs *CompactMapSegment) len() int {
return len(cs.list)
}
func (cs *CompactMapSegment) cap() int {
return cap(cs.list)
}
func (cs *CompactMapSegment) compactKey(key types.NeedleId) CompactKey {
return CompactKey(key - (types.NeedleId(SegmentChunkSize) * types.NeedleId(cs.chunk)))
}
// bsearchKey returns the CompactNeedleValue index for a given ID key.
// If the key is not found, it returns the index where it should be inserted instead.
func (cs *CompactMapSegment) bsearchKey(key types.NeedleId) (int, bool) {
ck := cs.compactKey(key)
switch {
case len(cs.list) == 0:
return 0, false
case ck == cs.firstKey:
return 0, true
case ck <= cs.firstKey:
return 0, false
case ck == cs.lastKey:
return len(cs.list) - 1, true
case ck > cs.lastKey:
return len(cs.list), false
}
i := sort.Search(len(cs.list), func(i int) bool {
return cs.list[i].key >= ck
})
return i, cs.list[i].key == ck
}
// set inserts/updates a CompactNeedleValue.
// If the operation is an update, returns the overwritten value's previous offset and size.
func (cs *CompactMapSegment) set(key types.NeedleId, offset types.Offset, size types.Size) (oldOffset types.Offset, oldSize types.Size) {
i, found := cs.bsearchKey(key)
if found {
// update
o := cs.list[i].offset.Offset()
oldOffset.OffsetLower = o.OffsetLower
oldOffset.OffsetHigher = o.OffsetHigher
oldSize = cs.list[i].size
o.OffsetLower = offset.OffsetLower
o.OffsetHigher = offset.OffsetHigher
cs.list[i].offset = OffsetToCompact(o)
cs.list[i].size = size
return
}
// insert
if len(cs.list) >= SegmentChunkSize {
panic(fmt.Sprintf("attempted to write more than %d entries on CompactMapSegment %p!!!", SegmentChunkSize, cs))
}
if len(cs.list) == SegmentChunkSize-1 {
// if we max out our segment storage, pin its capacity to minimize memory usage
nl := make([]CompactNeedleValue, SegmentChunkSize, SegmentChunkSize)
copy(nl, cs.list[:i])
copy(nl[i+1:], cs.list[i:])
cs.list = nl
} else {
cs.list = append(cs.list, CompactNeedleValue{})
copy(cs.list[i+1:], cs.list[i:])
}
ck := cs.compactKey(key)
cs.list[i] = CompactNeedleValue{
key: ck,
offset: OffsetToCompact(offset),
size: size,
}
if ck < cs.firstKey {
cs.firstKey = ck
}
if ck > cs.lastKey {
cs.lastKey = ck
}
return
}
// get seeks a map entry by key. Returns an entry pointer, with a boolean specifiying if the entry was found.
func (cs *CompactMapSegment) get(key types.NeedleId) (*CompactNeedleValue, bool) {
if i, found := cs.bsearchKey(key); found {
return &cs.list[i], true
}
return nil, false
}
// delete deletes a map entry by key. Returns the entries' previous Size, if available.
func (cs *CompactMapSegment) delete(key types.NeedleId) types.Size {
if i, found := cs.bsearchKey(key); found {
if cs.list[i].size > 0 && cs.list[i].size.IsValid() {
ret := cs.list[i].size
cs.list[i].size = -cs.list[i].size
return ret
}
}
return types.Size(0)
}
func NewCompactMap() *CompactMap {
return &CompactMap{
segments: map[Chunk]*CompactMapSegment{},
}
}
func (cm *CompactMap) Len() int {
l := 0
for _, s := range cm.segments {
l += s.len()
}
return l
}
func (cm *CompactMap) Cap() int {
c := 0
for _, s := range cm.segments {
c += s.cap()
}
return c
}
func (cm *CompactMap) String() string {
if cm.Len() == 0 {
return "empty"
}
return fmt.Sprintf(
"%d/%d elements on %d segments, %.02f%% efficiency",
cm.Len(), cm.Cap(), len(cm.segments),
float64(100)*float64(cm.Len())/float64(cm.Cap()))
}
func (cm *CompactMap) segmentForKey(key types.NeedleId) *CompactMapSegment {
chunk := Chunk(key / SegmentChunkSize)
if cs, ok := cm.segments[chunk]; ok {
return cs
}
cs := newCompactMapSegment(chunk)
cm.segments[chunk] = cs
return cs
}
// Set inserts/updates a NeedleValue.
// If the operation is an update, returns the overwritten value's previous offset and size.
func (cm *CompactMap) Set(key types.NeedleId, offset types.Offset, size types.Size) (oldOffset types.Offset, oldSize types.Size) {
cm.RLock()
defer cm.RUnlock()
cs := cm.segmentForKey(key)
return cs.set(key, offset, size)
}
// Get seeks a map entry by key. Returns an entry pointer, with a boolean specifiying if the entry was found.
func (cm *CompactMap) Get(key types.NeedleId) (*NeedleValue, bool) {
cm.RLock()
defer cm.RUnlock()
cs := cm.segmentForKey(key)
if cnv, found := cs.get(key); found {
nv := cnv.NeedleValue(cs.chunk)
return &nv, true
}
return nil, false
}
// Delete deletes a map entry by key. Returns the entries' previous Size, if available.
func (cm *CompactMap) Delete(key types.NeedleId) types.Size {
cm.RLock()
defer cm.RUnlock()
cs := cm.segmentForKey(key)
return cs.delete(key)
}
// AscendingVisit runs a function on all entries, in ascending key order. Returns any errors hit while visiting.
func (cm *CompactMap) AscendingVisit(visit func(NeedleValue) error) error {
cm.RLock()
defer cm.RUnlock()
chunks := []Chunk{}
for c := range cm.segments {
chunks = append(chunks, c)
}
slices.Sort(chunks)
for _, c := range chunks {
cs := cm.segments[c]
for _, cnv := range cs.list {
nv := cnv.NeedleValue(cs.chunk)
if err := visit(nv); err != nil {
return err
}
}
}
return nil
}
|