/
iblt.go
413 lines (377 loc) · 7.96 KB
/
iblt.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
package iblt
// Invertible Bloom Lookup Table from
// What’s the Difference?
// Efficient Set Reconciliation without Prior Context
// David Eppstein1 Michael T. Goodrich1 Frank Uyeda2 George Varghese
// https://www.ics.uci.edu/~eppstein/pubs/EppGooUye-SIGCOMM-11.pdf
// IBFL with N cells (N>=50) and K=4 can safely recover diffs of size at least N/2.
// For large N the space overhead is less than 1.3 (so we can decode diffs of size of n * 0.77).
import (
"bytes"
"encoding/binary"
"encoding/gob"
"errors"
"log"
"github.com/dchest/siphash"
)
// Filter is totally NOT thread-safe!
type Filter struct {
mask uint64
keySums []uint64
valueSums []*bts
counts []int
seen bitset
buf []byte
shift uint16
idx []int
}
func (f *Filter) Clone() *Filter {
r := &Filter{
mask: f.mask,
keySums: make([]uint64, f.N()),
valueSums: make([]*bts, 0, f.N()),
counts: make([]int, f.N()),
seen: make(bitset, len(f.seen)),
shift: f.shift,
idx: make([]int, len(f.idx)),
}
copy(r.keySums, f.keySums)
for _, v := range f.keySums {
r.keySums = append(r.keySums, v)
}
for _, v := range f.valueSums {
b := make([]byte, len(v.b))
copy(b, v.b)
r.valueSums = append(r.valueSums, &bts{b})
}
copy(r.counts, f.counts)
return r
}
type serializableFilter struct {
KeySums []uint64
ValueSums [][]byte
Counts []int
K int
}
func (f Filter) MarshalBinary() (data []byte, err error) {
s := serializableFilter{
KeySums: f.keySums,
Counts: f.counts,
ValueSums: make([][]byte, len(f.valueSums)),
K: len(f.idx),
}
for k, v := range f.valueSums {
s.ValueSums[k] = v.b
}
buf := &bytes.Buffer{}
err = gob.NewEncoder(buf).Encode(s)
if err != nil {
return nil, err
}
return buf.Bytes(), nil
}
func (f *Filter) UnmarshalBinary(data []byte) error {
var s serializableFilter
err := gob.NewDecoder(bytes.NewReader(data)).Decode(&s)
if err != nil {
return err
}
*f = *New(s.K, len(s.Counts))
f.keySums = s.KeySums
f.counts = s.Counts
for k, v := range s.ValueSums {
f.valueSums[k].b = v
}
return nil
}
// New constructs a new Filter.
func New(k, l int) *Filter {
if k >= 10 || k < 1 {
panic("k should be between 1 and 10")
}
if l/k < 2 {
panic("l should be at least 2*k")
}
if l&(l-1) != 0 {
panic("l should be a power of two")
}
var shift uint16
for ll := l; ll != 0; ll >>= 1 {
shift++
}
values := make([]*bts, l)
for i := range values {
values[i] = &bts{}
}
return &Filter{
mask: uint64(l - 1),
keySums: make([]uint64, l),
valueSums: values,
counts: make([]int, l),
seen: newBitSet(l),
shift: shift,
idx: make([]int, k),
}
}
func (f Filter) K() int {
return len(f.idx)
}
func (f Filter) N() int {
return len(f.counts)
}
func xor(a, b []byte) []byte {
if len(b) > len(a) {
return _xor(b, a)
}
return _xor(a, b)
}
func (f *Filter) getIdxHash(b []byte) ([]int, uint64) {
hash := hash(b)
return f.getIdx(hash), hash
}
func (f *Filter) getIdx(hash uint64) []int {
f.seen.ClearAll()
v := hash
bits := uint16(64)
for k := range f.idx {
for {
if bits < f.shift {
v = xorShiftStarRound(&hash)
bits = 64
}
pos := int(v & f.mask)
v >>= f.shift
bits -= f.shift
if f.seen.Test(pos) {
continue
}
f.seen.Set(pos)
f.idx[k] = pos
break
}
}
return f.idx
}
func (f *Filter) Add(b []byte) {
idx, hash := f.getIdxHash(b)
enc := f.encode(b)
for _, k := range idx {
f.counts[k] = f.counts[k] + 1
f.keySums[k] = f.keySums[k] ^ hash
f.valueSums[k].xorInPlace(enc)
}
}
func (f *Filter) Remove(b []byte) {
idx, hash := f.getIdxHash(b)
enc := f.encode(b)
for _, k := range idx {
f.counts[k] = f.counts[k] - 1
f.keySums[k] = f.keySums[k] ^ hash
f.valueSums[k].xorInPlace(enc)
}
// TODO: consider cleaning up pure cells:
// the value could get long if a very long values were inserted and then removed.
}
type Diff struct {
Added [][]byte
Removed [][]byte
}
func (f Filter) Diff(other Filter, diff *Filter) error {
if f.shift != other.shift || f.shift != diff.shift {
return errors.New("sizes should match")
}
if len(f.idx) != len(other.idx) || len(f.idx) != len(diff.idx) {
return errors.New("ks should match")
}
for k := range f.counts {
diff.counts[k] = f.counts[k] - other.counts[k]
diff.keySums[k] = f.keySums[k] ^ other.keySums[k]
diff.valueSums[k] = &bts{xor(f.valueSums[k].b, other.valueSums[k].b)}
}
return nil
}
// Inplace
func (f *Filter) Sub(other Filter) error {
if f.shift != other.shift {
return errors.New("sizes should match")
}
if len(f.idx) != len(other.idx) {
return errors.New("ks should match")
}
for k := range f.counts {
f.counts[k] = f.counts[k] - other.counts[k]
f.keySums[k] = f.keySums[k] ^ other.keySums[k]
f.valueSums[k].xorInPlace(other.valueSums[k].b)
}
return nil
}
// Decode is distructive!
// One can apply the diff produced (even in case of error) to restore the previous state.
func (f *Filter) Decode() (*Diff, error) {
pure := make([]int, len(f.counts))
numZ := 0
diff := &Diff{}
lp := 0
for k := range f.counts {
c := f.counts[k]
switch c {
case 0:
if f.keySums[k] == 0 {
numZ++
}
case -1, 1:
_, err := f.decode(f.valueSums[k].b)
if err != nil {
continue
}
pure[lp] = k
lp++
}
}
head := 0
tail := lp - 1
for lp > 0 {
// Deque pop
lp--
pos := pure[head]
c := f.counts[pos]
head++
if head == len(f.counts) {
head = 0
}
if c != 1 && c != -1 {
continue
}
dec, err := f.decode(f.valueSums[pos].b)
if err != nil {
continue
}
h := hash(dec)
if h == 0 || h != f.keySums[pos] {
continue
}
if c == 1 {
diff.Added = append(diff.Added, dec)
} else {
diff.Removed = append(diff.Removed, dec)
}
idx := f.getIdx(h)
val := f.valueSums[pos].b
numZ++
for _, k := range idx {
if k != pos {
if f.keySums[k] == 0 && f.counts[k] == 0 {
continue
}
f.keySums[k] = f.keySums[k] ^ h
f.valueSums[k].xorInPlace(val)
f.counts[k] = f.counts[k] - c
c := f.counts[k]
switch c {
case 0:
if f.keySums[k] == 0 {
numZ++
}
case -1, 1:
_, err := f.decode(f.valueSums[k].b)
if err != nil {
continue
}
// Deque push
lp++
tail++
if tail == len(f.counts) {
tail = 0
}
pure[tail] = k
}
}
}
f.keySums[pos] = 0
f.counts[pos] = 0
}
var err error
if numZ != len(f.counts) {
err = errors.New("failed to decode")
}
return diff, err
}
func _xor(a, b []byte) []byte {
// len(a) >= len(b)
// TODO: use append to reduce allocations?
r := make([]byte, len(a))
for i, v := range b {
r[i] = v ^ a[i]
}
copy(r[len(b):], a[len(b):])
return r
}
type bts struct {
b []byte
}
func (b *bts) xorInPlace(a []byte) {
if len(b.b) < len(a) {
b.b = _xor(a, b.b)
return
}
for i, v := range a {
b.b[i] = b.b[i] ^ v
}
}
// Self-delimited encoding.
func (f *Filter) encode(b []byte) []byte {
if len(b) >= 1<<16 {
log.Panicln("len(b) is too large", len(b))
}
l := len(b) + 2
if len(f.buf) < l {
f.buf = make([]byte, l)
}
binary.LittleEndian.PutUint16(f.buf, uint16(len(b)))
copy(f.buf[2:], b)
return f.buf[:l]
}
func (f *Filter) decode(b []byte) ([]byte, error) {
if len(b) == 0 {
return nil, nil
}
if len(b) < 2 {
return nil, errors.New("bad length")
}
l16 := binary.LittleEndian.Uint16(b)
l := int(l16)
if l+2 > len(b) {
return nil, errors.New("too short")
}
return b[2 : l+2], nil
}
type bitset []uint64
func (b bitset) Set(i int) {
b[i>>6] |= 1 << (uint32(i) & 63)
}
func (b bitset) Clear(i int) {
b[i>>6] &= ^(1 << (uint32(i) & 63))
}
func (b bitset) Test(i int) bool {
return (b[i>>6] & (1 << (uint32(i) & 63))) != 0
}
func (b bitset) ClearAll() {
for i := range b {
b[i] = 0
}
}
func newBitSet(l int) bitset {
return make([]uint64, (l+63)>>6)
}
func hash(v []byte) uint64 {
return siphash.Hash(2, 57, v)
}
func xorShiftStarRound(x *uint64) uint64 {
if *x == 0 {
*x = 1
}
*x ^= (*x >> 12)
*x ^= (*x << 25)
*x ^= (*x >> 27)
return *x * 2685821657736338717
}