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
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
|
/*
{s.get_tombstone_count()} -> std::convertible_to<size_t>;
* include/shard/WSS.h
*
* Copyright (C) 2023 Dong Xie <dongx@psu.edu>
*
* All rights reserved. Published under the Modified BSD License.
*
*/
#pragma once
#include <vector>
#include <cassert>
#include <queue>
#include <memory>
#include <concepts>
#include "ds/PriorityQueue.h"
#include "util/Cursor.h"
#include "ds/Alias.h"
#include "ds/BloomFilter.h"
#include "util/bf_config.h"
#include "framework/MutableBuffer.h"
#include "framework/RecordInterface.h"
#include "framework/ShardInterface.h"
#include "framework/QueryInterface.h"
namespace de {
thread_local size_t wss_cancelations = 0;
template <WeightedRecordInterface R>
struct wss_query_parms {
size_t sample_size;
gsl_rng *rng;
};
template <WeightedRecordInterface R, bool Rejection>
class WSSQuery;
template <WeightedRecordInterface R>
struct WSSState {
decltype(R::weight) total_weight;
size_t sample_size;
WSSState() {
total_weight = 0;
}
};
template <WeightedRecordInterface R>
struct WSSBufferState {
size_t cutoff;
size_t sample_size;
Alias* alias;
decltype(R::weight) max_weight;
decltype(R::weight) total_weight;
~WSSBufferState() {
delete alias;
}
};
template <WeightedRecordInterface R>
class WSS {
private:
typedef decltype(R::key) K;
typedef decltype(R::value) V;
typedef decltype(R::weight) W;
public:
// FIXME: there has to be a better way to do this
friend class WSSQuery<R, true>;
friend class WSSQuery<R, false>;
WSS(MutableBuffer<R>* buffer)
: m_reccnt(0), m_tombstone_cnt(0), m_total_weight(0), m_alias(nullptr), m_bf(nullptr) {
size_t alloc_size = (buffer->get_record_count() * sizeof(Wrapped<R>)) + (CACHELINE_SIZE - (buffer->get_record_count() * sizeof(Wrapped<R>)) % CACHELINE_SIZE);
assert(alloc_size % CACHELINE_SIZE == 0);
m_data = (Wrapped<R>*)std::aligned_alloc(CACHELINE_SIZE, alloc_size);
m_bf = new BloomFilter<K>(BF_FPR, buffer->get_tombstone_count(), BF_HASH_FUNCS);
size_t offset = 0;
m_reccnt = 0;
auto base = buffer->get_data();
auto stop = base + buffer->get_record_count();
std::sort(base, stop, std::less<Wrapped<R>>());
std::vector<W> weights;
while (base < stop) {
if (!(base->is_tombstone()) && (base + 1) < stop) {
if (base->rec == (base + 1)->rec && (base + 1)->is_tombstone()) {
base += 2;
wss_cancelations++;
continue;
}
} else if (base->is_deleted()) {
base += 1;
continue;
}
base->header &= 1;
m_data[m_reccnt++] = *base;
m_total_weight+= base->rec.weight;
weights.push_back(base->rec.weight);
if (m_bf && base->is_tombstone()) {
m_tombstone_cnt++;
m_bf->insert(base->rec.key);
}
base++;
}
if (m_reccnt > 0) {
build_alias_structure(weights);
}
}
WSS(WSS** shards, size_t len)
: m_reccnt(0), m_tombstone_cnt(0), m_total_weight(0), m_alias(nullptr), m_bf(nullptr) {
std::vector<Cursor<Wrapped<R>>> cursors;
cursors.reserve(len);
PriorityQueue<Wrapped<R>> pq(len);
size_t attemp_reccnt = 0;
size_t tombstone_count = 0;
for (size_t i = 0; i < len; ++i) {
if (shards[i]) {
auto base = shards[i]->get_data();
cursors.emplace_back(Cursor{base, base + shards[i]->get_record_count(), 0, shards[i]->get_record_count()});
attemp_reccnt += shards[i]->get_record_count();
tombstone_count += shards[i]->get_tombstone_count();
pq.push(cursors[i].ptr, i);
} else {
cursors.emplace_back(Cursor<Wrapped<R>>{nullptr, nullptr, 0, 0});
}
}
m_bf = new BloomFilter<K>(BF_FPR, tombstone_count, BF_HASH_FUNCS);
size_t alloc_size = (attemp_reccnt * sizeof(Wrapped<R>)) + (CACHELINE_SIZE - (attemp_reccnt * sizeof(Wrapped<R>)) % CACHELINE_SIZE);
assert(alloc_size % CACHELINE_SIZE == 0);
m_data = (Wrapped<R>*)std::aligned_alloc(CACHELINE_SIZE, alloc_size);
std::vector<W> weights;
while (pq.size()) {
auto now = pq.peek();
auto next = pq.size() > 1 ? pq.peek(1) : queue_record<Wrapped<R>>{nullptr, 0};
if (!now.data->is_tombstone() && next.data != nullptr &&
now.data->rec == next.data->rec && next.data->is_tombstone()) {
pq.pop(); pq.pop();
auto& cursor1 = cursors[now.version];
auto& cursor2 = cursors[next.version];
if (advance_cursor<Wrapped<R>>(cursor1)) pq.push(cursor1.ptr, now.version);
if (advance_cursor<Wrapped<R>>(cursor2)) pq.push(cursor2.ptr, next.version);
} else {
auto& cursor = cursors[now.version];
if (!cursor.ptr->is_deleted()) {
m_data[m_reccnt++] = *cursor.ptr;
m_total_weight += cursor.ptr->rec.weight;
weights.push_back(cursor.ptr->rec.weight);
if (m_bf && cursor.ptr->is_tombstone()) {
++m_tombstone_cnt;
if (m_bf) m_bf->insert(cursor.ptr->rec.key);
}
}
pq.pop();
if (advance_cursor<Wrapped<R>>(cursor)) pq.push(cursor.ptr, now.version);
}
}
if (m_reccnt > 0) {
build_alias_structure(weights);
}
}
~WSS() {
if (m_data) free(m_data);
if (m_alias) delete m_alias;
if (m_bf) delete m_bf;
}
Wrapped<R> *point_lookup(const R &rec, bool filter=false) {
if (filter && !m_bf->lookup(rec.key)) {
return nullptr;
}
size_t idx = get_lower_bound(rec.key);
if (idx >= m_reccnt) {
return nullptr;
}
while (idx < m_reccnt && m_data[idx].rec < rec) ++idx;
if (m_data[idx].rec == rec) {
return m_data + idx;
}
return nullptr;
}
Wrapped<R>* get_data() const {
return m_data;
}
size_t get_record_count() const {
return m_reccnt;
}
size_t get_tombstone_count() const {
return m_tombstone_cnt;
}
const Wrapped<R>* get_record_at(size_t idx) const {
if (idx >= m_reccnt) return nullptr;
return m_data + idx;
}
size_t get_memory_usage() {
return 0;
}
private:
size_t get_lower_bound(const K& key) const {
size_t min = 0;
size_t max = m_reccnt - 1;
const char * record_key;
while (min < max) {
size_t mid = (min + max) / 2;
if (key > m_data[mid].rec.key) {
min = mid + 1;
} else {
max = mid;
}
}
return min;
}
void build_alias_structure(std::vector<W> &weights) {
// normalize the weights vector
std::vector<double> norm_weights(weights.size());
for (size_t i=0; i<weights.size(); i++) {
norm_weights[i] = (double) weights[i] / (double) m_total_weight;
}
// build the alias structure
m_alias = new Alias(norm_weights);
}
Wrapped<R>* m_data;
Alias *m_alias;
W m_total_weight;
size_t m_reccnt;
size_t m_tombstone_cnt;
size_t m_group_size;
BloomFilter<K> *m_bf;
};
template <WeightedRecordInterface R, bool Rejection=true>
class WSSQuery {
public:
static void *get_query_state(WSS<R> *wss, void *parms) {
auto res = new WSSState<R>();
return res;
}
static void* get_buffer_query_state(MutableBuffer<R> *buffer, void *parms) {
WSSBufferState<R> *state = new WSSBufferState<R>();
auto parameters = (wss_query_parms<R>*) parms;
if constexpr (Rejection) {
state->cutoff = buffer->get_record_count() - 1;
state->max_weight = buffer->get_max_weight();
return state;
}
std::vector<double> weights;
state->cutoff = buffer->get_record_count() - 1;
double total_weight = 0.0;
for (size_t i = 0; i <= state->cutoff; i++) {
auto rec = buffer->get_data() + i;
weights.push_back(rec->rec.weight);
total_weight += rec->rec.weight;
}
for (size_t i = 0; i < weights.size(); i++) {
weights[i] = weights[i] / total_weight;
}
state->alias = new Alias(weights);
return state;
}
static void process_query_states(void *query_parms, std::vector<void*> shard_states, void *buff_state) {
auto p = (wss_query_parms<R> *) query_parms;
auto bs = (WSSBufferState<R> *) buff_state;
std::vector<size_t> shard_sample_sizes = {0};
size_t buffer_sz = 0;
std::vector<decltype(R::weight)> weights;
weights.push_back(bs->total_weight);
decltype(R::weight) total_weight;
for (auto &s : shard_states) {
auto state = (WSSState<R> *) s;
total_weight += state->total_weight;
weights.push_back(state->total_weight);
}
std::vector<double> normalized_weights;
for (auto w : weights) {
normalized_weights.push_back((double) w / (double) total_weight);
}
auto shard_alias = Alias(normalized_weights);
for (size_t i=0; i<p->sample_size; i++) {
auto idx = shard_alias.get(p->rng);
if (idx == 0) {
buffer_sz++;
} else {
shard_sample_sizes[idx - 1]++;
}
}
bs->sample_size = buffer_sz;
size_t i=1;
for (auto &s : shard_states) {
auto state = (WSSState<R> *) s;
state->sample_size = shard_sample_sizes[i++];
}
}
static std::vector<Wrapped<R>> query(WSS<R> *wss, void *q_state, void *parms) {
auto sample_size = ((WSSState<R> *) q_state)->sample_size;
auto rng = ((wss_query_parms<R> *) parms)->rng;
auto state = (WSSState<R> *) q_state;
std::vector<Wrapped<R>> result_set;
if (sample_size == 0) {
return result_set;
}
size_t attempts = 0;
do {
attempts++;
size_t idx = wss->m_alias->get(rng);
result_set.emplace_back(*wss->get_record_at(idx));
} while (attempts < sample_size);
return result_set;
}
static std::vector<Wrapped<R>> buffer_query(MutableBuffer<R> *buffer, void *state, void *parms) {
auto st = (WSSBufferState<R> *) state;
auto p = (wss_query_parms<R> *) parms;
std::vector<Wrapped<R>> result;
result.reserve(st->sample_size);
if constexpr (Rejection) {
for (size_t i=0; i<st->sample_size; i++) {
auto idx = gsl_rng_uniform_int(p->rng, st->cutoff);
auto rec = buffer->get_data() + idx;
auto test = gsl_rng_uniform(p->rng) * st->max_weight;
if (test <= rec->rec.weight) {
result.emplace_back(*rec);
}
}
return result;
}
for (size_t i=0; i<st->sample_size; i++) {
auto idx = st->alias->get(p->rng);
result.emplace_back(*(buffer->get_data() + idx));
}
return result;
}
static std::vector<R> merge(std::vector<std::vector<R>> &results) {
std::vector<R> output;
for (size_t i=0; i<results.size(); i++) {
for (size_t j=0; j<results[i].size(); j++) {
output.emplace_back(results[i][j]);
}
}
return output;
}
static void delete_query_state(void *state) {
auto s = (WSSState<R> *) state;
delete s;
}
static void delete_buffer_query_state(void *state) {
auto s = (WSSBufferState<R> *) state;
delete s;
}
};
}
|