forked from Disposrestfully/DB
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathblock_based_table_builder.cc
2148 lines (1955 loc) · 81.7 KB
/
block_based_table_builder.cc
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
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
// Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
// This source code is licensed under both the GPLv2 (found in the
// COPYING file in the root directory) and Apache 2.0 License
// (found in the LICENSE.Apache file in the root directory).
//
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file. See the AUTHORS file for names of contributors.
#include "table/block_based/block_based_table_builder.h"
#include <atomic>
#include <cassert>
#include <cstdio>
#include <list>
#include <map>
#include <memory>
#include <numeric>
#include <string>
#include <unordered_map>
#include <utility>
#include "block_cache.h"
#include "cache/cache_entry_roles.h"
#include "cache/cache_helpers.h"
#include "cache/cache_key.h"
#include "cache/cache_reservation_manager.h"
#include "db/dbformat.h"
#include "index_builder.h"
#include "logging/logging.h"
#include "memory/memory_allocator_impl.h"
#include "rocksdb/cache.h"
#include "rocksdb/comparator.h"
#include "rocksdb/env.h"
#include "rocksdb/filter_policy.h"
#include "rocksdb/flush_block_policy.h"
#include "rocksdb/merge_operator.h"
#include "rocksdb/table.h"
#include "rocksdb/types.h"
#include "table/block_based/block.h"
#include "table/block_based/block_based_table_factory.h"
#include "table/block_based/block_based_table_reader.h"
#include "table/block_based/block_builder.h"
#include "table/block_based/filter_block.h"
#include "table/block_based/filter_policy_internal.h"
#include "table/block_based/full_filter_block.h"
#include "table/block_based/partitioned_filter_block.h"
#include "table/format.h"
#include "table/meta_blocks.h"
#include "table/table_builder.h"
#include "util/coding.h"
#include "util/compression.h"
#include "util/stop_watch.h"
#include "util/string_util.h"
#include "util/work_queue.h"
namespace ROCKSDB_NAMESPACE {
extern const std::string kHashIndexPrefixesBlock;
extern const std::string kHashIndexPrefixesMetadataBlock;
// Without anonymous namespace here, we fail the warning -Wmissing-prototypes
namespace {
constexpr size_t kBlockTrailerSize = BlockBasedTable::kBlockTrailerSize;
// Create a filter block builder based on its type.
FilterBlockBuilder* CreateFilterBlockBuilder(
const ImmutableCFOptions& /*opt*/, const MutableCFOptions& mopt,
const FilterBuildingContext& context,
const bool use_delta_encoding_for_index_values,
PartitionedIndexBuilder* const p_index_builder, size_t ts_sz,
const bool persist_user_defined_timestamps) {
const BlockBasedTableOptions& table_opt = context.table_options;
assert(table_opt.filter_policy); // precondition
FilterBitsBuilder* filter_bits_builder =
BloomFilterPolicy::GetBuilderFromContext(context);
if (filter_bits_builder == nullptr) {
return nullptr;
} else {
if (table_opt.partition_filters) {
assert(p_index_builder != nullptr);
// Since after partition cut request from filter builder it takes time
// until index builder actully cuts the partition, until the end of a
// data block potentially with many keys, we take the lower bound as
// partition size.
assert(table_opt.block_size_deviation <= 100);
auto partition_size =
static_cast<uint32_t>(((table_opt.metadata_block_size *
(100 - table_opt.block_size_deviation)) +
99) /
100);
partition_size = std::max(partition_size, static_cast<uint32_t>(1));
return new PartitionedFilterBlockBuilder(
mopt.prefix_extractor.get(), table_opt.whole_key_filtering,
filter_bits_builder, table_opt.index_block_restart_interval,
use_delta_encoding_for_index_values, p_index_builder, partition_size,
ts_sz, persist_user_defined_timestamps);
} else {
return new FullFilterBlockBuilder(mopt.prefix_extractor.get(),
table_opt.whole_key_filtering,
filter_bits_builder);
}
}
}
bool GoodCompressionRatio(size_t compressed_size, size_t uncomp_size,
int max_compressed_bytes_per_kb) {
// For efficiency, avoid floating point and division
return compressed_size <=
(static_cast<uint64_t>(max_compressed_bytes_per_kb) * uncomp_size) >>
10;
}
} // namespace
// format_version is the block format as defined in include/rocksdb/table.h
Slice CompressBlock(const Slice& uncompressed_data, const CompressionInfo& info,
CompressionType* type, uint32_t format_version,
bool allow_sample, std::string* compressed_output,
std::string* sampled_output_fast,
std::string* sampled_output_slow) {
assert(type);
assert(compressed_output);
assert(compressed_output->empty());
// If requested, we sample one in every N block with a
// fast and slow compression algorithm and report the stats.
// The users can use these stats to decide if it is worthwhile
// enabling compression and they also get a hint about which
// compression algorithm wil be beneficial.
if (allow_sample && info.SampleForCompression() &&
Random::GetTLSInstance()->OneIn(
static_cast<int>(info.SampleForCompression()))) {
// Sampling with a fast compression algorithm
if (sampled_output_fast && (LZ4_Supported() || Snappy_Supported())) {
CompressionType c =
LZ4_Supported() ? kLZ4Compression : kSnappyCompression;
CompressionOptions options;
CompressionContext context(c, options);
CompressionInfo info_tmp(options, context,
CompressionDict::GetEmptyDict(), c,
info.SampleForCompression());
CompressData(uncompressed_data, info_tmp,
GetCompressFormatForVersion(format_version),
sampled_output_fast);
}
// Sampling with a slow but high-compression algorithm
if (sampled_output_slow && (ZSTD_Supported() || Zlib_Supported())) {
CompressionType c = ZSTD_Supported() ? kZSTD : kZlibCompression;
CompressionOptions options;
CompressionContext context(c, options);
CompressionInfo info_tmp(options, context,
CompressionDict::GetEmptyDict(), c,
info.SampleForCompression());
CompressData(uncompressed_data, info_tmp,
GetCompressFormatForVersion(format_version),
sampled_output_slow);
}
}
int max_compressed_bytes_per_kb = info.options().max_compressed_bytes_per_kb;
if (info.type() == kNoCompression || max_compressed_bytes_per_kb <= 0) {
*type = kNoCompression;
return uncompressed_data;
}
// Actually compress the data; if the compression method is not supported,
// or the compression fails etc., just fall back to uncompressed
if (!CompressData(uncompressed_data, info,
GetCompressFormatForVersion(format_version),
compressed_output)) {
*type = kNoCompression;
return uncompressed_data;
}
// Check the compression ratio; if it's not good enough, just fall back to
// uncompressed
if (!GoodCompressionRatio(compressed_output->size(), uncompressed_data.size(),
max_compressed_bytes_per_kb)) {
*type = kNoCompression;
return uncompressed_data;
}
*type = info.type();
return *compressed_output;
}
// kBlockBasedTableMagicNumber was picked by running
// echo rocksdb.table.block_based | sha1sum
// and taking the leading 64 bits.
// Please note that kBlockBasedTableMagicNumber may also be accessed by other
// .cc files
// for that reason we declare it extern in the header but to get the space
// allocated
// it must be not extern in one place.
const uint64_t kBlockBasedTableMagicNumber = 0x88e241b785f4cff7ull;
// We also support reading and writing legacy block based table format (for
// backwards compatibility)
const uint64_t kLegacyBlockBasedTableMagicNumber = 0xdb4775248b80fb57ull;
// A collector that collects properties of interest to block-based table.
// For now this class looks heavy-weight since we only write one additional
// property.
// But in the foreseeable future, we will add more and more properties that are
// specific to block-based table.
class BlockBasedTableBuilder::BlockBasedTablePropertiesCollector
: public InternalTblPropColl {
public:
explicit BlockBasedTablePropertiesCollector(
BlockBasedTableOptions::IndexType index_type, bool whole_key_filtering,
bool prefix_filtering)
: index_type_(index_type),
whole_key_filtering_(whole_key_filtering),
prefix_filtering_(prefix_filtering) {}
Status InternalAdd(const Slice& /*key*/, const Slice& /*value*/,
uint64_t /*file_size*/) override {
// Intentionally left blank. Have no interest in collecting stats for
// individual key/value pairs.
return Status::OK();
}
void BlockAdd(uint64_t /* block_uncomp_bytes */,
uint64_t /* block_compressed_bytes_fast */,
uint64_t /* block_compressed_bytes_slow */) override {
// Intentionally left blank. No interest in collecting stats for
// blocks.
}
Status Finish(UserCollectedProperties* properties) override {
std::string val;
PutFixed32(&val, static_cast<uint32_t>(index_type_));
properties->insert({BlockBasedTablePropertyNames::kIndexType, val});
properties->insert({BlockBasedTablePropertyNames::kWholeKeyFiltering,
whole_key_filtering_ ? kPropTrue : kPropFalse});
properties->insert({BlockBasedTablePropertyNames::kPrefixFiltering,
prefix_filtering_ ? kPropTrue : kPropFalse});
return Status::OK();
}
// The name of the properties collector can be used for debugging purpose.
const char* Name() const override {
return "BlockBasedTablePropertiesCollector";
}
UserCollectedProperties GetReadableProperties() const override {
// Intentionally left blank.
return UserCollectedProperties();
}
private:
BlockBasedTableOptions::IndexType index_type_;
bool whole_key_filtering_;
bool prefix_filtering_;
};
struct BlockBasedTableBuilder::Rep {
const ImmutableOptions ioptions;
// BEGIN from MutableCFOptions
std::shared_ptr<const SliceTransform> prefix_extractor;
// END from MutableCFOptions
const WriteOptions write_options;
const BlockBasedTableOptions table_options;
const InternalKeyComparator& internal_comparator;
// Size in bytes for the user-defined timestamps.
size_t ts_sz;
// When `ts_sz` > 0 and this flag is false, the user-defined timestamp in the
// user key will be stripped when creating the block based table. This
// stripping happens for all user keys, including the keys in data block,
// index block for data block, index block for index block (if index type is
// `kTwoLevelIndexSearch`), index for filter blocks (if using partitioned
// filters), the `first_internal_key` in `IndexValue`, the `end_key` for range
// deletion entries.
// As long as the user keys are sorted when added via `Add` API, their logic
// ordering won't change after timestamps are stripped. However, for each user
// key to be logically equivalent before and after timestamp is stripped, the
// user key should contain the minimum timestamp.
bool persist_user_defined_timestamps;
WritableFileWriter* file;
std::atomic<uint64_t> offset;
size_t alignment;
BlockBuilder data_block;
// Buffers uncompressed data blocks to replay later. Needed when
// compression dictionary is enabled so we can finalize the dictionary before
// compressing any data blocks.
std::vector<std::string> data_block_buffers;
BlockBuilder range_del_block;
InternalKeySliceTransform internal_prefix_transform;
std::unique_ptr<IndexBuilder> index_builder;
std::string index_separator_scratch;
PartitionedIndexBuilder* p_index_builder_ = nullptr;
std::string last_ikey; // Internal key or empty (unset)
const Slice* first_key_in_next_block = nullptr;
CompressionType compression_type;
uint64_t sample_for_compression;
std::atomic<uint64_t> compressible_input_data_bytes;
std::atomic<uint64_t> uncompressible_input_data_bytes;
std::atomic<uint64_t> sampled_input_data_bytes;
std::atomic<uint64_t> sampled_output_slow_data_bytes;
std::atomic<uint64_t> sampled_output_fast_data_bytes;
CompressionOptions compression_opts;
std::unique_ptr<CompressionDict> compression_dict;
std::vector<std::unique_ptr<CompressionContext>> compression_ctxs;
std::vector<std::unique_ptr<UncompressionContext>> verify_ctxs;
std::unique_ptr<UncompressionDict> verify_dict;
size_t data_begin_offset = 0;
TableProperties props;
// States of the builder.
//
// - `kBuffered`: This is the initial state where zero or more data blocks are
// accumulated uncompressed in-memory. From this state, call
// `EnterUnbuffered()` to finalize the compression dictionary if enabled,
// compress/write out any buffered blocks, and proceed to the `kUnbuffered`
// state.
//
// - `kUnbuffered`: This is the state when compression dictionary is finalized
// either because it wasn't enabled in the first place or it's been created
// from sampling previously buffered data. In this state, blocks are simply
// compressed/written out as they fill up. From this state, call `Finish()`
// to complete the file (write meta-blocks, etc.), or `Abandon()` to delete
// the partially created file.
//
// - `kClosed`: This indicates either `Finish()` or `Abandon()` has been
// called, so the table builder is no longer usable. We must be in this
// state by the time the destructor runs.
enum class State {
kBuffered,
kUnbuffered,
kClosed,
};
State state;
// `kBuffered` state is allowed only as long as the buffering of uncompressed
// data blocks (see `data_block_buffers`) does not exceed `buffer_limit`.
uint64_t buffer_limit;
std::shared_ptr<CacheReservationManager>
compression_dict_buffer_cache_res_mgr;
const bool use_delta_encoding_for_index_values;
std::unique_ptr<FilterBlockBuilder> filter_builder;
OffsetableCacheKey base_cache_key;
const TableFileCreationReason reason;
BlockHandle pending_handle; // Handle to add to index block
std::string compressed_output;
std::unique_ptr<FlushBlockPolicy> flush_block_policy;
std::vector<std::unique_ptr<InternalTblPropColl>> table_properties_collectors;
std::unique_ptr<ParallelCompressionRep> pc_rep;
BlockCreateContext create_context;
// The size of the "tail" part of a SST file. "Tail" refers to
// all blocks after data blocks till the end of the SST file.
uint64_t tail_size;
// See class Footer
uint32_t base_context_checksum;
uint64_t get_offset() { return offset.load(std::memory_order_relaxed); }
void set_offset(uint64_t o) { offset.store(o, std::memory_order_relaxed); }
bool IsParallelCompressionEnabled() const {
return compression_opts.parallel_threads > 1;
}
Status GetStatus() {
// We need to make modifications of status visible when status_ok is set
// to false, and this is ensured by status_mutex, so no special memory
// order for status_ok is required.
if (status_ok.load(std::memory_order_relaxed)) {
return Status::OK();
} else {
return CopyStatus();
}
}
Status CopyStatus() {
std::lock_guard<std::mutex> lock(status_mutex);
return status;
}
IOStatus GetIOStatus() {
// We need to make modifications of io_status visible when status_ok is set
// to false, and this is ensured by io_status_mutex, so no special memory
// order for io_status_ok is required.
if (io_status_ok.load(std::memory_order_relaxed)) {
#ifdef ROCKSDB_ASSERT_STATUS_CHECKED // Avoid unnecessary lock acquisition
auto ios = CopyIOStatus();
ios.PermitUncheckedError();
// Assume no races in unit tests
assert(ios.ok());
#endif // ROCKSDB_ASSERT_STATUS_CHECKED
return IOStatus::OK();
} else {
return CopyIOStatus();
}
}
IOStatus CopyIOStatus() {
std::lock_guard<std::mutex> lock(io_status_mutex);
return io_status;
}
// Never erase an existing status that is not OK.
void SetStatus(Status s) {
if (!s.ok() && status_ok.load(std::memory_order_relaxed)) {
// Locking is an overkill for non compression_opts.parallel_threads
// case but since it's unlikely that s is not OK, we take this cost
// to be simplicity.
std::lock_guard<std::mutex> lock(status_mutex);
status = s;
status_ok.store(false, std::memory_order_relaxed);
}
}
// Never erase an existing I/O status that is not OK.
// Calling this will also SetStatus(ios)
void SetIOStatus(IOStatus ios) {
if (!ios.ok() && io_status_ok.load(std::memory_order_relaxed)) {
// Locking is an overkill for non compression_opts.parallel_threads
// case but since it's unlikely that s is not OK, we take this cost
// to be simplicity.
std::lock_guard<std::mutex> lock(io_status_mutex);
io_status = ios;
io_status_ok.store(false, std::memory_order_relaxed);
}
SetStatus(ios);
}
Rep(const BlockBasedTableOptions& table_opt, const TableBuilderOptions& tbo,
WritableFileWriter* f)
: ioptions(tbo.ioptions),
prefix_extractor(tbo.moptions.prefix_extractor),
write_options(tbo.write_options),
table_options(table_opt),
internal_comparator(tbo.internal_comparator),
ts_sz(tbo.internal_comparator.user_comparator()->timestamp_size()),
persist_user_defined_timestamps(
tbo.ioptions.persist_user_defined_timestamps),
file(f),
offset(0),
alignment(table_options.block_align
? std::min(static_cast<size_t>(table_options.block_size),
kDefaultPageSize)
: 0),
data_block(table_options.block_restart_interval,
table_options.use_delta_encoding,
false /* use_value_delta_encoding */,
tbo.internal_comparator.user_comparator()
->CanKeysWithDifferentByteContentsBeEqual()
? BlockBasedTableOptions::kDataBlockBinarySearch
: table_options.data_block_index_type,
table_options.data_block_hash_table_util_ratio, ts_sz,
persist_user_defined_timestamps),
range_del_block(
1 /* block_restart_interval */, true /* use_delta_encoding */,
false /* use_value_delta_encoding */,
BlockBasedTableOptions::kDataBlockBinarySearch /* index_type */,
0.75 /* data_block_hash_table_util_ratio */, ts_sz,
persist_user_defined_timestamps),
internal_prefix_transform(prefix_extractor.get()),
compression_type(tbo.compression_type),
sample_for_compression(tbo.moptions.sample_for_compression),
compressible_input_data_bytes(0),
uncompressible_input_data_bytes(0),
sampled_input_data_bytes(0),
sampled_output_slow_data_bytes(0),
sampled_output_fast_data_bytes(0),
compression_opts(tbo.compression_opts),
compression_dict(),
compression_ctxs(tbo.compression_opts.parallel_threads),
verify_ctxs(tbo.compression_opts.parallel_threads),
verify_dict(),
state((tbo.compression_opts.max_dict_bytes > 0 &&
tbo.compression_type != kNoCompression)
? State::kBuffered
: State::kUnbuffered),
use_delta_encoding_for_index_values(table_opt.format_version >= 4 &&
!table_opt.block_align),
reason(tbo.reason),
flush_block_policy(
table_options.flush_block_policy_factory->NewFlushBlockPolicy(
table_options, data_block)),
create_context(&table_options, &ioptions, ioptions.stats,
compression_type == kZSTD ||
compression_type == kZSTDNotFinalCompression,
tbo.moptions.block_protection_bytes_per_key,
tbo.internal_comparator.user_comparator(),
!use_delta_encoding_for_index_values,
table_opt.index_type ==
BlockBasedTableOptions::kBinarySearchWithFirstKey),
tail_size(0),
status_ok(true),
io_status_ok(true) {
if (tbo.target_file_size == 0) {
buffer_limit = compression_opts.max_dict_buffer_bytes;
} else if (compression_opts.max_dict_buffer_bytes == 0) {
buffer_limit = tbo.target_file_size;
} else {
buffer_limit = std::min(tbo.target_file_size,
compression_opts.max_dict_buffer_bytes);
}
const auto compress_dict_build_buffer_charged =
table_options.cache_usage_options.options_overrides
.at(CacheEntryRole::kCompressionDictionaryBuildingBuffer)
.charged;
if (table_options.block_cache &&
(compress_dict_build_buffer_charged ==
CacheEntryRoleOptions::Decision::kEnabled ||
compress_dict_build_buffer_charged ==
CacheEntryRoleOptions::Decision::kFallback)) {
compression_dict_buffer_cache_res_mgr =
std::make_shared<CacheReservationManagerImpl<
CacheEntryRole::kCompressionDictionaryBuildingBuffer>>(
table_options.block_cache);
} else {
compression_dict_buffer_cache_res_mgr = nullptr;
}
assert(compression_ctxs.size() >= compression_opts.parallel_threads);
for (uint32_t i = 0; i < compression_opts.parallel_threads; i++) {
compression_ctxs[i].reset(
new CompressionContext(compression_type, compression_opts));
}
if (table_options.index_type ==
BlockBasedTableOptions::kTwoLevelIndexSearch) {
p_index_builder_ = PartitionedIndexBuilder::CreateIndexBuilder(
&internal_comparator, use_delta_encoding_for_index_values,
table_options, ts_sz, persist_user_defined_timestamps);
index_builder.reset(p_index_builder_);
} else {
index_builder.reset(IndexBuilder::CreateIndexBuilder(
table_options.index_type, &internal_comparator,
&this->internal_prefix_transform, use_delta_encoding_for_index_values,
table_options, ts_sz, persist_user_defined_timestamps));
}
if (ioptions.optimize_filters_for_hits && tbo.is_bottommost) {
// Apply optimize_filters_for_hits setting here when applicable by
// skipping filter generation
filter_builder.reset();
} else if (tbo.skip_filters) {
// For SstFileWriter skip_filters
filter_builder.reset();
} else if (!table_options.filter_policy) {
// Null filter_policy -> no filter
filter_builder.reset();
} else {
FilterBuildingContext filter_context(table_options);
filter_context.info_log = ioptions.logger;
filter_context.column_family_name = tbo.column_family_name;
filter_context.reason = reason;
// Only populate other fields if known to be in LSM rather than
// generating external SST file
if (reason != TableFileCreationReason::kMisc) {
filter_context.compaction_style = ioptions.compaction_style;
filter_context.num_levels = ioptions.num_levels;
filter_context.level_at_creation = tbo.level_at_creation;
filter_context.is_bottommost = tbo.is_bottommost;
assert(filter_context.level_at_creation < filter_context.num_levels);
}
filter_builder.reset(CreateFilterBlockBuilder(
ioptions, tbo.moptions, filter_context,
use_delta_encoding_for_index_values, p_index_builder_, ts_sz,
persist_user_defined_timestamps));
}
assert(tbo.internal_tbl_prop_coll_factories);
for (auto& factory : *tbo.internal_tbl_prop_coll_factories) {
assert(factory);
std::unique_ptr<InternalTblPropColl> collector{
factory->CreateInternalTblPropColl(
tbo.column_family_id, tbo.level_at_creation,
tbo.ioptions.num_levels,
tbo.last_level_inclusive_max_seqno_threshold)};
if (collector) {
table_properties_collectors.emplace_back(std::move(collector));
}
}
table_properties_collectors.emplace_back(
new BlockBasedTablePropertiesCollector(
table_options.index_type, table_options.whole_key_filtering,
prefix_extractor != nullptr));
if (ts_sz > 0 && persist_user_defined_timestamps) {
table_properties_collectors.emplace_back(
new TimestampTablePropertiesCollector(
tbo.internal_comparator.user_comparator()));
}
if (table_options.verify_compression) {
for (uint32_t i = 0; i < compression_opts.parallel_threads; i++) {
verify_ctxs[i].reset(new UncompressionContext(compression_type));
}
}
// These are only needed for populating table properties
props.column_family_id = tbo.column_family_id;
props.column_family_name = tbo.column_family_name;
props.oldest_key_time = tbo.oldest_key_time;
props.file_creation_time = tbo.file_creation_time;
props.orig_file_number = tbo.cur_file_num;
props.db_id = tbo.db_id;
props.db_session_id = tbo.db_session_id;
props.db_host_id = ioptions.db_host_id;
if (!ReifyDbHostIdProperty(ioptions.env, &props.db_host_id).ok()) {
ROCKS_LOG_INFO(ioptions.logger, "db_host_id property will not be set");
}
if (FormatVersionUsesContextChecksum(table_options.format_version)) {
// Must be non-zero and semi- or quasi-random
// TODO: ideally guaranteed different for related files (e.g. use file
// number and db_session, for benefit of SstFileWriter)
do {
base_context_checksum = Random::GetTLSInstance()->Next();
} while (UNLIKELY(base_context_checksum == 0));
} else {
base_context_checksum = 0;
}
if (alignment > 0 && compression_type != kNoCompression) {
// With better sanitization in `CompactionPicker::CompactFiles()`, we
// would not need to handle this case here and could change it to an
// assertion instead.
SetStatus(Status::InvalidArgument(
"Enable block_align, but compression enabled"));
}
}
Rep(const Rep&) = delete;
Rep& operator=(const Rep&) = delete;
private:
// Synchronize status & io_status accesses across threads from main thread,
// compression thread and write thread in parallel compression.
std::mutex status_mutex;
std::atomic<bool> status_ok;
Status status;
std::mutex io_status_mutex;
std::atomic<bool> io_status_ok;
IOStatus io_status;
};
struct BlockBasedTableBuilder::ParallelCompressionRep {
// Keys is a wrapper of vector of strings avoiding
// releasing string memories during vector clear()
// in order to save memory allocation overhead
class Keys {
public:
Keys() : keys_(kKeysInitSize), size_(0) {}
void PushBack(const Slice& key) {
if (size_ == keys_.size()) {
keys_.emplace_back(key.data(), key.size());
} else {
keys_[size_].assign(key.data(), key.size());
}
size_++;
}
void SwapAssign(std::vector<std::string>& keys) {
size_ = keys.size();
std::swap(keys_, keys);
}
void Clear() { size_ = 0; }
size_t Size() { return size_; }
std::string& Back() { return keys_[size_ - 1]; }
std::string& operator[](size_t idx) {
assert(idx < size_);
return keys_[idx];
}
private:
const size_t kKeysInitSize = 32;
std::vector<std::string> keys_;
size_t size_;
};
std::unique_ptr<Keys> curr_block_keys;
class BlockRepSlot;
// BlockRep instances are fetched from and recycled to
// block_rep_pool during parallel compression.
struct BlockRep {
Slice contents;
Slice compressed_contents;
std::unique_ptr<std::string> data;
std::unique_ptr<std::string> compressed_data;
CompressionType compression_type;
std::unique_ptr<std::string> first_key_in_next_block;
std::unique_ptr<Keys> keys;
std::unique_ptr<BlockRepSlot> slot;
Status status;
};
// Use a vector of BlockRep as a buffer for a determined number
// of BlockRep structures. All data referenced by pointers in
// BlockRep will be freed when this vector is destructed.
using BlockRepBuffer = std::vector<BlockRep>;
BlockRepBuffer block_rep_buf;
// Use a thread-safe queue for concurrent access from block
// building thread and writer thread.
using BlockRepPool = WorkQueue<BlockRep*>;
BlockRepPool block_rep_pool;
// Use BlockRepSlot to keep block order in write thread.
// slot_ will pass references to BlockRep
class BlockRepSlot {
public:
BlockRepSlot() : slot_(1) {}
template <typename T>
void Fill(T&& rep) {
slot_.push(std::forward<T>(rep));
}
void Take(BlockRep*& rep) { slot_.pop(rep); }
private:
// slot_ will pass references to BlockRep in block_rep_buf,
// and those references are always valid before the destruction of
// block_rep_buf.
WorkQueue<BlockRep*> slot_;
};
// Compression queue will pass references to BlockRep in block_rep_buf,
// and those references are always valid before the destruction of
// block_rep_buf.
using CompressQueue = WorkQueue<BlockRep*>;
CompressQueue compress_queue;
std::vector<port::Thread> compress_thread_pool;
// Write queue will pass references to BlockRep::slot in block_rep_buf,
// and those references are always valid before the corresponding
// BlockRep::slot is destructed, which is before the destruction of
// block_rep_buf.
using WriteQueue = WorkQueue<BlockRepSlot*>;
WriteQueue write_queue;
std::unique_ptr<port::Thread> write_thread;
// Estimate output file size when parallel compression is enabled. This is
// necessary because compression & flush are no longer synchronized,
// and BlockBasedTableBuilder::FileSize() is no longer accurate.
// memory_order_relaxed suffices because accurate statistics is not required.
class FileSizeEstimator {
public:
explicit FileSizeEstimator()
: uncomp_bytes_compressed(0),
uncomp_bytes_curr_block(0),
uncomp_bytes_curr_block_set(false),
uncomp_bytes_inflight(0),
blocks_inflight(0),
curr_compression_ratio(0),
estimated_file_size(0) {}
// Estimate file size when a block is about to be emitted to
// compression thread
void EmitBlock(uint64_t uncomp_block_size, uint64_t curr_file_size) {
uint64_t new_uncomp_bytes_inflight =
uncomp_bytes_inflight.fetch_add(uncomp_block_size,
std::memory_order_relaxed) +
uncomp_block_size;
uint64_t new_blocks_inflight =
blocks_inflight.fetch_add(1, std::memory_order_relaxed) + 1;
estimated_file_size.store(
curr_file_size +
static_cast<uint64_t>(
static_cast<double>(new_uncomp_bytes_inflight) *
curr_compression_ratio.load(std::memory_order_relaxed)) +
new_blocks_inflight * kBlockTrailerSize,
std::memory_order_relaxed);
}
// Estimate file size when a block is already reaped from
// compression thread
void ReapBlock(uint64_t compressed_block_size, uint64_t curr_file_size) {
assert(uncomp_bytes_curr_block_set);
uint64_t new_uncomp_bytes_compressed =
uncomp_bytes_compressed + uncomp_bytes_curr_block;
assert(new_uncomp_bytes_compressed > 0);
curr_compression_ratio.store(
(curr_compression_ratio.load(std::memory_order_relaxed) *
uncomp_bytes_compressed +
compressed_block_size) /
static_cast<double>(new_uncomp_bytes_compressed),
std::memory_order_relaxed);
uncomp_bytes_compressed = new_uncomp_bytes_compressed;
uint64_t new_uncomp_bytes_inflight =
uncomp_bytes_inflight.fetch_sub(uncomp_bytes_curr_block,
std::memory_order_relaxed) -
uncomp_bytes_curr_block;
uint64_t new_blocks_inflight =
blocks_inflight.fetch_sub(1, std::memory_order_relaxed) - 1;
estimated_file_size.store(
curr_file_size +
static_cast<uint64_t>(
static_cast<double>(new_uncomp_bytes_inflight) *
curr_compression_ratio.load(std::memory_order_relaxed)) +
new_blocks_inflight * kBlockTrailerSize,
std::memory_order_relaxed);
uncomp_bytes_curr_block_set = false;
}
void SetEstimatedFileSize(uint64_t size) {
estimated_file_size.store(size, std::memory_order_relaxed);
}
uint64_t GetEstimatedFileSize() {
return estimated_file_size.load(std::memory_order_relaxed);
}
void SetCurrBlockUncompSize(uint64_t size) {
uncomp_bytes_curr_block = size;
uncomp_bytes_curr_block_set = true;
}
private:
// Input bytes compressed so far.
uint64_t uncomp_bytes_compressed;
// Size of current block being appended.
uint64_t uncomp_bytes_curr_block;
// Whether uncomp_bytes_curr_block has been set for next
// ReapBlock call.
bool uncomp_bytes_curr_block_set;
// Input bytes under compression and not appended yet.
std::atomic<uint64_t> uncomp_bytes_inflight;
// Number of blocks under compression and not appended yet.
std::atomic<uint64_t> blocks_inflight;
// Current compression ratio, maintained by BGWorkWriteMaybeCompressedBlock.
std::atomic<double> curr_compression_ratio;
// Estimated SST file size.
std::atomic<uint64_t> estimated_file_size;
};
FileSizeEstimator file_size_estimator;
// Facilities used for waiting first block completion. Need to Wait for
// the completion of first block compression and flush to get a non-zero
// compression ratio.
std::atomic<bool> first_block_processed;
std::condition_variable first_block_cond;
std::mutex first_block_mutex;
explicit ParallelCompressionRep(uint32_t parallel_threads)
: curr_block_keys(new Keys()),
block_rep_buf(parallel_threads),
block_rep_pool(parallel_threads),
compress_queue(parallel_threads),
write_queue(parallel_threads),
first_block_processed(false) {
for (uint32_t i = 0; i < parallel_threads; i++) {
block_rep_buf[i].contents = Slice();
block_rep_buf[i].compressed_contents = Slice();
block_rep_buf[i].data.reset(new std::string());
block_rep_buf[i].compressed_data.reset(new std::string());
block_rep_buf[i].compression_type = CompressionType();
block_rep_buf[i].first_key_in_next_block.reset(new std::string());
block_rep_buf[i].keys.reset(new Keys());
block_rep_buf[i].slot.reset(new BlockRepSlot());
block_rep_buf[i].status = Status::OK();
block_rep_pool.push(&block_rep_buf[i]);
}
}
~ParallelCompressionRep() { block_rep_pool.finish(); }
// Make a block prepared to be emitted to compression thread
// Used in non-buffered mode
BlockRep* PrepareBlock(CompressionType compression_type,
const Slice* first_key_in_next_block,
BlockBuilder* data_block) {
BlockRep* block_rep =
PrepareBlockInternal(compression_type, first_key_in_next_block);
assert(block_rep != nullptr);
data_block->SwapAndReset(*(block_rep->data));
block_rep->contents = *(block_rep->data);
std::swap(block_rep->keys, curr_block_keys);
curr_block_keys->Clear();
return block_rep;
}
// Used in EnterUnbuffered
BlockRep* PrepareBlock(CompressionType compression_type,
const Slice* first_key_in_next_block,
std::string* data_block,
std::vector<std::string>* keys) {
BlockRep* block_rep =
PrepareBlockInternal(compression_type, first_key_in_next_block);
assert(block_rep != nullptr);
std::swap(*(block_rep->data), *data_block);
block_rep->contents = *(block_rep->data);
block_rep->keys->SwapAssign(*keys);
return block_rep;
}
// Emit a block to compression thread
void EmitBlock(BlockRep* block_rep) {
assert(block_rep != nullptr);
assert(block_rep->status.ok());
if (!write_queue.push(block_rep->slot.get())) {
return;
}
if (!compress_queue.push(block_rep)) {
return;
}
if (!first_block_processed.load(std::memory_order_relaxed)) {
std::unique_lock<std::mutex> lock(first_block_mutex);
first_block_cond.wait(lock, [this] {
return first_block_processed.load(std::memory_order_relaxed);
});
}
}
// Reap a block from compression thread
void ReapBlock(BlockRep* block_rep) {
assert(block_rep != nullptr);
block_rep->compressed_data->clear();
block_rep_pool.push(block_rep);
if (!first_block_processed.load(std::memory_order_relaxed)) {
std::lock_guard<std::mutex> lock(first_block_mutex);
first_block_processed.store(true, std::memory_order_relaxed);
first_block_cond.notify_one();
}
}
private:
BlockRep* PrepareBlockInternal(CompressionType compression_type,
const Slice* first_key_in_next_block) {
BlockRep* block_rep = nullptr;
block_rep_pool.pop(block_rep);
assert(block_rep != nullptr);
assert(block_rep->data);
block_rep->compression_type = compression_type;
if (first_key_in_next_block == nullptr) {
block_rep->first_key_in_next_block.reset(nullptr);
} else {
block_rep->first_key_in_next_block->assign(
first_key_in_next_block->data(), first_key_in_next_block->size());
}
return block_rep;
}
};
BlockBasedTableBuilder::BlockBasedTableBuilder(
const BlockBasedTableOptions& table_options, const TableBuilderOptions& tbo,
WritableFileWriter* file) {
BlockBasedTableOptions sanitized_table_options(table_options);
if (sanitized_table_options.format_version == 0 &&
sanitized_table_options.checksum != kCRC32c) {
ROCKS_LOG_WARN(
tbo.ioptions.logger,
"Silently converting format_version to 1 because checksum is "
"non-default");
// silently convert format_version to 1 to keep consistent with current
// behavior
sanitized_table_options.format_version = 1;
}
auto ucmp = tbo.internal_comparator.user_comparator();
assert(ucmp);
(void)ucmp; // avoids unused variable error.
rep_ = new Rep(sanitized_table_options, tbo, file);
TEST_SYNC_POINT_CALLBACK(
"BlockBasedTableBuilder::BlockBasedTableBuilder:PreSetupBaseCacheKey",
const_cast<TableProperties*>(&rep_->props));
BlockBasedTable::SetupBaseCacheKey(&rep_->props, tbo.db_session_id,
tbo.cur_file_num, &rep_->base_cache_key);
if (rep_->IsParallelCompressionEnabled()) {
StartParallelCompression();
}
}
BlockBasedTableBuilder::~BlockBasedTableBuilder() {
// Catch errors where caller forgot to call Finish()
assert(rep_->state == Rep::State::kClosed);
delete rep_;
}