forked from apache/doris
-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathbase_tablet.cpp
1651 lines (1517 loc) · 78.6 KB
/
base_tablet.cpp
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
// Licensed to the Apache Software Foundation (ASF) under one
// or more contributor license agreements. See the NOTICE file
// distributed with this work for additional information
// regarding copyright ownership. The ASF licenses this file
// to you 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.
#include "olap/base_tablet.h"
#include <fmt/format.h>
#include <rapidjson/prettywriter.h>
#include "common/status.h"
#include "olap/calc_delete_bitmap_executor.h"
#include "olap/delete_bitmap_calculator.h"
#include "olap/memtable.h"
#include "olap/partial_update_info.h"
#include "olap/primary_key_index.h"
#include "olap/rowid_conversion.h"
#include "olap/rowset/beta_rowset.h"
#include "olap/rowset/rowset.h"
#include "olap/rowset/rowset_fwd.h"
#include "olap/rowset/rowset_reader.h"
#include "olap/tablet_fwd.h"
#include "olap/txn_manager.h"
#include "service/point_query_executor.h"
#include "util/bvar_helper.h"
#include "util/crc32c.h"
#include "util/debug_points.h"
#include "util/doris_metrics.h"
#include "vec/common/assert_cast.h"
#include "vec/common/schema_util.h"
#include "vec/data_types/data_type_factory.hpp"
#include "vec/jsonb/serialize.h"
namespace doris {
using namespace ErrorCode;
namespace {
bvar::LatencyRecorder g_tablet_commit_phase_update_delete_bitmap_latency(
"doris_pk", "commit_phase_update_delete_bitmap");
bvar::LatencyRecorder g_tablet_lookup_rowkey_latency("doris_pk", "tablet_lookup_rowkey");
bvar::Adder<uint64_t> g_tablet_pk_not_found("doris_pk", "lookup_not_found");
bvar::PerSecond<bvar::Adder<uint64_t>> g_tablet_pk_not_found_per_second(
"doris_pk", "lookup_not_found_per_second", &g_tablet_pk_not_found, 60);
bvar::LatencyRecorder g_tablet_update_delete_bitmap_latency("doris_pk", "update_delete_bitmap");
static bvar::Adder<size_t> g_total_tablet_num("doris_total_tablet_num");
Status _get_segment_column_iterator(const BetaRowsetSharedPtr& rowset, uint32_t segid,
const TabletColumn& target_column,
SegmentCacheHandle* segment_cache_handle,
std::unique_ptr<segment_v2::ColumnIterator>* column_iterator,
OlapReaderStatistics* stats) {
RETURN_IF_ERROR(SegmentLoader::instance()->load_segments(rowset, segment_cache_handle, true));
// find segment
auto it = std::find_if(
segment_cache_handle->get_segments().begin(),
segment_cache_handle->get_segments().end(),
[&segid](const segment_v2::SegmentSharedPtr& seg) { return seg->id() == segid; });
if (it == segment_cache_handle->get_segments().end()) {
return Status::NotFound(fmt::format("rowset {} 's segemnt not found, seg_id {}",
rowset->rowset_id().to_string(), segid));
}
segment_v2::SegmentSharedPtr segment = *it;
RETURN_IF_ERROR(segment->new_column_iterator(target_column, column_iterator, nullptr));
segment_v2::ColumnIteratorOptions opt {
.use_page_cache = !config::disable_storage_page_cache,
.file_reader = segment->file_reader().get(),
.stats = stats,
.io_ctx = io::IOContext {.reader_type = ReaderType::READER_QUERY},
};
RETURN_IF_ERROR((*column_iterator)->init(opt));
return Status::OK();
}
} // namespace
extern MetricPrototype METRIC_query_scan_bytes;
extern MetricPrototype METRIC_query_scan_rows;
extern MetricPrototype METRIC_query_scan_count;
DEFINE_COUNTER_METRIC_PROTOTYPE_2ARG(flush_bytes, MetricUnit::BYTES);
DEFINE_COUNTER_METRIC_PROTOTYPE_2ARG(flush_finish_count, MetricUnit::OPERATIONS);
BaseTablet::BaseTablet(TabletMetaSharedPtr tablet_meta) : _tablet_meta(std::move(tablet_meta)) {
_metric_entity = DorisMetrics::instance()->metric_registry()->register_entity(
fmt::format("Tablet.{}", tablet_id()), {{"tablet_id", std::to_string(tablet_id())}},
MetricEntityType::kTablet);
INT_COUNTER_METRIC_REGISTER(_metric_entity, query_scan_bytes);
INT_COUNTER_METRIC_REGISTER(_metric_entity, query_scan_rows);
INT_COUNTER_METRIC_REGISTER(_metric_entity, query_scan_count);
INT_COUNTER_METRIC_REGISTER(_metric_entity, flush_bytes);
INT_COUNTER_METRIC_REGISTER(_metric_entity, flush_finish_count);
// construct _timestamped_versioned_tracker from rs and stale rs meta
_timestamped_version_tracker.construct_versioned_tracker(_tablet_meta->all_rs_metas(),
_tablet_meta->all_stale_rs_metas());
// if !_tablet_meta->all_rs_metas()[0]->tablet_schema(),
// that mean the tablet_meta is still no upgrade to doris 1.2 versions.
// Before doris 1.2 version, rowset metas don't have tablet schema.
// And when upgrade to doris 1.2 version,
// all rowset metas will be set the tablet schmea from tablet meta.
if (_tablet_meta->all_rs_metas().empty() || !_tablet_meta->all_rs_metas()[0]->tablet_schema()) {
_max_version_schema = _tablet_meta->tablet_schema();
} else {
_max_version_schema =
tablet_schema_with_merged_max_schema_version(_tablet_meta->all_rs_metas());
}
DCHECK(_max_version_schema);
g_total_tablet_num << 1;
}
BaseTablet::~BaseTablet() {
DorisMetrics::instance()->metric_registry()->deregister_entity(_metric_entity);
g_total_tablet_num << -1;
}
TabletSchemaSPtr BaseTablet::tablet_schema_with_merged_max_schema_version(
const std::vector<RowsetMetaSharedPtr>& rowset_metas) {
RowsetMetaSharedPtr max_schema_version_rs = *std::max_element(
rowset_metas.begin(), rowset_metas.end(),
[](const RowsetMetaSharedPtr& a, const RowsetMetaSharedPtr& b) {
return !a->tablet_schema()
? true
: (!b->tablet_schema()
? false
: a->tablet_schema()->schema_version() <
b->tablet_schema()->schema_version());
});
TabletSchemaSPtr target_schema = max_schema_version_rs->tablet_schema();
if (target_schema->num_variant_columns() > 0) {
// For variant columns tablet schema need to be the merged wide tablet schema
std::vector<TabletSchemaSPtr> schemas;
std::transform(rowset_metas.begin(), rowset_metas.end(), std::back_inserter(schemas),
[](const RowsetMetaSharedPtr& rs_meta) { return rs_meta->tablet_schema(); });
static_cast<void>(
vectorized::schema_util::get_least_common_schema(schemas, nullptr, target_schema));
VLOG_DEBUG << "dump schema: " << target_schema->dump_full_schema();
}
return target_schema;
}
Status BaseTablet::set_tablet_state(TabletState state) {
if (_tablet_meta->tablet_state() == TABLET_SHUTDOWN && state != TABLET_SHUTDOWN) {
return Status::Error<META_INVALID_ARGUMENT>(
"could not change tablet state from shutdown to {}", state);
}
_tablet_meta->set_tablet_state(state);
return Status::OK();
}
void BaseTablet::update_max_version_schema(const TabletSchemaSPtr& tablet_schema) {
std::lock_guard wrlock(_meta_lock);
// Double Check for concurrent update
if (!_max_version_schema ||
tablet_schema->schema_version() > _max_version_schema->schema_version()) {
_max_version_schema = tablet_schema;
}
}
Status BaseTablet::update_by_least_common_schema(const TabletSchemaSPtr& update_schema) {
std::lock_guard wrlock(_meta_lock);
CHECK(_max_version_schema->schema_version() >= update_schema->schema_version());
TabletSchemaSPtr final_schema;
bool check_column_size = true;
RETURN_IF_ERROR(vectorized::schema_util::get_least_common_schema(
{_max_version_schema, update_schema}, _max_version_schema, final_schema,
check_column_size));
_max_version_schema = final_schema;
VLOG_DEBUG << "dump updated tablet schema: " << final_schema->dump_structure();
return Status::OK();
}
uint32_t BaseTablet::get_real_compaction_score() const {
const auto& rs_metas = _tablet_meta->all_rs_metas();
return std::accumulate(rs_metas.begin(), rs_metas.end(), 0,
[](uint32_t score, const RowsetMetaSharedPtr& rs_meta) {
return score + rs_meta->get_compaction_score();
});
}
Status BaseTablet::capture_rs_readers_unlocked(const Versions& version_path,
std::vector<RowSetSplits>* rs_splits) const {
DCHECK(rs_splits != nullptr && rs_splits->empty());
for (auto version : version_path) {
auto it = _rs_version_map.find(version);
if (it == _rs_version_map.end()) {
VLOG_NOTICE << "fail to find Rowset in rs_version for version. tablet=" << tablet_id()
<< ", version='" << version.first << "-" << version.second;
it = _stale_rs_version_map.find(version);
if (it == _stale_rs_version_map.end()) {
return Status::Error<CAPTURE_ROWSET_READER_ERROR>(
"fail to find Rowset in stale_rs_version for version. tablet={}, "
"version={}-{}",
tablet_id(), version.first, version.second);
}
}
RowsetReaderSharedPtr rs_reader;
auto res = it->second->create_reader(&rs_reader);
if (!res.ok()) {
return Status::Error<CAPTURE_ROWSET_READER_ERROR>(
"failed to create reader for rowset:{}", it->second->rowset_id().to_string());
}
rs_splits->emplace_back(std::move(rs_reader));
}
return Status::OK();
}
// snapshot manager may call this api to check if version exists, so that
// the version maybe not exist
RowsetSharedPtr BaseTablet::get_rowset_by_version(const Version& version,
bool find_in_stale) const {
auto iter = _rs_version_map.find(version);
if (iter == _rs_version_map.end()) {
if (find_in_stale) {
return get_stale_rowset_by_version(version);
}
return nullptr;
}
return iter->second;
}
RowsetSharedPtr BaseTablet::get_stale_rowset_by_version(const Version& version) const {
auto iter = _stale_rs_version_map.find(version);
if (iter == _stale_rs_version_map.end()) {
VLOG_NOTICE << "no rowset for version:" << version << ", tablet: " << tablet_id();
return nullptr;
}
return iter->second;
}
// Already under _meta_lock
RowsetSharedPtr BaseTablet::get_rowset_with_max_version() const {
Version max_version = _tablet_meta->max_version();
if (max_version.first == -1) {
return nullptr;
}
auto iter = _rs_version_map.find(max_version);
if (iter == _rs_version_map.end()) {
DCHECK(false) << "invalid version:" << max_version;
return nullptr;
}
return iter->second;
}
Status BaseTablet::get_all_rs_id(int64_t max_version, RowsetIdUnorderedSet* rowset_ids) const {
std::shared_lock rlock(_meta_lock);
return get_all_rs_id_unlocked(max_version, rowset_ids);
}
Status BaseTablet::get_all_rs_id_unlocked(int64_t max_version,
RowsetIdUnorderedSet* rowset_ids) const {
// Ensure that the obtained versions of rowsets are continuous
Version spec_version(0, max_version);
Versions version_path;
auto st = _timestamped_version_tracker.capture_consistent_versions(spec_version, &version_path);
if (!st.ok()) [[unlikely]] {
return st;
}
for (auto& ver : version_path) {
if (ver.second == 1) {
// [0-1] rowset is empty for each tablet, skip it
continue;
}
auto it = _rs_version_map.find(ver);
if (it == _rs_version_map.end()) {
return Status::Error<CAPTURE_ROWSET_ERROR, false>(
"fail to find Rowset for version. tablet={}, version={}", tablet_id(),
ver.to_string());
}
rowset_ids->emplace(it->second->rowset_id());
}
return Status::OK();
}
Versions BaseTablet::get_missed_versions(int64_t spec_version) const {
DCHECK(spec_version > 0) << "invalid spec_version: " << spec_version;
Versions existing_versions;
{
std::shared_lock rdlock(_meta_lock);
for (const auto& rs : _tablet_meta->all_rs_metas()) {
existing_versions.emplace_back(rs->version());
}
}
return calc_missed_versions(spec_version, std::move(existing_versions));
}
Versions BaseTablet::get_missed_versions_unlocked(int64_t spec_version) const {
DCHECK(spec_version > 0) << "invalid spec_version: " << spec_version;
Versions existing_versions;
for (const auto& rs : _tablet_meta->all_rs_metas()) {
existing_versions.emplace_back(rs->version());
}
return calc_missed_versions(spec_version, std::move(existing_versions));
}
void BaseTablet::_print_missed_versions(const Versions& missed_versions) const {
std::stringstream ss;
ss << tablet_id() << " has " << missed_versions.size() << " missed version:";
// print at most 10 version
for (int i = 0; i < 10 && i < missed_versions.size(); ++i) {
ss << missed_versions[i] << ",";
}
LOG(WARNING) << ss.str();
}
bool BaseTablet::_reconstruct_version_tracker_if_necessary() {
double orphan_vertex_ratio = _timestamped_version_tracker.get_orphan_vertex_ratio();
if (orphan_vertex_ratio >= config::tablet_version_graph_orphan_vertex_ratio) {
_timestamped_version_tracker.construct_versioned_tracker(
_tablet_meta->all_rs_metas(), _tablet_meta->all_stale_rs_metas());
return true;
}
return false;
}
// should use this method to get a copy of current tablet meta
// there are some rowset meta in local meta store and in in-memory tablet meta
// but not in tablet meta in local meta store
void BaseTablet::generate_tablet_meta_copy(TabletMeta& new_tablet_meta) const {
TabletMetaPB tablet_meta_pb;
{
std::shared_lock rdlock(_meta_lock);
_tablet_meta->to_meta_pb(&tablet_meta_pb);
}
generate_tablet_meta_copy_unlocked(new_tablet_meta);
}
// this is a unlocked version of generate_tablet_meta_copy()
// some method already hold the _meta_lock before calling this,
// such as EngineCloneTask::_finish_clone -> tablet->revise_tablet_meta
void BaseTablet::generate_tablet_meta_copy_unlocked(TabletMeta& new_tablet_meta) const {
TabletMetaPB tablet_meta_pb;
_tablet_meta->to_meta_pb(&tablet_meta_pb);
new_tablet_meta.init_from_pb(tablet_meta_pb);
}
Status BaseTablet::calc_delete_bitmap_between_segments(
RowsetSharedPtr rowset, const std::vector<segment_v2::SegmentSharedPtr>& segments,
DeleteBitmapPtr delete_bitmap) {
size_t const num_segments = segments.size();
if (num_segments < 2) {
return Status::OK();
}
OlapStopWatch watch;
auto const rowset_id = rowset->rowset_id();
size_t seq_col_length = 0;
if (_tablet_meta->tablet_schema()->has_sequence_col()) {
auto seq_col_idx = _tablet_meta->tablet_schema()->sequence_col_idx();
seq_col_length = _tablet_meta->tablet_schema()->column(seq_col_idx).length() + 1;
}
size_t rowid_length = 0;
if (!_tablet_meta->tablet_schema()->cluster_key_idxes().empty()) {
rowid_length = PrimaryKeyIndexReader::ROW_ID_LENGTH;
}
MergeIndexDeleteBitmapCalculator calculator;
RETURN_IF_ERROR(calculator.init(rowset_id, segments, seq_col_length, rowid_length));
RETURN_IF_ERROR(calculator.calculate_all(delete_bitmap));
LOG(INFO) << fmt::format(
"construct delete bitmap between segments, "
"tablet: {}, rowset: {}, number of segments: {}, bitmap size: {}, cost {} (us)",
tablet_id(), rowset_id.to_string(), num_segments, delete_bitmap->delete_bitmap.size(),
watch.get_elapse_time_us());
return Status::OK();
}
std::vector<RowsetSharedPtr> BaseTablet::get_rowset_by_ids(
const RowsetIdUnorderedSet* specified_rowset_ids) {
std::vector<RowsetSharedPtr> rowsets;
for (auto& rs : _rs_version_map) {
if (!specified_rowset_ids ||
specified_rowset_ids->find(rs.second->rowset_id()) != specified_rowset_ids->end()) {
rowsets.push_back(rs.second);
}
}
std::sort(rowsets.begin(), rowsets.end(), [](RowsetSharedPtr& lhs, RowsetSharedPtr& rhs) {
return lhs->end_version() > rhs->end_version();
});
return rowsets;
}
Status BaseTablet::lookup_row_data(const Slice& encoded_key, const RowLocation& row_location,
RowsetSharedPtr input_rowset, const TupleDescriptor* desc,
OlapReaderStatistics& stats, std::string& values,
bool write_to_cache) {
MonotonicStopWatch watch;
size_t row_size = 1;
watch.start();
Defer _defer([&]() {
LOG_EVERY_N(INFO, 500) << "get a single_row, cost(us):" << watch.elapsed_time() / 1000
<< ", row_size:" << row_size;
});
BetaRowsetSharedPtr rowset = std::static_pointer_cast<BetaRowset>(input_rowset);
CHECK(rowset);
const TabletSchemaSPtr tablet_schema = rowset->tablet_schema();
SegmentCacheHandle segment_cache_handle;
std::unique_ptr<segment_v2::ColumnIterator> column_iterator;
const auto& column = *DORIS_TRY(tablet_schema->column(BeConsts::ROW_STORE_COL));
RETURN_IF_ERROR(_get_segment_column_iterator(rowset, row_location.segment_id, column,
&segment_cache_handle, &column_iterator, &stats));
// get and parse tuple row
vectorized::MutableColumnPtr column_ptr = vectorized::ColumnString::create();
std::vector<segment_v2::rowid_t> rowids {static_cast<segment_v2::rowid_t>(row_location.row_id)};
RETURN_IF_ERROR(column_iterator->read_by_rowids(rowids.data(), 1, column_ptr));
assert(column_ptr->size() == 1);
auto* string_column = static_cast<vectorized::ColumnString*>(column_ptr.get());
StringRef value = string_column->get_data_at(0);
values = value.to_string();
if (write_to_cache) {
StringRef value = string_column->get_data_at(0);
RowCache::instance()->insert({tablet_id(), encoded_key}, Slice {value.data, value.size});
}
return Status::OK();
}
Status BaseTablet::lookup_row_key(const Slice& encoded_key, TabletSchema* latest_schema,
bool with_seq_col,
const std::vector<RowsetSharedPtr>& specified_rowsets,
RowLocation* row_location, uint32_t version,
std::vector<std::unique_ptr<SegmentCacheHandle>>& segment_caches,
RowsetSharedPtr* rowset, bool with_rowid) {
SCOPED_BVAR_LATENCY(g_tablet_lookup_rowkey_latency);
size_t seq_col_length = 0;
// use the latest tablet schema to decide if the tablet has sequence column currently
const TabletSchema* schema =
(latest_schema == nullptr ? _tablet_meta->tablet_schema().get() : latest_schema);
if (schema->has_sequence_col() && with_seq_col) {
seq_col_length = schema->column(schema->sequence_col_idx()).length() + 1;
}
size_t rowid_length = 0;
if (with_rowid && !schema->cluster_key_idxes().empty()) {
rowid_length = PrimaryKeyIndexReader::ROW_ID_LENGTH;
}
Slice key_without_seq =
Slice(encoded_key.get_data(), encoded_key.get_size() - seq_col_length - rowid_length);
RowLocation loc;
for (size_t i = 0; i < specified_rowsets.size(); i++) {
auto& rs = specified_rowsets[i];
auto& segments_key_bounds = rs->rowset_meta()->get_segments_key_bounds();
int num_segments = rs->num_segments();
DCHECK_EQ(segments_key_bounds.size(), num_segments);
std::vector<uint32_t> picked_segments;
for (int i = num_segments - 1; i >= 0; i--) {
// If mow table has cluster keys, the key bounds is short keys, not primary keys
// use PrimaryKeyIndexMetaPB in primary key index?
if (schema->cluster_key_idxes().empty()) {
if (key_without_seq.compare(segments_key_bounds[i].max_key()) > 0 ||
key_without_seq.compare(segments_key_bounds[i].min_key()) < 0) {
continue;
}
}
picked_segments.emplace_back(i);
}
if (picked_segments.empty()) {
continue;
}
if (UNLIKELY(segment_caches[i] == nullptr)) {
segment_caches[i] = std::make_unique<SegmentCacheHandle>();
RETURN_IF_ERROR(SegmentLoader::instance()->load_segments(
std::static_pointer_cast<BetaRowset>(rs), segment_caches[i].get(), true, true));
}
auto& segments = segment_caches[i]->get_segments();
DCHECK_EQ(segments.size(), num_segments);
for (auto id : picked_segments) {
Status s = segments[id]->lookup_row_key(encoded_key, schema, with_seq_col, with_rowid,
&loc);
if (s.is<KEY_NOT_FOUND>()) {
continue;
}
if (!s.ok() && !s.is<KEY_ALREADY_EXISTS>()) {
return s;
}
if (s.ok() && _tablet_meta->delete_bitmap().contains_agg_without_cache(
{loc.rowset_id, loc.segment_id, version}, loc.row_id)) {
// if has sequence col, we continue to compare the sequence_id of
// all rowsets, util we find an existing key.
if (schema->has_sequence_col()) {
continue;
}
// The key is deleted, we don't need to search for it any more.
break;
}
// `st` is either OK or KEY_ALREADY_EXISTS now.
// for partial update, even if the key is already exists, we still need to
// read it's original values to keep all columns align.
*row_location = loc;
if (rowset) {
// return it's rowset
*rowset = rs;
}
// find it and return
return s;
}
}
g_tablet_pk_not_found << 1;
return Status::Error<ErrorCode::KEY_NOT_FOUND>("can't find key in all rowsets");
}
// if user pass a token, then all calculation works will submit to a threadpool,
// user can get all delete bitmaps from that token.
// if `token` is nullptr, the calculation will run in local, and user can get the result
// delete bitmap from `delete_bitmap` directly.
Status BaseTablet::calc_delete_bitmap(const BaseTabletSPtr& tablet, RowsetSharedPtr rowset,
const std::vector<segment_v2::SegmentSharedPtr>& segments,
const std::vector<RowsetSharedPtr>& specified_rowsets,
DeleteBitmapPtr delete_bitmap, int64_t end_version,
CalcDeleteBitmapToken* token, RowsetWriter* rowset_writer) {
auto rowset_id = rowset->rowset_id();
if (specified_rowsets.empty() || segments.empty()) {
LOG(INFO) << "skip to construct delete bitmap tablet: " << tablet->tablet_id()
<< " rowset: " << rowset_id;
return Status::OK();
}
OlapStopWatch watch;
for (const auto& segment : segments) {
const auto& seg = segment;
if (token != nullptr) {
RETURN_IF_ERROR(token->submit(tablet, rowset, seg, specified_rowsets, end_version,
delete_bitmap, rowset_writer));
} else {
RETURN_IF_ERROR(tablet->calc_segment_delete_bitmap(
rowset, segment, specified_rowsets, delete_bitmap, end_version, rowset_writer));
}
}
return Status::OK();
}
Status BaseTablet::calc_segment_delete_bitmap(RowsetSharedPtr rowset,
const segment_v2::SegmentSharedPtr& seg,
const std::vector<RowsetSharedPtr>& specified_rowsets,
DeleteBitmapPtr delete_bitmap, int64_t end_version,
RowsetWriter* rowset_writer) {
OlapStopWatch watch;
auto rowset_id = rowset->rowset_id();
Version dummy_version(end_version + 1, end_version + 1);
auto rowset_schema = rowset->tablet_schema();
bool is_partial_update = rowset_writer && rowset_writer->is_partial_update();
bool have_input_seq_column = false;
if (is_partial_update && rowset_schema->has_sequence_col()) {
std::vector<uint32_t> including_cids =
rowset_writer->get_partial_update_info()->update_cids;
have_input_seq_column =
rowset_schema->has_sequence_col() &&
(std::find(including_cids.cbegin(), including_cids.cend(),
rowset_schema->sequence_col_idx()) != including_cids.cend());
}
if (rowset_schema->num_variant_columns() > 0) {
// During partial updates, the extracted columns of a variant should not be included in the rowset schema.
// This is because the partial update for a variant needs to ignore the extracted columns.
// Otherwise, the schema types in different rowsets might be inconsistent. When performing a partial update,
// the complete variant is constructed by reading all the sub-columns of the variant.
rowset_schema = rowset_schema->copy_without_variant_extracted_columns();
}
// use for partial update
PartialUpdateReadPlan read_plan_ori;
PartialUpdateReadPlan read_plan_update;
int64_t conflict_rows = 0;
int64_t new_generated_rows = 0;
std::map<RowsetId, RowsetSharedPtr> rsid_to_rowset;
rsid_to_rowset[rowset_id] = rowset;
vectorized::Block block = rowset_schema->create_block();
vectorized::Block ordered_block = block.clone_empty();
uint32_t pos = 0;
RETURN_IF_ERROR(seg->load_pk_index_and_bf()); // We need index blocks to iterate
const auto* pk_idx = seg->get_primary_key_index();
int total = pk_idx->num_rows();
uint32_t row_id = 0;
int32_t remaining = total;
bool exact_match = false;
std::string last_key;
int batch_size = 1024;
// The data for each segment may be lookup multiple times. Creating a SegmentCacheHandle
// will update the lru cache, and there will be obvious lock competition in multithreading
// scenarios, so using a segment_caches to cache SegmentCacheHandle.
std::vector<std::unique_ptr<SegmentCacheHandle>> segment_caches(specified_rowsets.size());
while (remaining > 0) {
std::unique_ptr<segment_v2::IndexedColumnIterator> iter;
RETURN_IF_ERROR(pk_idx->new_iterator(&iter));
size_t num_to_read = std::min(batch_size, remaining);
auto index_type = vectorized::DataTypeFactory::instance().create_data_type(
pk_idx->type_info()->type(), 1, 0);
auto index_column = index_type->create_column();
Slice last_key_slice(last_key);
RETURN_IF_ERROR(iter->seek_at_or_after(&last_key_slice, &exact_match));
auto current_ordinal = iter->get_current_ordinal();
DCHECK(total == remaining + current_ordinal)
<< "total: " << total << ", remaining: " << remaining
<< ", current_ordinal: " << current_ordinal;
size_t num_read = num_to_read;
RETURN_IF_ERROR(iter->next_batch(&num_read, index_column));
DCHECK(num_to_read == num_read)
<< "num_to_read: " << num_to_read << ", num_read: " << num_read;
last_key = index_column->get_data_at(num_read - 1).to_string();
// exclude last_key, last_key will be read in next batch.
if (num_read == batch_size && num_read != remaining) {
num_read -= 1;
}
for (size_t i = 0; i < num_read; i++, row_id++) {
Slice key = Slice(index_column->get_data_at(i).data, index_column->get_data_at(i).size);
RowLocation loc;
// calculate row id
if (!_tablet_meta->tablet_schema()->cluster_key_idxes().empty()) {
size_t seq_col_length = 0;
if (_tablet_meta->tablet_schema()->has_sequence_col()) {
seq_col_length =
_tablet_meta->tablet_schema()
->column(_tablet_meta->tablet_schema()->sequence_col_idx())
.length() +
1;
}
size_t rowid_length = PrimaryKeyIndexReader::ROW_ID_LENGTH;
Slice key_without_seq =
Slice(key.get_data(), key.get_size() - seq_col_length - rowid_length);
Slice rowid_slice =
Slice(key.get_data() + key_without_seq.get_size() + seq_col_length + 1,
rowid_length - 1);
const auto* type_info =
get_scalar_type_info<FieldType::OLAP_FIELD_TYPE_UNSIGNED_INT>();
const auto* rowid_coder = get_key_coder(type_info->type());
RETURN_IF_ERROR(rowid_coder->decode_ascending(&rowid_slice, rowid_length,
(uint8_t*)&row_id));
}
// same row in segments should be filtered
if (delete_bitmap->contains({rowset_id, seg->id(), DeleteBitmap::TEMP_VERSION_COMMON},
row_id)) {
continue;
}
RowsetSharedPtr rowset_find;
auto st = lookup_row_key(key, rowset_schema.get(), true, specified_rowsets, &loc,
dummy_version.first - 1, segment_caches, &rowset_find);
bool expected_st = st.ok() || st.is<KEY_NOT_FOUND>() || st.is<KEY_ALREADY_EXISTS>();
// It's a defensive DCHECK, we need to exclude some common errors to avoid core-dump
// while stress test
DCHECK(expected_st || st.is<MEM_LIMIT_EXCEEDED>())
<< "unexpected error status while lookup_row_key:" << st;
if (!expected_st) {
return st;
}
if (st.is<KEY_NOT_FOUND>()) {
continue;
}
if (st.is<KEY_ALREADY_EXISTS>() && (!is_partial_update || have_input_seq_column)) {
// `st.is<KEY_ALREADY_EXISTS>()` means that there exists a row with the same key and larger value
// in seqeunce column.
// - If the current load is not a partial update, we just delete current row.
// - Otherwise, it means that we are doing the alignment process in publish phase due to conflicts
// during concurrent partial updates. And there exists another load which introduces a row with
// the same keys and larger sequence column value published successfully after the commit phase
// of the current load.
// - If the columns we update include sequence column, we should delete the current row becase the
// partial update on the current row has been `overwritten` by the previous one with larger sequence
// column value.
// - Otherwise, we should combine the values of the missing columns in the previous row and the values
// of the including columns in the current row into a new row.
delete_bitmap->add({rowset_id, seg->id(), DeleteBitmap::TEMP_VERSION_COMMON},
row_id);
++conflict_rows;
continue;
}
if (is_partial_update && rowset_writer != nullptr) {
// In publish version, record rows to be deleted for concurrent update
// For example, if version 5 and 6 update a row, but version 6 only see
// version 4 when write, and when publish version, version 5's value will
// be marked as deleted and it's update is losed.
// So here we should read version 5's columns and build a new row, which is
// consists of version 6's update columns and version 5's origin columns
// here we build 2 read plan for ori values and update values
read_plan_ori.prepare_to_read(loc, pos);
read_plan_update.prepare_to_read(RowLocation {rowset_id, seg->id(), row_id}, pos);
rsid_to_rowset[rowset_find->rowset_id()] = rowset_find;
++pos;
// delete bitmap will be calculate when memtable flush and
// publish. The two stages may see different versions.
// When there is sequence column, the currently imported data
// of rowset may be marked for deletion at memtablet flush or
// publish because the seq column is smaller than the previous
// rowset.
// just set 0 as a unified temporary version number, and update to
// the real version number later.
delete_bitmap->add(
{loc.rowset_id, loc.segment_id, DeleteBitmap::TEMP_VERSION_COMMON},
loc.row_id);
delete_bitmap->add({rowset_id, seg->id(), DeleteBitmap::TEMP_VERSION_COMMON},
row_id);
++conflict_rows;
++new_generated_rows;
continue;
}
// when st = ok
delete_bitmap->add({loc.rowset_id, loc.segment_id, DeleteBitmap::TEMP_VERSION_COMMON},
loc.row_id);
++conflict_rows;
}
remaining -= num_read;
}
// DCHECK_EQ(total, row_id) << "segment total rows: " << total << " row_id:" << row_id;
if (config::enable_merge_on_write_correctness_check) {
RowsetIdUnorderedSet rowsetids;
for (const auto& rowset : specified_rowsets) {
rowsetids.emplace(rowset->rowset_id());
VLOG_NOTICE << "[tabletID:" << tablet_id() << "]"
<< "[add_sentinel_mark_to_delete_bitmap][end_version:" << end_version << "]"
<< "add:" << rowset->rowset_id();
}
add_sentinel_mark_to_delete_bitmap(delete_bitmap.get(), rowsetids);
}
if (pos > 0) {
auto partial_update_info = rowset_writer->get_partial_update_info();
DCHECK(partial_update_info);
RETURN_IF_ERROR(generate_new_block_for_partial_update(
rowset_schema, partial_update_info.get(), read_plan_ori, read_plan_update,
rsid_to_rowset, &block));
RETURN_IF_ERROR(sort_block(block, ordered_block));
RETURN_IF_ERROR(rowset_writer->flush_single_block(&ordered_block));
if (new_generated_rows != rowset_writer->num_rows()) {
LOG(WARNING) << "partial update correctness warning: conflict new generated rows ("
<< new_generated_rows << ") not equal to the new flushed rows ("
<< rowset_writer->num_rows() << "), tablet: " << tablet_id();
}
LOG(INFO) << "calc segment delete bitmap for partial update, tablet: " << tablet_id()
<< " rowset: " << rowset_id << " seg_id: " << seg->id()
<< " dummy_version: " << end_version + 1 << " rows: " << seg->num_rows()
<< " conflict rows: " << conflict_rows
<< " new generated rows: " << new_generated_rows
<< " bimap num: " << delete_bitmap->delete_bitmap.size()
<< " cost: " << watch.get_elapse_time_us() << "(us)";
return Status::OK();
}
LOG(INFO) << "calc segment delete bitmap, tablet: " << tablet_id() << " rowset: " << rowset_id
<< " seg_id: " << seg->id() << " dummy_version: " << end_version + 1
<< " rows: " << seg->num_rows() << " conflict rows: " << conflict_rows
<< " bitmap num: " << delete_bitmap->delete_bitmap.size()
<< " cost: " << watch.get_elapse_time_us() << "(us)";
return Status::OK();
}
Status BaseTablet::sort_block(vectorized::Block& in_block, vectorized::Block& output_block) {
vectorized::MutableBlock mutable_input_block =
vectorized::MutableBlock::build_mutable_block(&in_block);
vectorized::MutableBlock mutable_output_block =
vectorized::MutableBlock::build_mutable_block(&output_block);
std::shared_ptr<RowInBlockComparator> vec_row_comparator =
std::make_shared<RowInBlockComparator>(_tablet_meta->tablet_schema());
vec_row_comparator->set_block(&mutable_input_block);
std::vector<std::unique_ptr<RowInBlock>> row_in_blocks;
DCHECK(in_block.rows() <= std::numeric_limits<int>::max());
row_in_blocks.reserve(in_block.rows());
for (size_t i = 0; i < in_block.rows(); ++i) {
row_in_blocks.emplace_back(std::make_unique<RowInBlock>(i));
}
std::sort(row_in_blocks.begin(), row_in_blocks.end(),
[&](const std::unique_ptr<RowInBlock>& l,
const std::unique_ptr<RowInBlock>& r) -> bool {
auto value = (*vec_row_comparator)(l.get(), r.get());
DCHECK(value != 0) << "value equel when sort block, l_pos: " << l->_row_pos
<< " r_pos: " << r->_row_pos;
return value < 0;
});
std::vector<uint32_t> row_pos_vec;
row_pos_vec.reserve(in_block.rows());
for (auto& block : row_in_blocks) {
row_pos_vec.emplace_back(block->_row_pos);
}
return mutable_output_block.add_rows(&in_block, row_pos_vec.data(),
row_pos_vec.data() + in_block.rows());
}
// fetch value by row column
Status BaseTablet::fetch_value_through_row_column(RowsetSharedPtr input_rowset,
const TabletSchema& tablet_schema, uint32_t segid,
const std::vector<uint32_t>& rowids,
const std::vector<uint32_t>& cids,
vectorized::Block& block) {
MonotonicStopWatch watch;
watch.start();
Defer _defer([&]() {
LOG_EVERY_N(INFO, 500) << "fetch_value_by_rowids, cost(us):" << watch.elapsed_time() / 1000
<< ", row_batch_size:" << rowids.size();
});
BetaRowsetSharedPtr rowset = std::static_pointer_cast<BetaRowset>(input_rowset);
CHECK(rowset);
CHECK(tablet_schema.has_row_store_for_all_columns());
SegmentCacheHandle segment_cache_handle;
std::unique_ptr<segment_v2::ColumnIterator> column_iterator;
OlapReaderStatistics stats;
const auto& column = *DORIS_TRY(tablet_schema.column(BeConsts::ROW_STORE_COL));
RETURN_IF_ERROR(_get_segment_column_iterator(rowset, segid, column, &segment_cache_handle,
&column_iterator, &stats));
// get and parse tuple row
vectorized::MutableColumnPtr column_ptr = vectorized::ColumnString::create();
RETURN_IF_ERROR(column_iterator->read_by_rowids(rowids.data(), rowids.size(), column_ptr));
assert(column_ptr->size() == rowids.size());
auto* string_column = static_cast<vectorized::ColumnString*>(column_ptr.get());
vectorized::DataTypeSerDeSPtrs serdes;
serdes.resize(cids.size());
std::unordered_map<uint32_t, uint32_t> col_uid_to_idx;
std::vector<std::string> default_values;
default_values.resize(cids.size());
for (int i = 0; i < cids.size(); ++i) {
const TabletColumn& column = tablet_schema.column(cids[i]);
vectorized::DataTypePtr type =
vectorized::DataTypeFactory::instance().create_data_type(column);
col_uid_to_idx[column.unique_id()] = i;
default_values[i] = column.default_value();
serdes[i] = type->get_serde();
}
vectorized::JsonbSerializeUtil::jsonb_to_block(serdes, *string_column, col_uid_to_idx, block,
default_values, {});
return Status::OK();
}
Status BaseTablet::fetch_value_by_rowids(RowsetSharedPtr input_rowset, uint32_t segid,
const std::vector<uint32_t>& rowids,
const TabletColumn& tablet_column,
vectorized::MutableColumnPtr& dst) {
MonotonicStopWatch watch;
watch.start();
Defer _defer([&]() {
LOG_EVERY_N(INFO, 500) << "fetch_value_by_rowids, cost(us):" << watch.elapsed_time() / 1000
<< ", row_batch_size:" << rowids.size();
});
// read row data
BetaRowsetSharedPtr rowset = std::static_pointer_cast<BetaRowset>(input_rowset);
CHECK(rowset);
SegmentCacheHandle segment_cache_handle;
std::unique_ptr<segment_v2::ColumnIterator> column_iterator;
OlapReaderStatistics stats;
RETURN_IF_ERROR(_get_segment_column_iterator(rowset, segid, tablet_column,
&segment_cache_handle, &column_iterator, &stats));
RETURN_IF_ERROR(column_iterator->read_by_rowids(rowids.data(), rowids.size(), dst));
return Status::OK();
}
const signed char* BaseTablet::get_delete_sign_column_data(vectorized::Block& block,
size_t rows_at_least) {
if (const vectorized::ColumnWithTypeAndName* delete_sign_column =
block.try_get_by_name(DELETE_SIGN);
delete_sign_column != nullptr) {
const auto& delete_sign_col =
reinterpret_cast<const vectorized::ColumnInt8&>(*(delete_sign_column->column));
if (delete_sign_col.size() >= rows_at_least) {
return delete_sign_col.get_data().data();
}
}
return nullptr;
};
Status BaseTablet::generate_default_value_block(const TabletSchema& schema,
const std::vector<uint32_t>& cids,
const std::vector<std::string>& default_values,
const vectorized::Block& ref_block,
vectorized::Block& default_value_block) {
auto mutable_default_value_columns = default_value_block.mutate_columns();
for (auto i = 0; i < cids.size(); ++i) {
const auto& column = schema.column(cids[i]);
if (column.has_default_value()) {
const auto& default_value = default_values[i];
vectorized::ReadBuffer rb(const_cast<char*>(default_value.c_str()),
default_value.size());
RETURN_IF_ERROR(ref_block.get_by_position(i).type->from_string(
rb, mutable_default_value_columns[i].get()));
}
}
default_value_block.set_columns(std::move(mutable_default_value_columns));
return Status::OK();
}
Status BaseTablet::generate_new_block_for_partial_update(
TabletSchemaSPtr rowset_schema, const PartialUpdateInfo* partial_update_info,
const PartialUpdateReadPlan& read_plan_ori, const PartialUpdateReadPlan& read_plan_update,
const std::map<RowsetId, RowsetSharedPtr>& rsid_to_rowset,
vectorized::Block* output_block) {
// do partial update related works
// 1. read columns by read plan
// 2. generate new block
// 3. write a new segment and modify rowset meta
// 4. mark current keys deleted
CHECK(output_block);
auto full_mutable_columns = output_block->mutate_columns();
const auto& missing_cids = partial_update_info->missing_cids;
const auto& update_cids = partial_update_info->update_cids;
auto old_block = rowset_schema->create_block_by_cids(missing_cids);
auto update_block = rowset_schema->create_block_by_cids(update_cids);
// rowid in the final block(start from 0, increase continuously) -> rowid to read in update_block
std::map<uint32_t, uint32_t> read_index_update;
// read current rowset first, if a row in the current rowset has delete sign mark
// we don't need to read values from old block
RETURN_IF_ERROR(read_plan_update.read_columns_by_plan(
*rowset_schema, update_cids, rsid_to_rowset, update_block, &read_index_update));
size_t update_rows = read_index_update.size();
for (auto i = 0; i < update_cids.size(); ++i) {
for (auto idx = 0; idx < update_rows; ++idx) {
full_mutable_columns[update_cids[i]]->insert_from(
*update_block.get_columns_with_type_and_name()[i].column.get(),
read_index_update[idx]);
}
}
// if there is sequence column in the table, we need to read the sequence column,
// otherwise it may cause the merge-on-read based compaction policy to produce incorrect results
const auto* __restrict new_block_delete_signs =
rowset_schema->has_sequence_col()
? nullptr
: get_delete_sign_column_data(update_block, update_rows);
// rowid in the final block(start from 0, increase, may not continuous becasue we skip to read some rows) -> rowid to read in old_block
std::map<uint32_t, uint32_t> read_index_old;
RETURN_IF_ERROR(read_plan_ori.read_columns_by_plan(*rowset_schema, missing_cids, rsid_to_rowset,
old_block, &read_index_old,
new_block_delete_signs));
size_t old_rows = read_index_old.size();
const auto* __restrict old_block_delete_signs =
get_delete_sign_column_data(old_block, old_rows);
// build default value block
auto default_value_block = old_block.clone_empty();
if (old_block_delete_signs != nullptr || new_block_delete_signs != nullptr) {
RETURN_IF_ERROR(BaseTablet::generate_default_value_block(
*rowset_schema, missing_cids, partial_update_info->default_values, old_block,
default_value_block));
}
auto mutable_default_value_columns = default_value_block.mutate_columns();
CHECK(update_rows >= old_rows);
// build full block
for (auto i = 0; i < missing_cids.size(); ++i) {
const auto& rs_column = rowset_schema->column(missing_cids[i]);
auto& mutable_column = full_mutable_columns[missing_cids[i]];
for (auto idx = 0; idx < update_rows; ++idx) {
// There are two cases we don't need to read values from old data:
// 1. if the conflicting new row's delete sign is marked, which means the value columns
// of the row will not be read. So we don't need to read the missing values from the previous rows.
// 2. if the conflicting old row's delete sign is marked, which means that the key is not exist now,
// we should not read old values from the deleted data, and should use default value instead.
// NOTE: since now we are in the publishing phase, all data is commited
// before, even the `strict_mode` is true (which requires partial update
// load job can't insert new keys), this "new" key MUST be written into
// the new generated segment file.
if (new_block_delete_signs != nullptr && new_block_delete_signs[idx]) {
mutable_column->insert_default();
} else if (old_block_delete_signs != nullptr &&
old_block_delete_signs[read_index_old[idx]] != 0) {
if (rs_column.has_default_value()) {
mutable_column->insert_from(*mutable_default_value_columns[i].get(), 0);
} else if (rs_column.is_nullable()) {
assert_cast<vectorized::ColumnNullable*, TypeCheckOnRelease::DISABLE>(
mutable_column.get())
->insert_null_elements(1);
} else {
mutable_column->insert_default();
}
} else {
mutable_column->insert_from(
*old_block.get_columns_with_type_and_name()[i].column.get(),
read_index_old[idx]);
}