/
Aggregator.cpp
3488 lines (2919 loc) · 129 KB
/
Aggregator.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
#include <algorithm>
#include <future>
#include <numeric>
#include <Poco/Util/Application.h>
#ifdef OS_LINUX
# include <unistd.h>
#endif
#include <base/sort.h>
#include <DataTypes/DataTypeAggregateFunction.h>
#include <DataTypes/DataTypeNullable.h>
#include <DataTypes/DataTypeLowCardinality.h>
#include <Columns/ColumnArray.h>
#include <Columns/ColumnTuple.h>
#include <Columns/ColumnSparse.h>
#include <Formats/NativeWriter.h>
#include <IO/WriteBufferFromFile.h>
#include <Compression/CompressedWriteBuffer.h>
#include <Interpreters/Aggregator.h>
#include <AggregateFunctions/Combinators/AggregateFunctionArray.h>
#include <AggregateFunctions/Combinators/AggregateFunctionState.h>
#include <IO/Operators.h>
#include <Interpreters/JIT/compileFunction.h>
#include <Interpreters/JIT/CompiledExpressionCache.h>
#include <Core/ProtocolDefines.h>
#include <Disks/TemporaryFileOnDisk.h>
#include <Interpreters/TemporaryDataOnDisk.h>
#include <Common/Stopwatch.h>
#include <Common/setThreadName.h>
#include <Common/formatReadable.h>
#include <Common/logger_useful.h>
#include <Common/CacheBase.h>
#include <Common/MemoryTracker.h>
#include <Common/CurrentThread.h>
#include <Common/CurrentMetrics.h>
#include <Common/typeid_cast.h>
#include <Common/assert_cast.h>
#include <Common/JSONBuilder.h>
#include <Common/filesystemHelpers.h>
#include <Common/scope_guard_safe.h>
#include <Parsers/ASTSelectQuery.h>
#include <Interpreters/AggregationUtils.h>
namespace ProfileEvents
{
extern const Event ExternalAggregationWritePart;
extern const Event ExternalAggregationCompressedBytes;
extern const Event ExternalAggregationUncompressedBytes;
extern const Event ExternalProcessingCompressedBytesTotal;
extern const Event ExternalProcessingUncompressedBytesTotal;
extern const Event AggregationPreallocatedElementsInHashTables;
extern const Event AggregationHashTablesInitializedAsTwoLevel;
extern const Event OverflowThrow;
extern const Event OverflowBreak;
extern const Event OverflowAny;
}
namespace CurrentMetrics
{
extern const Metric TemporaryFilesForAggregation;
extern const Metric AggregatorThreads;
extern const Metric AggregatorThreadsActive;
}
namespace DB
{
namespace ErrorCodes
{
extern const int UNKNOWN_AGGREGATED_DATA_VARIANT;
extern const int TOO_MANY_ROWS;
extern const int EMPTY_DATA_PASSED;
extern const int CANNOT_MERGE_DIFFERENT_AGGREGATED_DATA_VARIANTS;
extern const int LOGICAL_ERROR;
}
}
namespace
{
/** Collects observed HashMap-s sizes to avoid redundant intermediate resizes.
*/
class HashTablesStatistics
{
public:
struct Entry
{
size_t sum_of_sizes; // used to determine if it's better to convert aggregation to two-level from the beginning
size_t median_size; // roughly the size we're going to preallocate on each thread
};
using Cache = DB::CacheBase<UInt64, Entry>;
using CachePtr = std::shared_ptr<Cache>;
using Params = DB::Aggregator::Params::StatsCollectingParams;
/// Collection and use of the statistics should be enabled.
std::optional<Entry> getSizeHint(const Params & params)
{
if (!params.isCollectionAndUseEnabled())
throw DB::Exception(DB::ErrorCodes::LOGICAL_ERROR, "Collection and use of the statistics should be enabled.");
std::lock_guard lock(mutex);
const auto cache = getHashTableStatsCache(params, lock);
if (const auto hint = cache->get(params.key))
{
LOG_TRACE(
&Poco::Logger::get("Aggregator"),
"An entry for key={} found in cache: sum_of_sizes={}, median_size={}",
params.key,
hint->sum_of_sizes,
hint->median_size);
return *hint;
}
return std::nullopt;
}
/// Collection and use of the statistics should be enabled.
void update(size_t sum_of_sizes, size_t median_size, const Params & params)
{
if (!params.isCollectionAndUseEnabled())
throw DB::Exception(DB::ErrorCodes::LOGICAL_ERROR, "Collection and use of the statistics should be enabled.");
std::lock_guard lock(mutex);
const auto cache = getHashTableStatsCache(params, lock);
const auto hint = cache->get(params.key);
// We'll maintain the maximum among all the observed values until the next prediction turns out to be too wrong.
if (!hint || sum_of_sizes < hint->sum_of_sizes / 2 || hint->sum_of_sizes < sum_of_sizes || median_size < hint->median_size / 2
|| hint->median_size < median_size)
{
LOG_TRACE(
&Poco::Logger::get("Aggregator"),
"Statistics updated for key={}: new sum_of_sizes={}, median_size={}",
params.key,
sum_of_sizes,
median_size);
cache->set(params.key, std::make_shared<Entry>(Entry{.sum_of_sizes = sum_of_sizes, .median_size = median_size}));
}
}
std::optional<DB::HashTablesCacheStatistics> getCacheStats() const
{
std::lock_guard lock(mutex);
if (hash_table_stats)
{
size_t hits = 0, misses = 0;
hash_table_stats->getStats(hits, misses);
return DB::HashTablesCacheStatistics{.entries = hash_table_stats->count(), .hits = hits, .misses = misses};
}
return std::nullopt;
}
static size_t calculateCacheKey(const DB::ASTPtr & select_query)
{
if (!select_query)
throw DB::Exception(DB::ErrorCodes::LOGICAL_ERROR, "Query ptr cannot be null");
const auto & select = select_query->as<DB::ASTSelectQuery &>();
// It may happen in some corner cases like `select 1 as num group by num`.
if (!select.tables())
return 0;
SipHash hash;
hash.update(select.tables()->getTreeHash());
if (const auto where = select.where())
hash.update(where->getTreeHash());
if (const auto group_by = select.groupBy())
hash.update(group_by->getTreeHash());
return hash.get64();
}
private:
CachePtr getHashTableStatsCache(const Params & params, const std::lock_guard<std::mutex> &)
{
if (!hash_table_stats || hash_table_stats->maxSizeInBytes() != params.max_entries_for_hash_table_stats)
hash_table_stats = std::make_shared<Cache>(params.max_entries_for_hash_table_stats);
return hash_table_stats;
}
mutable std::mutex mutex;
CachePtr hash_table_stats;
};
HashTablesStatistics & getHashTablesStatistics()
{
static HashTablesStatistics hash_tables_stats;
return hash_tables_stats;
}
bool worthConvertToTwoLevel(
size_t group_by_two_level_threshold, size_t result_size, size_t group_by_two_level_threshold_bytes, auto result_size_bytes)
{
// params.group_by_two_level_threshold will be equal to 0 if we have only one thread to execute aggregation (refer to AggregatingStep::transformPipeline).
return (group_by_two_level_threshold && result_size >= group_by_two_level_threshold)
|| (group_by_two_level_threshold_bytes && result_size_bytes >= static_cast<Int64>(group_by_two_level_threshold_bytes));
}
DB::AggregatedDataVariants::Type convertToTwoLevelTypeIfPossible(DB::AggregatedDataVariants::Type type)
{
using Type = DB::AggregatedDataVariants::Type;
switch (type)
{
#define M(NAME) \
case Type::NAME: \
return Type::NAME##_two_level;
APPLY_FOR_VARIANTS_CONVERTIBLE_TO_TWO_LEVEL(M)
#undef M
default:
return type;
}
UNREACHABLE();
}
void initDataVariantsWithSizeHint(
DB::AggregatedDataVariants & result, DB::AggregatedDataVariants::Type method_chosen, const DB::Aggregator::Params & params)
{
const auto & stats_collecting_params = params.stats_collecting_params;
if (stats_collecting_params.isCollectionAndUseEnabled())
{
if (auto hint = getHashTablesStatistics().getSizeHint(stats_collecting_params))
{
const auto max_threads = params.group_by_two_level_threshold != 0 ? std::max(params.max_threads, 1ul) : 1;
const auto lower_limit = hint->sum_of_sizes / max_threads;
const auto upper_limit = stats_collecting_params.max_size_to_preallocate_for_aggregation / max_threads;
if (hint->median_size > upper_limit)
{
/// Since we cannot afford to preallocate as much as we want, we will likely need to do resize anyway.
/// But we will also work with the big (i.e. not so cache friendly) HT from the beginning which may result in a slight slowdown.
/// So let's just do nothing.
LOG_TRACE(
&Poco::Logger::get("Aggregator"),
"No space were preallocated in hash tables because 'max_size_to_preallocate_for_aggregation' has too small value: {}, "
"should be at least {}",
stats_collecting_params.max_size_to_preallocate_for_aggregation,
hint->median_size * max_threads);
}
/// https://github.com/ClickHouse/ClickHouse/issues/44402#issuecomment-1359920703
else if ((max_threads > 1 && hint->sum_of_sizes > 100'000) || hint->sum_of_sizes > 500'000)
{
const auto adjusted = std::max(lower_limit, hint->median_size);
if (worthConvertToTwoLevel(
params.group_by_two_level_threshold,
hint->sum_of_sizes,
/*group_by_two_level_threshold_bytes*/ 0,
/*result_size_bytes*/ 0))
method_chosen = convertToTwoLevelTypeIfPossible(method_chosen);
result.init(method_chosen, adjusted);
ProfileEvents::increment(ProfileEvents::AggregationHashTablesInitializedAsTwoLevel, result.isTwoLevel());
return;
}
}
}
result.init(method_chosen);
}
/// Collection and use of the statistics should be enabled.
void updateStatistics(const DB::ManyAggregatedDataVariants & data_variants, const DB::Aggregator::Params::StatsCollectingParams & params)
{
if (!params.isCollectionAndUseEnabled())
throw DB::Exception(DB::ErrorCodes::LOGICAL_ERROR, "Collection and use of the statistics should be enabled.");
std::vector<size_t> sizes(data_variants.size());
for (size_t i = 0; i < data_variants.size(); ++i)
sizes[i] = data_variants[i]->size();
const auto median_size = sizes.begin() + sizes.size() / 2; // not precisely though...
std::nth_element(sizes.begin(), median_size, sizes.end());
const auto sum_of_sizes = std::accumulate(sizes.begin(), sizes.end(), 0ull);
getHashTablesStatistics().update(sum_of_sizes, *median_size, params);
}
// The std::is_constructible trait isn't suitable here because some classes have template constructors with semantics different from providing size hints.
// Also string hash table variants are not supported due to the fact that both local perf tests and tests in CI showed slowdowns for them.
template <typename...>
struct HasConstructorOfNumberOfElements : std::false_type
{
};
template <typename... Ts>
struct HasConstructorOfNumberOfElements<HashMapTable<Ts...>> : std::true_type
{
};
template <typename Key, typename Cell, typename Hash, typename Grower, typename Allocator, template <typename...> typename ImplTable>
struct HasConstructorOfNumberOfElements<TwoLevelHashMapTable<Key, Cell, Hash, Grower, Allocator, ImplTable>> : std::true_type
{
};
template <typename... Ts>
struct HasConstructorOfNumberOfElements<HashTable<Ts...>> : std::true_type
{
};
template <typename... Ts>
struct HasConstructorOfNumberOfElements<TwoLevelHashTable<Ts...>> : std::true_type
{
};
template <template <typename> typename Method, typename Base>
struct HasConstructorOfNumberOfElements<Method<Base>> : HasConstructorOfNumberOfElements<Base>
{
};
template <typename Method>
auto constructWithReserveIfPossible(size_t size_hint)
{
if constexpr (HasConstructorOfNumberOfElements<typename Method::Data>::value)
{
ProfileEvents::increment(ProfileEvents::AggregationPreallocatedElementsInHashTables, size_hint);
return std::make_unique<Method>(size_hint);
}
else
return std::make_unique<Method>();
}
DB::ColumnNumbers calculateKeysPositions(const DB::Block & header, const DB::Aggregator::Params & params)
{
DB::ColumnNumbers keys_positions(params.keys_size);
for (size_t i = 0; i < params.keys_size; ++i)
keys_positions[i] = header.getPositionByName(params.keys[i]);
return keys_positions;
}
template <typename HashTable, typename KeyHolder>
concept HasPrefetchMemberFunc = requires
{
{std::declval<HashTable>().prefetch(std::declval<KeyHolder>())};
};
size_t getMinBytesForPrefetch()
{
size_t l2_size = 0;
#if defined(OS_LINUX) && defined(_SC_LEVEL2_CACHE_SIZE)
if (auto ret = sysconf(_SC_LEVEL2_CACHE_SIZE); ret != -1)
l2_size = ret;
#endif
/// 256KB looks like a reasonable default L2 size. 4 is empirical constant.
return 4 * std::max<size_t>(l2_size, 256 * 1024);
}
}
namespace DB
{
AggregatedDataVariants::~AggregatedDataVariants()
{
if (aggregator && !aggregator->all_aggregates_has_trivial_destructor)
{
try
{
aggregator->destroyAllAggregateStates(*this);
}
catch (...)
{
tryLogCurrentException(__PRETTY_FUNCTION__);
}
}
}
std::optional<HashTablesCacheStatistics> getHashTablesCacheStatistics()
{
return getHashTablesStatistics().getCacheStats();
}
void AggregatedDataVariants::convertToTwoLevel()
{
if (aggregator)
LOG_TRACE(aggregator->log, "Converting aggregation data to two-level.");
switch (type)
{
#define M(NAME) \
case Type::NAME: \
NAME ## _two_level = std::make_unique<decltype(NAME ## _two_level)::element_type>(*(NAME)); \
(NAME).reset(); \
type = Type::NAME ## _two_level; \
break;
APPLY_FOR_VARIANTS_CONVERTIBLE_TO_TWO_LEVEL(M)
#undef M
default:
throw Exception(ErrorCodes::LOGICAL_ERROR, "Wrong data variant passed.");
}
}
void AggregatedDataVariants::init(Type type_, std::optional<size_t> size_hint)
{
switch (type_)
{
case Type::EMPTY:
case Type::without_key:
break;
#define M(NAME, IS_TWO_LEVEL) \
case Type::NAME: \
if (size_hint) \
(NAME) = constructWithReserveIfPossible<decltype(NAME)::element_type>(*size_hint); \
else \
(NAME) = std::make_unique<decltype(NAME)::element_type>(); \
break;
APPLY_FOR_AGGREGATED_VARIANTS(M)
#undef M
}
type = type_;
}
Aggregator::Params::StatsCollectingParams::StatsCollectingParams() = default;
Aggregator::Params::StatsCollectingParams::StatsCollectingParams(
const ASTPtr & select_query_,
bool collect_hash_table_stats_during_aggregation_,
size_t max_entries_for_hash_table_stats_,
size_t max_size_to_preallocate_for_aggregation_)
: key(collect_hash_table_stats_during_aggregation_ ? HashTablesStatistics::calculateCacheKey(select_query_) : 0)
, max_entries_for_hash_table_stats(max_entries_for_hash_table_stats_)
, max_size_to_preallocate_for_aggregation(max_size_to_preallocate_for_aggregation_)
{
}
Block Aggregator::getHeader(bool final) const
{
return params.getHeader(header, final);
}
Block Aggregator::Params::getHeader(
const Block & header, bool only_merge, const Names & keys, const AggregateDescriptions & aggregates, bool final)
{
Block res;
if (only_merge)
{
NameSet needed_columns(keys.begin(), keys.end());
for (const auto & aggregate : aggregates)
needed_columns.emplace(aggregate.column_name);
for (const auto & column : header)
{
if (needed_columns.contains(column.name))
res.insert(column.cloneEmpty());
}
if (final)
{
for (const auto & aggregate : aggregates)
{
auto & elem = res.getByName(aggregate.column_name);
elem.type = aggregate.function->getResultType();
elem.column = elem.type->createColumn();
}
}
}
else
{
for (const auto & key : keys)
res.insert(header.getByName(key).cloneEmpty());
for (const auto & aggregate : aggregates)
{
size_t arguments_size = aggregate.argument_names.size();
DataTypes argument_types(arguments_size);
for (size_t j = 0; j < arguments_size; ++j)
argument_types[j] = header.getByName(aggregate.argument_names[j]).type;
DataTypePtr type;
if (final)
type = aggregate.function->getResultType();
else
type = std::make_shared<DataTypeAggregateFunction>(aggregate.function, argument_types, aggregate.parameters);
res.insert({ type, aggregate.column_name });
}
}
return materializeBlock(res);
}
ColumnRawPtrs Aggregator::Params::makeRawKeyColumns(const Block & block) const
{
ColumnRawPtrs key_columns(keys_size);
for (size_t i = 0; i < keys_size; ++i)
key_columns[i] = block.safeGetByPosition(i).column.get();
return key_columns;
}
Aggregator::AggregateColumnsConstData Aggregator::Params::makeAggregateColumnsData(const Block & block) const
{
AggregateColumnsConstData aggregate_columns(aggregates_size);
for (size_t i = 0; i < aggregates_size; ++i)
{
const auto & aggregate_column_name = aggregates[i].column_name;
aggregate_columns[i] = &typeid_cast<const ColumnAggregateFunction &>(*block.getByName(aggregate_column_name).column).getData();
}
return aggregate_columns;
}
void Aggregator::Params::explain(WriteBuffer & out, size_t indent) const
{
String prefix(indent, ' ');
{
/// Dump keys.
out << prefix << "Keys: ";
bool first = true;
for (const auto & key : keys)
{
if (!first)
out << ", ";
first = false;
out << key;
}
out << '\n';
}
if (!aggregates.empty())
{
out << prefix << "Aggregates:\n";
for (const auto & aggregate : aggregates)
aggregate.explain(out, indent + 4);
}
}
void Aggregator::Params::explain(JSONBuilder::JSONMap & map) const
{
auto keys_array = std::make_unique<JSONBuilder::JSONArray>();
for (const auto & key : keys)
keys_array->add(key);
map.add("Keys", std::move(keys_array));
if (!aggregates.empty())
{
auto aggregates_array = std::make_unique<JSONBuilder::JSONArray>();
for (const auto & aggregate : aggregates)
{
auto aggregate_map = std::make_unique<JSONBuilder::JSONMap>();
aggregate.explain(*aggregate_map);
aggregates_array->add(std::move(aggregate_map));
}
map.add("Aggregates", std::move(aggregates_array));
}
}
#if USE_EMBEDDED_COMPILER
static CHJIT & getJITInstance()
{
static CHJIT jit;
return jit;
}
class CompiledAggregateFunctionsHolder final : public CompiledExpressionCacheEntry
{
public:
explicit CompiledAggregateFunctionsHolder(CompiledAggregateFunctions compiled_function_)
: CompiledExpressionCacheEntry(compiled_function_.compiled_module.size)
, compiled_aggregate_functions(compiled_function_)
{}
~CompiledAggregateFunctionsHolder() override
{
getJITInstance().deleteCompiledModule(compiled_aggregate_functions.compiled_module);
}
CompiledAggregateFunctions compiled_aggregate_functions;
};
#endif
Aggregator::Aggregator(const Block & header_, const Params & params_)
: header(header_)
, keys_positions(calculateKeysPositions(header, params_))
, params(params_)
, tmp_data(params.tmp_data_scope ? std::make_unique<TemporaryDataOnDisk>(params.tmp_data_scope, CurrentMetrics::TemporaryFilesForAggregation) : nullptr)
, min_bytes_for_prefetch(getMinBytesForPrefetch())
{
/// Use query-level memory tracker
if (auto * memory_tracker_child = CurrentThread::getMemoryTracker())
if (auto * memory_tracker = memory_tracker_child->getParent())
memory_usage_before_aggregation = memory_tracker->get();
aggregate_functions.resize(params.aggregates_size);
for (size_t i = 0; i < params.aggregates_size; ++i)
aggregate_functions[i] = params.aggregates[i].function.get();
/// Initialize sizes of aggregation states and its offsets.
offsets_of_aggregate_states.resize(params.aggregates_size);
total_size_of_aggregate_states = 0;
all_aggregates_has_trivial_destructor = true;
// aggregate_states will be aligned as below:
// |<-- state_1 -->|<-- pad_1 -->|<-- state_2 -->|<-- pad_2 -->| .....
//
// pad_N will be used to match alignment requirement for each next state.
// The address of state_1 is aligned based on maximum alignment requirements in states
for (size_t i = 0; i < params.aggregates_size; ++i)
{
offsets_of_aggregate_states[i] = total_size_of_aggregate_states;
total_size_of_aggregate_states += params.aggregates[i].function->sizeOfData();
// aggregate states are aligned based on maximum requirement
align_aggregate_states = std::max(align_aggregate_states, params.aggregates[i].function->alignOfData());
// If not the last aggregate_state, we need pad it so that next aggregate_state will be aligned.
if (i + 1 < params.aggregates_size)
{
size_t alignment_of_next_state = params.aggregates[i + 1].function->alignOfData();
if ((alignment_of_next_state & (alignment_of_next_state - 1)) != 0)
throw Exception(ErrorCodes::LOGICAL_ERROR, "Logical error: alignOfData is not 2^N");
/// Extend total_size to next alignment requirement
/// Add padding by rounding up 'total_size_of_aggregate_states' to be a multiplier of alignment_of_next_state.
total_size_of_aggregate_states = (total_size_of_aggregate_states + alignment_of_next_state - 1) / alignment_of_next_state * alignment_of_next_state;
}
if (!params.aggregates[i].function->hasTrivialDestructor())
all_aggregates_has_trivial_destructor = false;
}
method_chosen = chooseAggregationMethod();
HashMethodContext::Settings cache_settings;
cache_settings.max_threads = params.max_threads;
aggregation_state_cache = AggregatedDataVariants::createCache(method_chosen, cache_settings);
#if USE_EMBEDDED_COMPILER
compileAggregateFunctionsIfNeeded();
#endif
}
#if USE_EMBEDDED_COMPILER
void Aggregator::compileAggregateFunctionsIfNeeded()
{
static std::unordered_map<UInt128, UInt64, UInt128Hash> aggregate_functions_description_to_count;
static std::mutex mutex;
if (!params.compile_aggregate_expressions)
return;
std::vector<AggregateFunctionWithOffset> functions_to_compile;
String functions_description;
is_aggregate_function_compiled.resize(aggregate_functions.size());
/// Add values to the aggregate functions.
for (size_t i = 0; i < aggregate_functions.size(); ++i)
{
const auto * function = aggregate_functions[i];
size_t offset_of_aggregate_function = offsets_of_aggregate_states[i];
if (function->isCompilable())
{
AggregateFunctionWithOffset function_to_compile
{
.function = function,
.aggregate_data_offset = offset_of_aggregate_function
};
functions_to_compile.emplace_back(std::move(function_to_compile));
functions_description += function->getDescription();
functions_description += ' ';
functions_description += std::to_string(offset_of_aggregate_function);
functions_description += ' ';
}
is_aggregate_function_compiled[i] = function->isCompilable();
}
if (functions_to_compile.empty())
return;
SipHash aggregate_functions_description_hash;
aggregate_functions_description_hash.update(functions_description);
const auto aggregate_functions_description_hash_key = aggregate_functions_description_hash.get128();
{
std::lock_guard<std::mutex> lock(mutex);
if (aggregate_functions_description_to_count[aggregate_functions_description_hash_key]++ < params.min_count_to_compile_aggregate_expression)
return;
}
if (auto * compilation_cache = CompiledExpressionCacheFactory::instance().tryGetCache())
{
auto [compiled_function_cache_entry, _] = compilation_cache->getOrSet(aggregate_functions_description_hash_key, [&] ()
{
LOG_TRACE(log, "Compile expression {}", functions_description);
auto compiled_aggregate_functions = compileAggregateFunctions(getJITInstance(), functions_to_compile, functions_description);
return std::make_shared<CompiledAggregateFunctionsHolder>(std::move(compiled_aggregate_functions));
});
compiled_aggregate_functions_holder = std::static_pointer_cast<CompiledAggregateFunctionsHolder>(compiled_function_cache_entry);
}
else
{
LOG_TRACE(log, "Compile expression {}", functions_description);
auto compiled_aggregate_functions = compileAggregateFunctions(getJITInstance(), functions_to_compile, functions_description);
compiled_aggregate_functions_holder = std::make_shared<CompiledAggregateFunctionsHolder>(std::move(compiled_aggregate_functions));
}
}
#endif
AggregatedDataVariants::Type Aggregator::chooseAggregationMethod()
{
/// If no keys. All aggregating to single row.
if (params.keys_size == 0)
return AggregatedDataVariants::Type::without_key;
/// Check if at least one of the specified keys is nullable.
DataTypes types_removed_nullable;
types_removed_nullable.reserve(params.keys.size());
bool has_nullable_key = false;
bool has_low_cardinality = false;
for (const auto & key : params.keys)
{
DataTypePtr type = header.getByName(key).type;
if (type->lowCardinality())
{
has_low_cardinality = true;
type = removeLowCardinality(type);
}
if (type->isNullable())
{
has_nullable_key = true;
type = removeNullable(type);
}
types_removed_nullable.push_back(type);
}
/** Returns ordinary (not two-level) methods, because we start from them.
* Later, during aggregation process, data may be converted (partitioned) to two-level structure, if cardinality is high.
*/
size_t keys_bytes = 0;
size_t num_fixed_contiguous_keys = 0;
key_sizes.resize(params.keys_size);
for (size_t j = 0; j < params.keys_size; ++j)
{
if (types_removed_nullable[j]->isValueUnambiguouslyRepresentedInContiguousMemoryRegion())
{
if (types_removed_nullable[j]->isValueUnambiguouslyRepresentedInFixedSizeContiguousMemoryRegion())
{
++num_fixed_contiguous_keys;
key_sizes[j] = types_removed_nullable[j]->getSizeOfValueInMemory();
keys_bytes += key_sizes[j];
}
}
}
if (has_nullable_key)
{
/// Optimization for one key
if (params.keys_size == 1 && !has_low_cardinality)
{
if (types_removed_nullable[0]->isValueRepresentedByNumber())
{
size_t size_of_field = types_removed_nullable[0]->getSizeOfValueInMemory();
if (size_of_field == 1)
return AggregatedDataVariants::Type::nullable_key8;
if (size_of_field == 2)
return AggregatedDataVariants::Type::nullable_key16;
if (size_of_field == 4)
return AggregatedDataVariants::Type::nullable_key32;
if (size_of_field == 8)
return AggregatedDataVariants::Type::nullable_key64;
}
if (isFixedString(types_removed_nullable[0]))
{
return AggregatedDataVariants::Type::nullable_key_fixed_string;
}
if (isString(types_removed_nullable[0]))
{
return AggregatedDataVariants::Type::nullable_key_string;
}
}
if (params.keys_size == num_fixed_contiguous_keys && !has_low_cardinality)
{
/// Pack if possible all the keys along with information about which key values are nulls
/// into a fixed 16- or 32-byte blob.
if (std::tuple_size<KeysNullMap<UInt128>>::value + keys_bytes <= 16)
return AggregatedDataVariants::Type::nullable_keys128;
if (std::tuple_size<KeysNullMap<UInt256>>::value + keys_bytes <= 32)
return AggregatedDataVariants::Type::nullable_keys256;
}
if (has_low_cardinality && params.keys_size == 1)
{
if (types_removed_nullable[0]->isValueRepresentedByNumber())
{
size_t size_of_field = types_removed_nullable[0]->getSizeOfValueInMemory();
if (size_of_field == 1)
return AggregatedDataVariants::Type::low_cardinality_key8;
if (size_of_field == 2)
return AggregatedDataVariants::Type::low_cardinality_key16;
if (size_of_field == 4)
return AggregatedDataVariants::Type::low_cardinality_key32;
if (size_of_field == 8)
return AggregatedDataVariants::Type::low_cardinality_key64;
}
else if (isString(types_removed_nullable[0]))
return AggregatedDataVariants::Type::low_cardinality_key_string;
else if (isFixedString(types_removed_nullable[0]))
return AggregatedDataVariants::Type::low_cardinality_key_fixed_string;
}
/// Fallback case.
return AggregatedDataVariants::Type::serialized;
}
/// No key has been found to be nullable.
/// Single numeric key.
if (params.keys_size == 1 && types_removed_nullable[0]->isValueRepresentedByNumber())
{
size_t size_of_field = types_removed_nullable[0]->getSizeOfValueInMemory();
if (has_low_cardinality)
{
if (size_of_field == 1)
return AggregatedDataVariants::Type::low_cardinality_key8;
if (size_of_field == 2)
return AggregatedDataVariants::Type::low_cardinality_key16;
if (size_of_field == 4)
return AggregatedDataVariants::Type::low_cardinality_key32;
if (size_of_field == 8)
return AggregatedDataVariants::Type::low_cardinality_key64;
if (size_of_field == 16)
return AggregatedDataVariants::Type::low_cardinality_keys128;
if (size_of_field == 32)
return AggregatedDataVariants::Type::low_cardinality_keys256;
throw Exception(ErrorCodes::LOGICAL_ERROR, "Logical error: low cardinality numeric column has sizeOfField not in 1, 2, 4, 8, 16, 32.");
}
if (size_of_field == 1)
return AggregatedDataVariants::Type::key8;
if (size_of_field == 2)
return AggregatedDataVariants::Type::key16;
if (size_of_field == 4)
return AggregatedDataVariants::Type::key32;
if (size_of_field == 8)
return AggregatedDataVariants::Type::key64;
if (size_of_field == 16)
return AggregatedDataVariants::Type::keys128;
if (size_of_field == 32)
return AggregatedDataVariants::Type::keys256;
throw Exception(ErrorCodes::LOGICAL_ERROR, "Logical error: numeric column has sizeOfField not in 1, 2, 4, 8, 16, 32.");
}
if (params.keys_size == 1 && isFixedString(types_removed_nullable[0]))
{
if (has_low_cardinality)
return AggregatedDataVariants::Type::low_cardinality_key_fixed_string;
else
return AggregatedDataVariants::Type::key_fixed_string;
}
/// If all keys fits in N bits, will use hash table with all keys packed (placed contiguously) to single N-bit key.
if (params.keys_size == num_fixed_contiguous_keys)
{
if (has_low_cardinality)
{
if (keys_bytes <= 16)
return AggregatedDataVariants::Type::low_cardinality_keys128;
if (keys_bytes <= 32)
return AggregatedDataVariants::Type::low_cardinality_keys256;
}
if (keys_bytes <= 2)
return AggregatedDataVariants::Type::keys16;
if (keys_bytes <= 4)
return AggregatedDataVariants::Type::keys32;
if (keys_bytes <= 8)
return AggregatedDataVariants::Type::keys64;
if (keys_bytes <= 16)
return AggregatedDataVariants::Type::keys128;
if (keys_bytes <= 32)
return AggregatedDataVariants::Type::keys256;
}
/// If single string key - will use hash table with references to it. Strings itself are stored separately in Arena.
if (params.keys_size == 1 && isString(types_removed_nullable[0]))
{
if (has_low_cardinality)
return AggregatedDataVariants::Type::low_cardinality_key_string;
else
return AggregatedDataVariants::Type::key_string;
}
return AggregatedDataVariants::Type::serialized;
}
template <bool skip_compiled_aggregate_functions>
void Aggregator::createAggregateStates(AggregateDataPtr & aggregate_data) const
{
for (size_t j = 0; j < params.aggregates_size; ++j)
{
if constexpr (skip_compiled_aggregate_functions)
if (is_aggregate_function_compiled[j])
continue;
try
{
/** An exception may occur if there is a shortage of memory.
* In order that then everything is properly destroyed, we "roll back" some of the created states.
* The code is not very convenient.
*/
aggregate_functions[j]->create(aggregate_data + offsets_of_aggregate_states[j]);
}
catch (...)
{
for (size_t rollback_j = 0; rollback_j < j; ++rollback_j)
{
if constexpr (skip_compiled_aggregate_functions)
if (is_aggregate_function_compiled[j])
continue;
aggregate_functions[rollback_j]->destroy(aggregate_data + offsets_of_aggregate_states[rollback_j]);
}
throw;
}
}
}
bool Aggregator::hasSparseArguments(AggregateFunctionInstruction * aggregate_instructions)
{
for (auto * inst = aggregate_instructions; inst->that; ++inst)
if (inst->has_sparse_arguments)
return true;
return false;
}
void Aggregator::executeOnBlockSmall(
AggregatedDataVariants & result,
size_t row_begin,
size_t row_end,
ColumnRawPtrs & key_columns,
AggregateFunctionInstruction * aggregate_instructions) const
{
/// `result` will destroy the states of aggregate functions in the destructor
result.aggregator = this;
/// How to perform the aggregation?
if (result.empty())
{
if (method_chosen != AggregatedDataVariants::Type::without_key)
initDataVariantsWithSizeHint(result, method_chosen, params);
else
result.init(method_chosen);
result.keys_size = params.keys_size;
result.key_sizes = key_sizes;
}
executeImpl(result, row_begin, row_end, key_columns, aggregate_instructions);
CurrentMemoryTracker::check();
}
void Aggregator::mergeOnBlockSmall(
AggregatedDataVariants & result,
size_t row_begin,
size_t row_end,
const AggregateColumnsConstData & aggregate_columns_data,
const ColumnRawPtrs & key_columns) const
{
/// `result` will destroy the states of aggregate functions in the destructor
result.aggregator = this;
/// How to perform the aggregation?