/
Aggregator.cpp
2611 lines (2128 loc) · 90.5 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 <iomanip>
#include <thread>
#include <future>
#include <Poco/Version.h>
#include <Poco/Util/Application.h>
#include <Common/Stopwatch.h>
#include <Common/setThreadName.h>
#include <Common/formatReadable.h>
#include <DataTypes/DataTypeAggregateFunction.h>
#include <DataTypes/DataTypeNullable.h>
#include <DataTypes/DataTypeLowCardinality.h>
#include <Columns/ColumnsNumber.h>
#include <Columns/ColumnArray.h>
#include <Columns/ColumnTuple.h>
#include <Columns/ColumnLowCardinality.h>
#include <DataStreams/IBlockInputStream.h>
#include <DataStreams/NativeBlockOutputStream.h>
#include <DataStreams/NullBlockInputStream.h>
#include <DataStreams/materializeBlock.h>
#include <IO/WriteBufferFromFile.h>
#include <Compression/CompressedWriteBuffer.h>
#include <Interpreters/Aggregator.h>
#include <Common/ClickHouseRevision.h>
#include <Common/MemoryTracker.h>
#include <Common/CurrentThread.h>
#include <Common/typeid_cast.h>
#include <Common/assert_cast.h>
#include <common/demangle.h>
#include <AggregateFunctions/AggregateFunctionArray.h>
#include <AggregateFunctions/AggregateFunctionState.h>
#include <AggregateFunctions/AggregateFunctionResample.h>
#include <Disks/StoragePolicy.h>
#include <IO/Operators.h>
namespace ProfileEvents
{
extern const Event ExternalAggregationWritePart;
extern const Event ExternalAggregationCompressedBytes;
extern const Event ExternalAggregationUncompressedBytes;
}
namespace CurrentMetrics
{
extern const Metric QueryThread;
}
namespace DB
{
namespace ErrorCodes
{
extern const int UNKNOWN_AGGREGATED_DATA_VARIANT;
extern const int NOT_ENOUGH_SPACE;
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;
}
AggregatedDataVariants::~AggregatedDataVariants()
{
if (aggregator && !aggregator->all_aggregates_has_trivial_destructor)
{
try
{
aggregator->destroyAllAggregateStates(*this);
}
catch (...)
{
tryLogCurrentException(__PRETTY_FUNCTION__);
}
}
}
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("Wrong data variant passed.", ErrorCodes::LOGICAL_ERROR);
}
}
Block Aggregator::getHeader(bool final) const
{
return params.getHeader(final);
}
Block Aggregator::Params::getHeader(
const Block & src_header,
const Block & intermediate_header,
const ColumnNumbers & keys,
const AggregateDescriptions & aggregates,
bool final)
{
Block res;
if (src_header)
{
for (const auto & key : keys)
res.insert(src_header.safeGetByPosition(key).cloneEmpty());
for (const auto & aggregate : aggregates)
{
size_t arguments_size = aggregate.arguments.size();
DataTypes argument_types(arguments_size);
for (size_t j = 0; j < arguments_size; ++j)
argument_types[j] = src_header.safeGetByPosition(aggregate.arguments[j]).type;
DataTypePtr type;
if (final)
type = aggregate.function->getReturnType();
else
type = std::make_shared<DataTypeAggregateFunction>(aggregate.function, argument_types, aggregate.parameters);
res.insert({ type, aggregate.column_name });
}
}
else if (intermediate_header)
{
res = intermediate_header.cloneEmpty();
if (final)
{
for (const auto & aggregate : aggregates)
{
auto & elem = res.getByName(aggregate.column_name);
elem.type = aggregate.function->getReturnType();
elem.column = elem.type->createColumn();
}
}
}
return materializeBlock(res);
}
void Aggregator::Params::explain(WriteBuffer & out, size_t indent) const
{
Strings res;
const auto & header = src_header ? src_header
: intermediate_header;
String prefix(indent, ' ');
{
/// Dump keys.
out << prefix << "Keys: ";
bool first = true;
for (auto key : keys)
{
if (!first)
out << ", ";
first = false;
if (key >= header.columns())
out << "unknown position " << key;
else
out << header.getByPosition(key).name;
}
out << '\n';
}
if (!aggregates.empty())
{
out << prefix << "Aggregates:\n";
for (const auto & aggregate : aggregates)
aggregate.explain(out, indent + 4);
}
}
Aggregator::Aggregator(const Params & params_)
: params(params_),
isCancelled([]() { return false; })
{
/// 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("Logical error: alignOfData is not 2^N", ErrorCodes::LOGICAL_ERROR);
/// 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);
}
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 & pos : params.keys)
{
DataTypePtr type = (params.src_header ? params.src_header : params.intermediate_header).safeGetByPosition(pos).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)
{
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 == 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;
throw Exception("Logical error: numeric column has sizeOfField not in 1, 2, 4, 8, 16.", ErrorCodes::LOGICAL_ERROR);
}
/// 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;
}
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;
}
return AggregatedDataVariants::Type::serialized;
}
void Aggregator::createAggregateStates(AggregateDataPtr & aggregate_data) const
{
for (size_t j = 0; j < params.aggregates_size; ++j)
{
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)
aggregate_functions[rollback_j]->destroy(aggregate_data + offsets_of_aggregate_states[rollback_j]);
throw;
}
}
}
/** It's interesting - if you remove `noinline`, then gcc for some reason will inline this function, and the performance decreases (~ 10%).
* (Probably because after the inline of this function, more internal functions no longer be inlined.)
* Inline does not make sense, since the inner loop is entirely inside this function.
*/
template <typename Method>
void NO_INLINE Aggregator::executeImpl(
Method & method,
Arena * aggregates_pool,
size_t rows,
ColumnRawPtrs & key_columns,
AggregateFunctionInstruction * aggregate_instructions,
bool no_more_keys,
AggregateDataPtr overflow_row) const
{
typename Method::State state(key_columns, key_sizes, aggregation_state_cache);
if (!no_more_keys)
executeImplBatch(method, state, aggregates_pool, rows, aggregate_instructions);
else
executeImplCase<true>(method, state, aggregates_pool, rows, aggregate_instructions, overflow_row);
}
template <bool no_more_keys, typename Method>
void NO_INLINE Aggregator::executeImplCase(
Method & method,
typename Method::State & state,
Arena * aggregates_pool,
size_t rows,
AggregateFunctionInstruction * aggregate_instructions,
AggregateDataPtr overflow_row) const
{
/// NOTE When editing this code, also pay attention to SpecializedAggregator.h.
/// For all rows.
for (size_t i = 0; i < rows; ++i)
{
AggregateDataPtr aggregate_data = nullptr;
if constexpr (!no_more_keys) /// Insert.
{
auto emplace_result = state.emplaceKey(method.data, i, *aggregates_pool);
/// If a new key is inserted, initialize the states of the aggregate functions, and possibly something related to the key.
if (emplace_result.isInserted())
{
/// exception-safety - if you can not allocate memory or create states, then destructors will not be called.
emplace_result.setMapped(nullptr);
aggregate_data = aggregates_pool->alignedAlloc(total_size_of_aggregate_states, align_aggregate_states);
createAggregateStates(aggregate_data);
emplace_result.setMapped(aggregate_data);
}
else
aggregate_data = emplace_result.getMapped();
}
else
{
/// Add only if the key already exists.
auto find_result = state.findKey(method.data, i, *aggregates_pool);
if (find_result.isFound())
aggregate_data = find_result.getMapped();
}
/// aggregate_date == nullptr means that the new key did not fit in the hash table because of no_more_keys.
/// If the key does not fit, and the data does not need to be aggregated in a separate row, then there's nothing to do.
if (!aggregate_data && !overflow_row)
continue;
AggregateDataPtr value = aggregate_data ? aggregate_data : overflow_row;
/// Add values to the aggregate functions.
for (AggregateFunctionInstruction * inst = aggregate_instructions; inst->that; ++inst)
(*inst->func)(inst->that, value + inst->state_offset, inst->arguments, i, aggregates_pool);
}
}
template <typename Method>
void NO_INLINE Aggregator::executeImplBatch(
Method & method,
typename Method::State & state,
Arena * aggregates_pool,
size_t rows,
AggregateFunctionInstruction * aggregate_instructions) const
{
/// Optimization for special case when there are no aggregate functions.
if (params.aggregates_size == 0)
{
/// For all rows.
AggregateDataPtr place = aggregates_pool->alloc(0);
for (size_t i = 0; i < rows; ++i)
state.emplaceKey(method.data, i, *aggregates_pool).setMapped(place);
return;
}
/// Optimization for special case when aggregating by 8bit key.
if constexpr (std::is_same_v<Method, typename decltype(AggregatedDataVariants::key8)::element_type>)
{
/// We use another method if there are aggregate functions with -Array combinator.
bool has_arrays = false;
for (AggregateFunctionInstruction * inst = aggregate_instructions; inst->that; ++inst)
{
if (inst->offsets)
{
has_arrays = true;
break;
}
}
if (!has_arrays)
{
for (AggregateFunctionInstruction * inst = aggregate_instructions; inst->that; ++inst)
{
inst->batch_that->addBatchLookupTable8(
rows,
reinterpret_cast<AggregateDataPtr *>(method.data.data()),
inst->state_offset,
[&](AggregateDataPtr & aggregate_data)
{
aggregate_data = aggregates_pool->alignedAlloc(total_size_of_aggregate_states, align_aggregate_states);
createAggregateStates(aggregate_data);
},
state.getKeyData(),
inst->batch_arguments,
aggregates_pool);
}
return;
}
}
/// Generic case.
PODArray<AggregateDataPtr> places(rows);
/// For all rows.
for (size_t i = 0; i < rows; ++i)
{
AggregateDataPtr aggregate_data = nullptr;
auto emplace_result = state.emplaceKey(method.data, i, *aggregates_pool);
/// If a new key is inserted, initialize the states of the aggregate functions, and possibly something related to the key.
if (emplace_result.isInserted())
{
/// exception-safety - if you can not allocate memory or create states, then destructors will not be called.
emplace_result.setMapped(nullptr);
aggregate_data = aggregates_pool->alignedAlloc(total_size_of_aggregate_states, align_aggregate_states);
createAggregateStates(aggregate_data);
emplace_result.setMapped(aggregate_data);
}
else
aggregate_data = emplace_result.getMapped();
places[i] = aggregate_data;
assert(places[i] != nullptr);
}
/// Add values to the aggregate functions.
for (AggregateFunctionInstruction * inst = aggregate_instructions; inst->that; ++inst)
{
if (inst->offsets)
inst->batch_that->addBatchArray(rows, places.data(), inst->state_offset, inst->batch_arguments, inst->offsets, aggregates_pool);
else
inst->batch_that->addBatch(rows, places.data(), inst->state_offset, inst->batch_arguments, aggregates_pool);
}
}
void NO_INLINE Aggregator::executeWithoutKeyImpl(
AggregatedDataWithoutKey & res,
size_t rows,
AggregateFunctionInstruction * aggregate_instructions,
Arena * arena)
{
/// Adding values
for (AggregateFunctionInstruction * inst = aggregate_instructions; inst->that; ++inst)
{
if (inst->offsets)
inst->batch_that->addBatchSinglePlace(
inst->offsets[static_cast<ssize_t>(rows - 1)], res + inst->state_offset, inst->batch_arguments, arena);
else
inst->batch_that->addBatchSinglePlace(rows, res + inst->state_offset, inst->batch_arguments, arena);
}
}
void NO_INLINE Aggregator::executeOnIntervalWithoutKeyImpl(
AggregatedDataWithoutKey & res,
size_t row_begin,
size_t row_end,
AggregateFunctionInstruction * aggregate_instructions,
Arena * arena)
{
/// Adding values
for (AggregateFunctionInstruction * inst = aggregate_instructions; inst->that; ++inst)
{
if (inst->offsets)
inst->batch_that->addBatchSinglePlaceFromInterval(inst->offsets[row_begin], inst->offsets[row_end - 1], res + inst->state_offset, inst->batch_arguments, arena);
else
inst->batch_that->addBatchSinglePlaceFromInterval(row_begin, row_end, res + inst->state_offset, inst->batch_arguments, arena);
}
}
void Aggregator::prepareAggregateInstructions(Columns columns, AggregateColumns & aggregate_columns, Columns & materialized_columns,
AggregateFunctionInstructions & aggregate_functions_instructions, NestedColumnsHolder & nested_columns_holder)
{
for (size_t i = 0; i < params.aggregates_size; ++i)
aggregate_columns[i].resize(params.aggregates[i].arguments.size());
aggregate_functions_instructions.resize(params.aggregates_size + 1);
aggregate_functions_instructions[params.aggregates_size].that = nullptr;
for (size_t i = 0; i < params.aggregates_size; ++i)
{
for (size_t j = 0; j < aggregate_columns[i].size(); ++j)
{
materialized_columns.push_back(columns.at(params.aggregates[i].arguments[j])->convertToFullColumnIfConst());
aggregate_columns[i][j] = materialized_columns.back().get();
auto column_no_lc = recursiveRemoveLowCardinality(aggregate_columns[i][j]->getPtr());
if (column_no_lc.get() != aggregate_columns[i][j])
{
materialized_columns.emplace_back(std::move(column_no_lc));
aggregate_columns[i][j] = materialized_columns.back().get();
}
}
aggregate_functions_instructions[i].arguments = aggregate_columns[i].data();
aggregate_functions_instructions[i].state_offset = offsets_of_aggregate_states[i];
auto * that = aggregate_functions[i];
/// Unnest consecutive trailing -State combinators
while (const auto * func = typeid_cast<const AggregateFunctionState *>(that))
that = func->getNestedFunction().get();
aggregate_functions_instructions[i].that = that;
aggregate_functions_instructions[i].func = that->getAddressOfAddFunction();
if (const auto * func = typeid_cast<const AggregateFunctionArray *>(that))
{
/// Unnest consecutive -State combinators before -Array
that = func->getNestedFunction().get();
while (const auto * nested_func = typeid_cast<const AggregateFunctionState *>(that))
that = nested_func->getNestedFunction().get();
auto [nested_columns, offsets] = checkAndGetNestedArrayOffset(aggregate_columns[i].data(), that->getArgumentTypes().size());
nested_columns_holder.push_back(std::move(nested_columns));
aggregate_functions_instructions[i].batch_arguments = nested_columns_holder.back().data();
aggregate_functions_instructions[i].offsets = offsets;
}
else
aggregate_functions_instructions[i].batch_arguments = aggregate_columns[i].data();
aggregate_functions_instructions[i].batch_that = that;
}
}
bool Aggregator::executeOnBlock(const Block & block, AggregatedDataVariants & result,
ColumnRawPtrs & key_columns, AggregateColumns & aggregate_columns, bool & no_more_keys)
{
UInt64 num_rows = block.rows();
return executeOnBlock(block.getColumns(), num_rows, result, key_columns, aggregate_columns, no_more_keys);
}
bool Aggregator::executeOnBlock(Columns columns, UInt64 num_rows, AggregatedDataVariants & result,
ColumnRawPtrs & key_columns, AggregateColumns & aggregate_columns, bool & no_more_keys)
{
if (isCancelled())
return true;
/// `result` will destroy the states of aggregate functions in the destructor
result.aggregator = this;
/// How to perform the aggregation?
if (result.empty())
{
result.init(method_chosen);
result.keys_size = params.keys_size;
result.key_sizes = key_sizes;
LOG_TRACE(log, "Aggregation method: {}", result.getMethodName());
}
if (isCancelled())
return true;
/** Constant columns are not supported directly during aggregation.
* To make them work anyway, we materialize them.
*/
Columns materialized_columns;
/// Remember the columns we will work with
for (size_t i = 0; i < params.keys_size; ++i)
{
materialized_columns.push_back(columns.at(params.keys[i])->convertToFullColumnIfConst());
key_columns[i] = materialized_columns.back().get();
if (!result.isLowCardinality())
{
auto column_no_lc = recursiveRemoveLowCardinality(key_columns[i]->getPtr());
if (column_no_lc.get() != key_columns[i])
{
materialized_columns.emplace_back(std::move(column_no_lc));
key_columns[i] = materialized_columns.back().get();
}
}
}
NestedColumnsHolder nested_columns_holder;
AggregateFunctionInstructions aggregate_functions_instructions;
prepareAggregateInstructions(columns, aggregate_columns, materialized_columns, aggregate_functions_instructions, nested_columns_holder);
if (isCancelled())
return true;
if ((params.overflow_row || result.type == AggregatedDataVariants::Type::without_key) && !result.without_key)
{
AggregateDataPtr place = result.aggregates_pool->alignedAlloc(total_size_of_aggregate_states, align_aggregate_states);
createAggregateStates(place);
result.without_key = place;
}
/// We select one of the aggregation methods and call it.
/// For the case when there are no keys (all aggregate into one row).
if (result.type == AggregatedDataVariants::Type::without_key)
{
executeWithoutKeyImpl(result.without_key, num_rows, aggregate_functions_instructions.data(), result.aggregates_pool);
}
else
{
/// This is where data is written that does not fit in `max_rows_to_group_by` with `group_by_overflow_mode = any`.
AggregateDataPtr overflow_row_ptr = params.overflow_row ? result.without_key : nullptr;
#define M(NAME, IS_TWO_LEVEL) \
else if (result.type == AggregatedDataVariants::Type::NAME) \
executeImpl(*result.NAME, result.aggregates_pool, num_rows, key_columns, aggregate_functions_instructions.data(), \
no_more_keys, overflow_row_ptr);
if (false) {} // NOLINT
APPLY_FOR_AGGREGATED_VARIANTS(M)
#undef M
}
size_t result_size = result.sizeWithoutOverflowRow();
Int64 current_memory_usage = 0;
if (auto * memory_tracker_child = CurrentThread::getMemoryTracker())
if (auto * memory_tracker = memory_tracker_child->getParent())
current_memory_usage = memory_tracker->get();
/// Here all the results in the sum are taken into account, from different threads.
auto result_size_bytes = current_memory_usage - memory_usage_before_aggregation;
bool worth_convert_to_two_level
= (params.group_by_two_level_threshold && result_size >= params.group_by_two_level_threshold)
|| (params.group_by_two_level_threshold_bytes && result_size_bytes >= static_cast<Int64>(params.group_by_two_level_threshold_bytes));
/** Converting to a two-level data structure.
* It allows you to make, in the subsequent, an effective merge - either economical from memory or parallel.
*/
if (result.isConvertibleToTwoLevel() && worth_convert_to_two_level)
result.convertToTwoLevel();
/// Checking the constraints.
if (!checkLimits(result_size, no_more_keys))
return false;
/** Flush data to disk if too much RAM is consumed.
* Data can only be flushed to disk if a two-level aggregation structure is used.
*/
if (params.max_bytes_before_external_group_by
&& result.isTwoLevel()
&& current_memory_usage > static_cast<Int64>(params.max_bytes_before_external_group_by)
&& worth_convert_to_two_level)
{
size_t size = current_memory_usage + params.min_free_disk_space;
std::string tmp_path = params.tmp_volume->getDisk()->getPath();
// enoughSpaceInDirectory() is not enough to make it right, since
// another process (or another thread of aggregator) can consume all
// space.
//
// But true reservation (IVolume::reserve()) cannot be used here since
// current_memory_usage does not takes compression into account and
// will reserve way more that actually will be used.
//
// Hence let's do a simple check.
if (!enoughSpaceInDirectory(tmp_path, size))
throw Exception("Not enough space for external aggregation in " + tmp_path, ErrorCodes::NOT_ENOUGH_SPACE);
writeToTemporaryFile(result, tmp_path);
}
return true;
}
void Aggregator::writeToTemporaryFile(AggregatedDataVariants & data_variants, const String & tmp_path)
{
Stopwatch watch;
size_t rows = data_variants.size();
auto file = createTemporaryFile(tmp_path);
const std::string & path = file->path();
WriteBufferFromFile file_buf(path);
CompressedWriteBuffer compressed_buf(file_buf);
NativeBlockOutputStream block_out(compressed_buf, ClickHouseRevision::get(), getHeader(false));
LOG_DEBUG(log, "Writing part of aggregation data into temporary file {}.", path);
ProfileEvents::increment(ProfileEvents::ExternalAggregationWritePart);
/// Flush only two-level data and possibly overflow data.
#define M(NAME) \
else if (data_variants.type == AggregatedDataVariants::Type::NAME) \
writeToTemporaryFileImpl(data_variants, *data_variants.NAME, block_out);
if (false) {} // NOLINT
APPLY_FOR_VARIANTS_TWO_LEVEL(M)
#undef M
else
throw Exception("Unknown aggregated data variant.", ErrorCodes::UNKNOWN_AGGREGATED_DATA_VARIANT);
/// NOTE Instead of freeing up memory and creating new hash tables and arenas, you can re-use the old ones.
data_variants.init(data_variants.type);
data_variants.aggregates_pools = Arenas(1, std::make_shared<Arena>());
data_variants.aggregates_pool = data_variants.aggregates_pools.back().get();
data_variants.without_key = nullptr;
block_out.flush();
compressed_buf.next();
file_buf.next();
double elapsed_seconds = watch.elapsedSeconds();
double compressed_bytes = file_buf.count();
double uncompressed_bytes = compressed_buf.count();
{
std::lock_guard lock(temporary_files.mutex);
temporary_files.files.emplace_back(std::move(file));
temporary_files.sum_size_uncompressed += uncompressed_bytes;
temporary_files.sum_size_compressed += compressed_bytes;
}
ProfileEvents::increment(ProfileEvents::ExternalAggregationCompressedBytes, compressed_bytes);
ProfileEvents::increment(ProfileEvents::ExternalAggregationUncompressedBytes, uncompressed_bytes);
LOG_TRACE(log,
"Written part in {} sec., {} rows, {} uncompressed, {} compressed,"
" {} uncompressed bytes per row, {} compressed bytes per row, compression rate: {}"
" ({} rows/sec., {}/sec. uncompressed, {}/sec. compressed)",
elapsed_seconds,
rows,
ReadableSize(uncompressed_bytes),
ReadableSize(compressed_bytes),
uncompressed_bytes / rows,
compressed_bytes / rows,
uncompressed_bytes / compressed_bytes,
rows / elapsed_seconds,
ReadableSize(uncompressed_bytes / elapsed_seconds),
ReadableSize(compressed_bytes / elapsed_seconds));
}
void Aggregator::writeToTemporaryFile(AggregatedDataVariants & data_variants)
{
String tmp_path = params.tmp_volume->getDisk()->getPath();
return writeToTemporaryFile(data_variants, tmp_path);
}
template <typename Method>
Block Aggregator::convertOneBucketToBlock(
AggregatedDataVariants & data_variants,
Method & method,
bool final,
size_t bucket) const
{
Block block = prepareBlockAndFill(data_variants, final, method.data.impls[bucket].size(),
[bucket, &method, this] (
MutableColumns & key_columns,
AggregateColumnsData & aggregate_columns,
MutableColumns & final_aggregate_columns,
Arena * arena,
bool final_)
{
convertToBlockImpl(method, method.data.impls[bucket],
key_columns, aggregate_columns, final_aggregate_columns, arena, final_);
});
block.info.bucket_num = bucket;
return block;
}
Block Aggregator::mergeAndConvertOneBucketToBlock(
ManyAggregatedDataVariants & variants,
Arena * arena,
bool final,
size_t bucket,
std::atomic<bool> * is_cancelled) const
{
auto & merged_data = *variants[0];
auto method = merged_data.type;
Block block;
if (false) {} // NOLINT
#define M(NAME) \
else if (method == AggregatedDataVariants::Type::NAME) \
{ \
mergeBucketImpl<decltype(merged_data.NAME)::element_type>(variants, bucket, arena); \
if (is_cancelled && is_cancelled->load(std::memory_order_seq_cst)) \
return {}; \
block = convertOneBucketToBlock(merged_data, *merged_data.NAME, final, bucket); \
}
APPLY_FOR_VARIANTS_TWO_LEVEL(M)
#undef M
return block;
}
template <typename Method>
void Aggregator::writeToTemporaryFileImpl(
AggregatedDataVariants & data_variants,
Method & method,
IBlockOutputStream & out)
{
size_t max_temporary_block_size_rows = 0;
size_t max_temporary_block_size_bytes = 0;
auto update_max_sizes = [&](const Block & block)
{
size_t block_size_rows = block.rows();
size_t block_size_bytes = block.bytes();
if (block_size_rows > max_temporary_block_size_rows)
max_temporary_block_size_rows = block_size_rows;
if (block_size_bytes > max_temporary_block_size_bytes)
max_temporary_block_size_bytes = block_size_bytes;
};
for (size_t bucket = 0; bucket < Method::Data::NUM_BUCKETS; ++bucket)
{
Block block = convertOneBucketToBlock(data_variants, method, false, bucket);
out.write(block);
update_max_sizes(block);
}
if (params.overflow_row)
{
Block block = prepareBlockAndFillWithoutKey(data_variants, false, true);
out.write(block);
update_max_sizes(block);
}
/// Pass ownership of the aggregate functions states:
/// `data_variants` will not destroy them in the destructor, they are now owned by ColumnAggregateFunction objects.
data_variants.aggregator = nullptr;
LOG_TRACE(log, "Max size of temporary block: {} rows, {}.", max_temporary_block_size_rows, ReadableSize(max_temporary_block_size_bytes));
}