-
Notifications
You must be signed in to change notification settings - Fork 977
/
pruning.rs
1764 lines (1569 loc) · 61.1 KB
/
pruning.rs
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.
//! This module contains code to prune "containers" of row groups
//! based on statistics prior to execution. This can lead to
//! significant performance improvements by avoiding the need
//! to evaluate a plan on entire containers (e.g. an entire file)
//!
//! For example, it is used to prune (skip) row groups while reading
//! parquet files if it can be determined from the predicate that
//! nothing in the row group can match.
//!
//! This code is currently specific to Parquet, but soon (TM), via
//! <https://github.com/apache/arrow-datafusion/issues/363> it will
//! be genericized.
use std::convert::TryFrom;
use std::{collections::HashSet, sync::Arc};
use crate::execution::context::ExecutionProps;
use crate::prelude::lit;
use crate::{
error::{DataFusionError, Result},
logical_plan::{Column, DFSchema, Expr, Operator},
physical_plan::{ColumnarValue, PhysicalExpr},
};
use arrow::{
array::{new_null_array, ArrayRef, BooleanArray},
datatypes::{DataType, Field, Schema, SchemaRef},
record_batch::RecordBatch,
};
use datafusion_expr::binary_expr;
use datafusion_expr::expr_rewriter::{ExprRewritable, ExprRewriter};
use datafusion_expr::utils::expr_to_columns;
use datafusion_physical_expr::create_physical_expr;
/// Interface to pass statistics information to [`PruningPredicate`]
///
/// Returns statistics for containers / files of data in Arrays.
///
/// For example, for the following three files with a single column
/// ```text
/// file1: column a: min=5, max=10
/// file2: column a: No stats
/// file2: column a: min=20, max=30
/// ```
///
/// PruningStatistics should return:
///
/// ```text
/// min_values("a") -> Some([5, Null, 20])
/// max_values("a") -> Some([20, Null, 30])
/// min_values("X") -> None
/// ```
pub trait PruningStatistics {
/// return the minimum values for the named column, if known.
/// Note: the returned array must contain `num_containers()` rows
fn min_values(&self, column: &Column) -> Option<ArrayRef>;
/// return the maximum values for the named column, if known.
/// Note: the returned array must contain `num_containers()` rows.
fn max_values(&self, column: &Column) -> Option<ArrayRef>;
/// return the number of containers (e.g. row groups) being
/// pruned with these statistics
fn num_containers(&self) -> usize;
/// return the number of null values for the named column as an
/// `Option<UInt64Array>`.
///
/// Note: the returned array must contain `num_containers()` rows.
fn null_counts(&self, column: &Column) -> Option<ArrayRef>;
}
/// Evaluates filter expressions on statistics in order to
/// prune data containers (e.g. parquet row group)
///
/// See [`PruningPredicate::try_new`] for more information.
#[derive(Debug, Clone)]
pub struct PruningPredicate {
/// The input schema against which the predicate will be evaluated
schema: SchemaRef,
/// Actual pruning predicate (rewritten in terms of column min/max statistics)
predicate_expr: Arc<dyn PhysicalExpr>,
/// The statistics required to evaluate this predicate
required_columns: RequiredStatColumns,
/// Logical predicate from which this predicate expr is derived (required for serialization)
logical_expr: Expr,
}
impl PruningPredicate {
/// Try to create a new instance of [`PruningPredicate`]
///
/// This will translate the provided `expr` filter expression into
/// a *pruning predicate*.
///
/// A pruning predicate is one that has been rewritten in terms of
/// the min and max values of column references and that evaluates
/// to FALSE if the filter predicate would evaluate FALSE *for
/// every row* whose values fell within the min / max ranges (aka
/// could be pruned).
///
/// The pruning predicate evaluates to TRUE or NULL
/// if the filter predicate *might* evaluate to TRUE for at least
/// one row whose vaules fell within the min/max ranges (in other
/// words they might pass the predicate)
///
/// For example, the filter expression `(column / 2) = 4` becomes
/// the pruning predicate
/// `(column_min / 2) <= 4 && 4 <= (column_max / 2))`
pub fn try_new(expr: Expr, schema: SchemaRef) -> Result<Self> {
// build predicate expression once
let mut required_columns = RequiredStatColumns::new();
let logical_predicate_expr =
build_predicate_expression(&expr, schema.as_ref(), &mut required_columns)?;
let stat_fields = required_columns
.iter()
.map(|(_, _, f)| f.clone())
.collect::<Vec<_>>();
let stat_schema = Schema::new(stat_fields);
let stat_dfschema = DFSchema::try_from(stat_schema.clone())?;
// TODO allow these properties to be passed in
let execution_props = ExecutionProps::new();
let predicate_expr = create_physical_expr(
&logical_predicate_expr,
&stat_dfschema,
&stat_schema,
&execution_props,
)?;
Ok(Self {
schema,
predicate_expr,
required_columns,
logical_expr: expr,
})
}
/// For each set of statistics, evalates the pruning predicate
/// and returns a `bool` with the following meaning for a
/// all rows whose values match the statistics:
///
/// `true`: There MAY be rows that match the predicate
///
/// `false`: There are no rows that could match the predicate
///
/// Note this function takes a slice of statistics as a parameter
/// to amortize the cost of the evaluation of the predicate
/// against a single record batch.
///
/// Note: the predicate passed to `prune` should be simplified as
/// much as possible (e.g. this pass doesn't handle some
/// expressions like `b = false`, but it does handle the
/// simplified version `b`. The predicates are simplified via the
/// ConstantFolding optimizer pass
pub fn prune<S: PruningStatistics>(&self, statistics: &S) -> Result<Vec<bool>> {
// build statistics record batch
let predicate_array =
build_statistics_record_batch(statistics, &self.required_columns)
.and_then(|statistics_batch| {
// execute predicate expression
self.predicate_expr.evaluate(&statistics_batch)
})
.and_then(|v| match v {
ColumnarValue::Array(array) => Ok(array),
ColumnarValue::Scalar(_) => Err(DataFusionError::Internal(
"predicate expression didn't return an array".to_string(),
)),
})?;
let predicate_array = predicate_array
.as_any()
.downcast_ref::<BooleanArray>()
.ok_or_else(|| {
DataFusionError::Internal(format!(
"Expected pruning predicate evaluation to be BooleanArray, \
but was {:?}",
predicate_array
))
})?;
// when the result of the predicate expression for a row group is null / undefined,
// e.g. due to missing statistics, this row group can't be filtered out,
// so replace with true
Ok(predicate_array
.into_iter()
.map(|x| x.unwrap_or(true))
.collect::<Vec<_>>())
}
/// Return a reference to the input schema
pub fn schema(&self) -> &SchemaRef {
&self.schema
}
/// Returns a reference to the logical expr used to construct this pruning predicate
pub fn logical_expr(&self) -> &Expr {
&self.logical_expr
}
/// Returns a reference to the predicate expr
pub fn predicate_expr(&self) -> &Arc<dyn PhysicalExpr> {
&self.predicate_expr
}
}
/// Handles creating references to the min/max statistics
/// for columns as well as recording which statistics are needed
#[derive(Debug, Default, Clone)]
struct RequiredStatColumns {
/// The statistics required to evaluate this predicate:
/// * The unqualified column in the input schema
/// * Statistics type (e.g. Min or Max or Null_Count)
/// * The field the statistics value should be placed in for
/// pruning predicate evaluation
columns: Vec<(Column, StatisticsType, Field)>,
}
impl RequiredStatColumns {
fn new() -> Self {
Self::default()
}
/// Retur an iterator over items in columns (see doc on
/// `self.columns` for details)
fn iter(&self) -> impl Iterator<Item = &(Column, StatisticsType, Field)> {
self.columns.iter()
}
fn is_stat_column_missing(
&self,
column: &Column,
statistics_type: StatisticsType,
) -> bool {
!self
.columns
.iter()
.any(|(c, t, _f)| c == column && t == &statistics_type)
}
/// Rewrites column_expr so that all appearances of column
/// are replaced with a reference to either the min or max
/// statistics column, while keeping track that a reference to the statistics
/// column is required
///
/// for example, an expression like `col("foo") > 5`, when called
/// with Max would result in an expression like `col("foo_max") >
/// 5` with the approprate entry noted in self.columns
fn stat_column_expr(
&mut self,
column: &Column,
column_expr: &Expr,
field: &Field,
stat_type: StatisticsType,
suffix: &str,
) -> Result<Expr> {
let stat_column = Column {
relation: column.relation.clone(),
name: format!("{}_{}", column.flat_name(), suffix),
};
let stat_field = Field::new(
stat_column.flat_name().as_str(),
field.data_type().clone(),
field.is_nullable(),
);
if self.is_stat_column_missing(column, stat_type) {
// only add statistics column if not previously added
self.columns.push((column.clone(), stat_type, stat_field));
}
rewrite_column_expr(column_expr.clone(), column, &stat_column)
}
/// rewrite col --> col_min
fn min_column_expr(
&mut self,
column: &Column,
column_expr: &Expr,
field: &Field,
) -> Result<Expr> {
self.stat_column_expr(column, column_expr, field, StatisticsType::Min, "min")
}
/// rewrite col --> col_max
fn max_column_expr(
&mut self,
column: &Column,
column_expr: &Expr,
field: &Field,
) -> Result<Expr> {
self.stat_column_expr(column, column_expr, field, StatisticsType::Max, "max")
}
/// rewrite col --> col_null_count
fn null_count_column_expr(
&mut self,
column: &Column,
column_expr: &Expr,
field: &Field,
) -> Result<Expr> {
self.stat_column_expr(
column,
column_expr,
field,
StatisticsType::NullCount,
"null_count",
)
}
}
impl From<Vec<(Column, StatisticsType, Field)>> for RequiredStatColumns {
fn from(columns: Vec<(Column, StatisticsType, Field)>) -> Self {
Self { columns }
}
}
/// Build a RecordBatch from a list of statistics, creating arrays,
/// with one row for each PruningStatistics and columns specified in
/// in the required_columns parameter.
///
/// For example, if the requested columns are
/// ```text
/// ("s1", Min, Field:s1_min)
/// ("s2", Max, field:s2_max)
///```
///
/// And the input statistics had
/// ```text
/// S1(Min: 5, Max: 10)
/// S2(Min: 99, Max: 1000)
/// S3(Min: 1, Max: 2)
/// ```
///
/// Then this function would build a record batch with 2 columns and
/// one row s1_min and s2_max as follows (s3 is not requested):
///
/// ```text
/// s1_min | s2_max
/// -------+--------
/// 5 | 1000
/// ```
fn build_statistics_record_batch<S: PruningStatistics>(
statistics: &S,
required_columns: &RequiredStatColumns,
) -> Result<RecordBatch> {
let mut fields = Vec::<Field>::new();
let mut arrays = Vec::<ArrayRef>::new();
// For each needed statistics column:
for (column, statistics_type, stat_field) in required_columns.iter() {
let data_type = stat_field.data_type();
let num_containers = statistics.num_containers();
let array = match statistics_type {
StatisticsType::Min => statistics.min_values(column),
StatisticsType::Max => statistics.max_values(column),
StatisticsType::NullCount => statistics.null_counts(column),
};
let array = array.unwrap_or_else(|| new_null_array(data_type, num_containers));
if num_containers != array.len() {
return Err(DataFusionError::Internal(format!(
"mismatched statistics length. Expected {}, got {}",
num_containers,
array.len()
)));
}
// cast statistics array to required data type (e.g. parquet
// provides timestamp statistics as "Int64")
let array = arrow::compute::cast(&array, data_type)?;
fields.push(stat_field.clone());
arrays.push(array);
}
let schema = Arc::new(Schema::new(fields));
RecordBatch::try_new(schema, arrays)
.map_err(|err| DataFusionError::Plan(err.to_string()))
}
struct PruningExpressionBuilder<'a> {
column: Column,
column_expr: Expr,
op: Operator,
scalar_expr: Expr,
field: &'a Field,
required_columns: &'a mut RequiredStatColumns,
}
impl<'a> PruningExpressionBuilder<'a> {
fn try_new(
left: &'a Expr,
right: &'a Expr,
op: Operator,
schema: &'a Schema,
required_columns: &'a mut RequiredStatColumns,
) -> Result<Self> {
// find column name; input could be a more complicated expression
let mut left_columns = HashSet::<Column>::new();
expr_to_columns(left, &mut left_columns)?;
let mut right_columns = HashSet::<Column>::new();
expr_to_columns(right, &mut right_columns)?;
let (column_expr, scalar_expr, columns, correct_operator) =
match (left_columns.len(), right_columns.len()) {
(1, 0) => (left, right, left_columns, op),
(0, 1) => (right, left, right_columns, reverse_operator(op)),
_ => {
// if more than one column used in expression - not supported
return Err(DataFusionError::Plan(
"Multi-column expressions are not currently supported"
.to_string(),
));
}
};
let (column_expr, correct_operator, scalar_expr) =
match rewrite_expr_to_prunable(column_expr, correct_operator, scalar_expr) {
Ok(ret) => ret,
Err(e) => return Err(e),
};
let column = columns.iter().next().unwrap().clone();
let field = match schema.column_with_name(&column.flat_name()) {
Some((_, f)) => f,
_ => {
return Err(DataFusionError::Plan(
"Field not found in schema".to_string(),
));
}
};
Ok(Self {
column,
column_expr,
op: correct_operator,
scalar_expr,
field,
required_columns,
})
}
fn op(&self) -> Operator {
self.op
}
fn scalar_expr(&self) -> &Expr {
&self.scalar_expr
}
fn min_column_expr(&mut self) -> Result<Expr> {
self.required_columns
.min_column_expr(&self.column, &self.column_expr, self.field)
}
fn max_column_expr(&mut self) -> Result<Expr> {
self.required_columns
.max_column_expr(&self.column, &self.column_expr, self.field)
}
}
/// This function is designed to rewrite the column_expr to
/// ensure the column_expr is monotonically increasing.
///
/// For example,
/// 1. `col > 10`
/// 2. `-col > 10` should be rewritten to `col < -10`
/// 3. `!col = true` would be rewritten to `col = !true`
/// 4. `abs(a - 10) > 0` not supported
///
/// More rewrite rules are still in progress.
fn rewrite_expr_to_prunable(
column_expr: &Expr,
op: Operator,
scalar_expr: &Expr,
) -> Result<(Expr, Operator, Expr)> {
if !is_compare_op(op) {
return Err(DataFusionError::Plan(
"rewrite_expr_to_prunable only support compare expression".to_string(),
));
}
match column_expr {
// `col > lit()`
Expr::Column(_) => Ok((column_expr.clone(), op, scalar_expr.clone())),
// `-col > lit()` --> `col < -lit()`
Expr::Negative(c) => match c.as_ref() {
Expr::Column(_) => Ok((
c.as_ref().clone(),
reverse_operator(op),
Expr::Negative(Box::new(scalar_expr.clone())),
)),
_ => Err(DataFusionError::Plan(format!(
"negative with complex expression {:?} is not supported",
column_expr
))),
},
// `!col = true` --> `col = !true`
Expr::Not(c) => {
if op != Operator::Eq && op != Operator::NotEq {
return Err(DataFusionError::Plan(
"Not with operator other than Eq / NotEq is not supported"
.to_string(),
));
}
return match c.as_ref() {
Expr::Column(_) => Ok((
c.as_ref().clone(),
reverse_operator(op),
Expr::Not(Box::new(scalar_expr.clone())),
)),
_ => Err(DataFusionError::Plan(format!(
"Not with complex expression {:?} is not supported",
column_expr
))),
};
}
_ => Err(DataFusionError::Plan(format!(
"column expression {:?} is not supported",
column_expr
))),
}
}
fn is_compare_op(op: Operator) -> bool {
matches!(
op,
Operator::Eq
| Operator::NotEq
| Operator::Lt
| Operator::LtEq
| Operator::Gt
| Operator::GtEq
)
}
/// replaces a column with an old name with a new name in an expression
fn rewrite_column_expr(
e: Expr,
column_old: &Column,
column_new: &Column,
) -> Result<Expr> {
struct ColumnReplacer<'a> {
old: &'a Column,
new: &'a Column,
}
impl<'a> ExprRewriter for ColumnReplacer<'a> {
fn mutate(&mut self, expr: Expr) -> Result<Expr> {
match expr {
Expr::Column(c) if c == *self.old => Ok(Expr::Column(self.new.clone())),
_ => Ok(expr),
}
}
}
e.rewrite(&mut ColumnReplacer {
old: column_old,
new: column_new,
})
}
fn reverse_operator(op: Operator) -> Operator {
match op {
Operator::Lt => Operator::Gt,
Operator::Gt => Operator::Lt,
Operator::LtEq => Operator::GtEq,
Operator::GtEq => Operator::LtEq,
_ => op,
}
}
/// Given a column reference to `column`, returns a pruning
/// expression in terms of the min and max that will evaluate to true
/// if the column may contain values, and false if definitely does not
/// contain values
fn build_single_column_expr(
column: &Column,
schema: &Schema,
required_columns: &mut RequiredStatColumns,
is_not: bool, // if true, treat as !col
) -> Option<Expr> {
let field = schema.field_with_name(&column.name).ok()?;
if matches!(field.data_type(), &DataType::Boolean) {
let col_ref = Expr::Column(column.clone());
let min = required_columns
.min_column_expr(column, &col_ref, field)
.ok()?;
let max = required_columns
.max_column_expr(column, &col_ref, field)
.ok()?;
// remember -- we want an expression that is:
// TRUE: if there may be rows that match
// FALSE: if there are no rows that match
if is_not {
// The only way we know a column couldn't match is if both the min and max are true
// !(min && max)
Some(!(min.and(max)))
} else {
// the only way we know a column couldn't match is if both the min and max are false
// !(!min && !max) --> min || max
Some(min.or(max))
}
} else {
None
}
}
/// Given an expression reference to `expr`, if `expr` is a column expression,
/// returns a pruning expression in terms of IsNull that will evaluate to true
/// if the column may contain null, and false if definitely does not
/// contain null.
fn build_is_null_column_expr(
expr: &Expr,
schema: &Schema,
required_columns: &mut RequiredStatColumns,
) -> Option<Expr> {
match expr {
Expr::Column(ref col) => {
let field = schema.field_with_name(&col.name).ok()?;
let null_count_field = &Field::new(field.name(), DataType::UInt64, false);
required_columns
.null_count_column_expr(col, expr, null_count_field)
.map(|null_count_column_expr| {
// IsNull(column) => null_count > 0
null_count_column_expr.gt(lit::<u64>(0))
})
.ok()
}
_ => None,
}
}
/// Translate logical filter expression into pruning predicate
/// expression that will evaluate to FALSE if it can be determined no
/// rows between the min/max values could pass the predicates.
///
/// Returns the pruning predicate as an [`Expr`]
fn build_predicate_expression(
expr: &Expr,
schema: &Schema,
required_columns: &mut RequiredStatColumns,
) -> Result<Expr> {
use crate::logical_plan;
// Returned for unsupported expressions. Such expressions are
// converted to TRUE. This can still be useful when multiple
// conditions are joined using AND such as: column > 10 AND TRUE
let unhandled = logical_plan::lit(true);
// predicate expression can only be a binary expression
let (left, op, right) = match expr {
Expr::BinaryExpr { left, op, right } => (left, *op, right),
Expr::IsNull(expr) => {
let expr = build_is_null_column_expr(expr, schema, required_columns)
.unwrap_or(unhandled);
return Ok(expr);
}
Expr::Column(col) => {
let expr = build_single_column_expr(col, schema, required_columns, false)
.unwrap_or(unhandled);
return Ok(expr);
}
// match !col (don't do so recursively)
Expr::Not(input) => {
if let Expr::Column(col) = input.as_ref() {
let expr = build_single_column_expr(col, schema, required_columns, true)
.unwrap_or(unhandled);
return Ok(expr);
} else {
return Ok(unhandled);
}
}
Expr::InList {
expr,
list,
negated,
} if !list.is_empty() && list.len() < 20 => {
let eq_fun = if *negated { Expr::not_eq } else { Expr::eq };
let re_fun = if *negated { Expr::and } else { Expr::or };
let change_expr = list
.iter()
.map(|e| eq_fun(*expr.clone(), e.clone()))
.reduce(re_fun)
.unwrap();
return build_predicate_expression(&change_expr, schema, required_columns);
}
_ => {
return Ok(unhandled);
}
};
if op == Operator::And || op == Operator::Or {
let left_expr = build_predicate_expression(left, schema, required_columns)?;
let right_expr = build_predicate_expression(right, schema, required_columns)?;
return Ok(binary_expr(left_expr, op, right_expr));
}
let expr_builder =
PruningExpressionBuilder::try_new(left, right, op, schema, required_columns);
let mut expr_builder = match expr_builder {
Ok(builder) => builder,
// allow partial failure in predicate expression generation
// this can still produce a useful predicate when multiple conditions are joined using AND
Err(_) => {
return Ok(unhandled);
}
};
let statistics_expr = build_statistics_expr(&mut expr_builder).unwrap_or(unhandled);
Ok(statistics_expr)
}
fn build_statistics_expr(expr_builder: &mut PruningExpressionBuilder) -> Result<Expr> {
let statistics_expr =
match expr_builder.op() {
Operator::NotEq => {
// column != literal => (min, max) = literal =>
// !(min != literal && max != literal) ==>
// min != literal || literal != max
let min_column_expr = expr_builder.min_column_expr()?;
let max_column_expr = expr_builder.max_column_expr()?;
min_column_expr
.not_eq(expr_builder.scalar_expr().clone())
.or(expr_builder.scalar_expr().clone().not_eq(max_column_expr))
}
Operator::Eq => {
// column = literal => (min, max) = literal => min <= literal && literal <= max
// (column / 2) = 4 => (column_min / 2) <= 4 && 4 <= (column_max / 2)
let min_column_expr = expr_builder.min_column_expr()?;
let max_column_expr = expr_builder.max_column_expr()?;
min_column_expr
.lt_eq(expr_builder.scalar_expr().clone())
.and(expr_builder.scalar_expr().clone().lt_eq(max_column_expr))
}
Operator::Gt => {
// column > literal => (min, max) > literal => max > literal
expr_builder
.max_column_expr()?
.gt(expr_builder.scalar_expr().clone())
}
Operator::GtEq => {
// column >= literal => (min, max) >= literal => max >= literal
expr_builder
.max_column_expr()?
.gt_eq(expr_builder.scalar_expr().clone())
}
Operator::Lt => {
// column < literal => (min, max) < literal => min < literal
expr_builder
.min_column_expr()?
.lt(expr_builder.scalar_expr().clone())
}
Operator::LtEq => {
// column <= literal => (min, max) <= literal => min <= literal
expr_builder
.min_column_expr()?
.lt_eq(expr_builder.scalar_expr().clone())
}
// other expressions are not supported
_ => return Err(DataFusionError::Plan(
"expressions other than (neq, eq, gt, gteq, lt, lteq) are not superted"
.to_string(),
)),
};
Ok(statistics_expr)
}
#[derive(Debug, Copy, Clone, PartialEq, Eq)]
enum StatisticsType {
Min,
Max,
NullCount,
}
#[cfg(test)]
mod tests {
use super::*;
use crate::from_slice::FromSlice;
use crate::logical_plan::{col, lit};
use crate::{assert_batches_eq, physical_optimizer::pruning::StatisticsType};
use arrow::array::Decimal128Array;
use arrow::{
array::{BinaryArray, Int32Array, Int64Array, StringArray},
datatypes::{DataType, TimeUnit},
};
use datafusion_common::ScalarValue;
use std::collections::HashMap;
#[derive(Debug)]
/// Test for container stats
struct ContainerStats {
min: ArrayRef,
max: ArrayRef,
}
impl ContainerStats {
fn new_decimal128(
min: impl IntoIterator<Item = Option<i128>>,
max: impl IntoIterator<Item = Option<i128>>,
precision: usize,
scale: usize,
) -> Self {
Self {
min: Arc::new(
min.into_iter()
.collect::<Decimal128Array>()
.with_precision_and_scale(precision, scale)
.unwrap(),
),
max: Arc::new(
max.into_iter()
.collect::<Decimal128Array>()
.with_precision_and_scale(precision, scale)
.unwrap(),
),
}
}
fn new_i64(
min: impl IntoIterator<Item = Option<i64>>,
max: impl IntoIterator<Item = Option<i64>>,
) -> Self {
Self {
min: Arc::new(min.into_iter().collect::<Int64Array>()),
max: Arc::new(max.into_iter().collect::<Int64Array>()),
}
}
fn new_i32(
min: impl IntoIterator<Item = Option<i32>>,
max: impl IntoIterator<Item = Option<i32>>,
) -> Self {
Self {
min: Arc::new(min.into_iter().collect::<Int32Array>()),
max: Arc::new(max.into_iter().collect::<Int32Array>()),
}
}
fn new_utf8<'a>(
min: impl IntoIterator<Item = Option<&'a str>>,
max: impl IntoIterator<Item = Option<&'a str>>,
) -> Self {
Self {
min: Arc::new(min.into_iter().collect::<StringArray>()),
max: Arc::new(max.into_iter().collect::<StringArray>()),
}
}
fn new_bool(
min: impl IntoIterator<Item = Option<bool>>,
max: impl IntoIterator<Item = Option<bool>>,
) -> Self {
Self {
min: Arc::new(min.into_iter().collect::<BooleanArray>()),
max: Arc::new(max.into_iter().collect::<BooleanArray>()),
}
}
fn min(&self) -> Option<ArrayRef> {
Some(self.min.clone())
}
fn max(&self) -> Option<ArrayRef> {
Some(self.max.clone())
}
fn len(&self) -> usize {
assert_eq!(self.min.len(), self.max.len());
self.min.len()
}
}
#[derive(Debug, Default)]
struct TestStatistics {
// key: column name
stats: HashMap<Column, ContainerStats>,
}
impl TestStatistics {
fn new() -> Self {
Self::default()
}
fn with(
mut self,
name: impl Into<String>,
container_stats: ContainerStats,
) -> Self {
self.stats
.insert(Column::from_name(name.into()), container_stats);
self
}
}
impl PruningStatistics for TestStatistics {
fn min_values(&self, column: &Column) -> Option<ArrayRef> {
self.stats
.get(column)
.map(|container_stats| container_stats.min())
.unwrap_or(None)
}
fn max_values(&self, column: &Column) -> Option<ArrayRef> {
self.stats
.get(column)
.map(|container_stats| container_stats.max())
.unwrap_or(None)
}
fn num_containers(&self) -> usize {
self.stats
.values()
.next()
.map(|container_stats| container_stats.len())
.unwrap_or(0)
}
fn null_counts(&self, _column: &Column) -> Option<ArrayRef> {
None
}
}
/// Returns the specified min/max container values
struct OneContainerStats {
min_values: Option<ArrayRef>,
max_values: Option<ArrayRef>,
num_containers: usize,
}
impl PruningStatistics for OneContainerStats {
fn min_values(&self, _column: &Column) -> Option<ArrayRef> {
self.min_values.clone()
}
fn max_values(&self, _column: &Column) -> Option<ArrayRef> {
self.max_values.clone()
}
fn num_containers(&self) -> usize {
self.num_containers
}
fn null_counts(&self, _column: &Column) -> Option<ArrayRef> {
None
}
}
#[test]
fn test_build_statistics_record_batch() {
// Request a record batch with of s1_min, s2_max, s3_max, s3_min
let required_columns = RequiredStatColumns::from(vec![
// min of original column s1, named s1_min
(
"s1".into(),
StatisticsType::Min,
Field::new("s1_min", DataType::Int32, true),
),
// max of original column s2, named s2_max
(
"s2".into(),
StatisticsType::Max,
Field::new("s2_max", DataType::Int32, true),
),
// max of original column s3, named s3_max
(
"s3".into(),
StatisticsType::Max,
Field::new("s3_max", DataType::Utf8, true),
),
// min of original column s3, named s3_min
(
"s3".into(),
StatisticsType::Min,
Field::new("s3_min", DataType::Utf8, true),
),
]);
let statistics = TestStatistics::new()