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
use polars_core::utils::split_df_as_ref;
use super::*;
use crate::plans::hive::HivePartitionsDf;
use crate::plans::inputs::Inputs;
use crate::utils::deep_clone_ir;
fn is_hive_partitioned(node: Node, ir_arena: &Arena<IR>) -> Option<HivePartitionsDf> {
for (_, ir) in ir_arena.iter(node) {
match ir {
IR::Scan { hive_parts, .. } => return hive_parts.clone(),
// Only partition the first group-by.
IR::GroupBy { .. } => return None,
// They can modify hive names.
// Be conservative for now.
IR::Select { .. } | IR::HStack { .. } => return None,
// We only want to return hive partitions for the first joins
// Any node in between with more than one input (join, union, etc) will not return a
// match.
ir if matches!(ir.inputs(), Inputs::Single { .. }) => continue,
_ => return None,
}
}
None
}
#[cfg(feature = "is_in")]
fn hive_rewrite_supports_join_type(how: &JoinType) -> bool {
match how {
JoinType::Inner | JoinType::Left | JoinType::Right => true,
#[cfg(feature = "semi_anti_join")]
JoinType::Semi => true,
_ => false,
}
}
fn get_partitions(hive_df: &DataFrame) -> Vec<DataFrame> {
let n_parts = std::env::var("POLARS_HIVE_PARTITIONS")
.ok()
.and_then(|v| v.parse::<usize>().ok())
.unwrap_or(64);
split_df_as_ref(hive_df, std::cmp::min(n_parts, hive_df.height()), false)
}
// We must deduplicate so that we get unique files (keys) per partition.
#[cfg(feature = "is_in")]
fn unique_key_frame(hive_df: &DataFrame, key_name: &PlSmallStr) -> PolarsResult<DataFrame> {
hive_df
.column(key_name)?
.clone()
.into_frame()
.unique_stable(None, UniqueKeepStrategy::First, None)
}
#[allow(clippy::too_many_arguments)]
pub fn rewrite_hive(
ir: IR,
opt: &mut PredicatePushDown,
ir_arena: &mut Arena<IR>,
expr_arena: &mut Arena<AExpr>,
) -> PolarsResult<IR> {
#[cfg(not(feature = "is_in"))]
return Ok(ir);
#[cfg(feature = "is_in")]
{
if opt.hive_rewrite_active {
return Ok(ir);
}
match ir {
IR::GroupBy {
input,
keys,
aggs,
schema,
maintain_order,
options,
apply,
} if opt.partition_hive
&& !maintain_order
&& let Some(hive) = is_hive_partitioned(input, ir_arena) =>
{
// This replaces a group-by on a hive partitioned key
// by a union on hive partitioned group-by's.
// We do that by pushing down an is_in predicate
// Later in the optimizer we prune the hive paths
// based on all the predicates.
// Any hive column is valid to split on, prefer the leftmost as it is the coarsest.
let mut hive_col: Option<(usize, PlSmallStr)> = None;
let hive_schema = hive.schema();
for e in keys.iter() {
let key = expr_arena.get(e.node());
if let AExpr::Column(name) = key {
if let Some(idx) = hive_schema.index_of(name)
&& hive_col.as_ref().is_none_or(|(best, _)| idx < *best)
{
hive_col = Some((idx, name.clone()));
}
}
}
if let Some((_, key_name)) = hive_col {
let hive_df = unique_key_frame(hive.df(), &key_name)?;
let chunks = get_partitions(&hive_df);
let mut branches = Vec::with_capacity(chunks.len());
for chunk in chunks {
if chunk.height() == 0 {
continue;
}
let pred =
make_predicate(&chunk, key_name.clone(), key_name.clone(), expr_arena);
// We need to deep clone as each branch hits different predicate pd passes.
let branch = deep_clone_ir(input, ir_arena);
let mut acc_new = init_indexmap(Some(1));
insert_predicate_dedup(&mut acc_new, &pred, expr_arena);
opt.pushdown_and_assign(branch, acc_new, ir_arena, expr_arena)?;
branches.push(ir_arena.add(IR::GroupBy {
input: branch,
keys: keys.clone(),
aggs: aggs.clone(),
schema: schema.clone(),
maintain_order,
options: options.clone(),
apply: apply.clone(),
}));
}
Ok(IR::Union {
inputs: branches,
options: UnionOptions {
slice: options.slice,
maintain_order: false,
..Default::default()
},
})
} else {
Ok(IR::GroupBy {
input,
keys,
aggs,
schema,
maintain_order,
options,
apply,
})
}
},
IR::Join {
input_left,
input_right,
schema,
left_on,
right_on,
options,
} if let (MaintainOrderJoin::None, true, Some(hive_left), Some(hive_right)) = (
&options.args.maintain_order,
hive_rewrite_supports_join_type(&options.args.how),
is_hive_partitioned(input_left, ir_arena),
is_hive_partitioned(input_right, ir_arena),
) =>
{
// This replaces a join on a hive partitioned key
// by a union on hive partitioned joins.
// We do that by pushing down an is_in predicate
// Later in the optimizer we prune the hive paths
// based on all the predicates.
// Any hive column is valid to split on, prefer the leftmost as it is the coarsest.
let mut hive_cols: Option<(usize, PlSmallStr, PlSmallStr)> = None;
let hive_left_schema = hive_left.schema();
let hive_right_schema = hive_right.schema();
for (l, r) in left_on.iter().zip(right_on.iter()) {
let l = expr_arena.get(l.node());
let r = expr_arena.get(r.node());
if let (AExpr::Column(l), AExpr::Column(r)) = (l, r) {
if let Some(idx) = hive_left_schema.index_of(l)
&& hive_right_schema.contains(r)
&& hive_cols.as_ref().is_none_or(|(best, _, _)| idx < *best)
{
hive_cols = Some((idx, l.clone(), r.clone()));
}
}
}
if let Some((_, l, r)) = hive_cols {
let hive_l = unique_key_frame(hive_left.df(), &l)?;
let hive_r = unique_key_frame(hive_right.df(), &r)?;
let partitions = hive_l
.join(
&hive_r,
[l.as_str()],
[r.as_str()],
JoinArgs {
how: options.args.how.clone(),
nulls_equal: options.args.nulls_equal,
..Default::default()
},
None,
)
.unwrap();
let l_key_name = if partitions.schema().contains(l.as_str()) {
l.clone()
} else {
r.clone()
};
let r_key_name = if partitions.schema().contains(r.as_str()) {
r.clone()
} else {
l.clone()
};
if !opt.partition_hive {
let (l_pred, r_pred) = make_predicates(
&partitions,
l_key_name.clone(),
l.clone(),
r_key_name.clone(),
r.clone(),
expr_arena,
);
// TODO: this goes into the branch twice. We could optimize for a single pass.
let mut acc_left = init_indexmap(Some(1));
insert_predicate_dedup(&mut acc_left, &l_pred, expr_arena);
opt.pushdown_and_assign(input_left, acc_left, ir_arena, expr_arena)?;
let mut acc_right = init_indexmap(Some(1));
insert_predicate_dedup(&mut acc_right, &r_pred, expr_arena);
opt.pushdown_and_assign(input_right, acc_right, ir_arena, expr_arena)?;
} else {
let chunks = get_partitions(&partitions);
let mut branches = Vec::with_capacity(chunks.len());
for chunk in chunks {
if chunk.height() == 0 {
continue;
}
let (l_pred, r_pred) = make_predicates(
&chunk,
l_key_name.clone(),
l.clone(),
r_key_name.clone(),
r.clone(),
expr_arena,
);
// We need to deep clone as each branch hits different predicate pd passes.
let branch_left = deep_clone_ir(input_left, ir_arena);
let branch_right = deep_clone_ir(input_right, ir_arena);
let mut acc_left = init_indexmap(Some(1));
insert_predicate_dedup(&mut acc_left, &l_pred, expr_arena);
opt.pushdown_and_assign(branch_left, acc_left, ir_arena, expr_arena)?;
let mut acc_right = init_indexmap(Some(1));
insert_predicate_dedup(&mut acc_right, &r_pred, expr_arena);
opt.pushdown_and_assign(branch_right, acc_right, ir_arena, expr_arena)?;
branches.push(ir_arena.add(IR::Join {
input_left: branch_left,
input_right: branch_right,
left_on: left_on.clone(),
right_on: right_on.clone(),
schema: schema.clone(),
options: options.clone(),
}));
}
return Ok(IR::Union {
inputs: branches,
options: UnionOptions {
maintain_order: false,
slice: options.args.slice,
..Default::default()
},
});
}
}
Ok(IR::Join {
input_left,
input_right,
left_on,
right_on,
schema,
options,
})
},
_ => Ok(ir),
}
}
}
#[cfg(feature = "is_in")]
fn make_predicates(
partitions: &DataFrame,
extract_name_left: PlSmallStr,
predicate_name_left: PlSmallStr,
extract_name_right: PlSmallStr,
predicate_name_right: PlSmallStr,
expr_arena: &mut Arena<AExpr>,
) -> (ExprIR, ExprIR) {
(
make_predicate(
partitions,
extract_name_left,
predicate_name_left,
expr_arena,
),
make_predicate(
partitions,
extract_name_right,
predicate_name_right,
expr_arena,
),
)
}
#[cfg(feature = "is_in")]
fn make_predicate(
partitions: &DataFrame,
extract_name: PlSmallStr,
predicate_name: PlSmallStr,
expr_arena: &mut Arena<AExpr>,
) -> ExprIR {
let values = partitions
.column(&extract_name)
.unwrap()
.as_materialized_series()
.implode()
.unwrap()
.into_series();
AExprBuilder::col(predicate_name, expr_arena)
.is_in(
AExprBuilder::lit(LiteralValue::Series(SpecialEq::new(values)), expr_arena),
// Hive __HIVE_DEFAULT_PARTITION__ can produce nulls
true,
expr_arena,
)
.expr_ir_unnamed()
}