Coverage Report

Created: 2024-10-13 08:39

/Users/andrewlamb/Software/datafusion/datafusion/physical-expr/src/window/rank.rs
Line
Count
Source (jump to first uncovered line)
1
// Licensed to the Apache Software Foundation (ASF) under one
2
// or more contributor license agreements.  See the NOTICE file
3
// distributed with this work for additional information
4
// regarding copyright ownership.  The ASF licenses this file
5
// to you under the Apache License, Version 2.0 (the
6
// "License"); you may not use this file except in compliance
7
// with the License.  You may obtain a copy of the License at
8
//
9
//   http://www.apache.org/licenses/LICENSE-2.0
10
//
11
// Unless required by applicable law or agreed to in writing,
12
// software distributed under the License is distributed on an
13
// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
14
// KIND, either express or implied.  See the License for the
15
// specific language governing permissions and limitations
16
// under the License.
17
18
//! Defines physical expression for `rank`, `dense_rank`, and `percent_rank` that can evaluated
19
//! at runtime during query execution
20
21
use crate::expressions::Column;
22
use crate::window::window_expr::RankState;
23
use crate::window::BuiltInWindowFunctionExpr;
24
use crate::{PhysicalExpr, PhysicalSortExpr};
25
26
use arrow::array::ArrayRef;
27
use arrow::array::{Float64Array, UInt64Array};
28
use arrow::datatypes::{DataType, Field};
29
use arrow_schema::{SchemaRef, SortOptions};
30
use datafusion_common::utils::get_row_at_idx;
31
use datafusion_common::{exec_err, Result, ScalarValue};
32
use datafusion_expr::PartitionEvaluator;
33
34
use std::any::Any;
35
use std::iter;
36
use std::ops::Range;
37
use std::sync::Arc;
38
39
/// Rank calculates the rank in the window function with order by
40
#[derive(Debug)]
41
pub struct Rank {
42
    name: String,
43
    rank_type: RankType,
44
    /// Output data type
45
    data_type: DataType,
46
}
47
48
impl Rank {
49
    /// Get rank_type of the rank in window function with order by
50
0
    pub fn get_type(&self) -> RankType {
51
0
        self.rank_type
52
0
    }
53
}
54
55
#[derive(Debug, Copy, Clone)]
56
pub enum RankType {
57
    Basic,
58
    Dense,
59
    Percent,
60
}
61
62
/// Create a rank window function
63
0
pub fn rank(name: String, data_type: &DataType) -> Rank {
64
0
    Rank {
65
0
        name,
66
0
        rank_type: RankType::Basic,
67
0
        data_type: data_type.clone(),
68
0
    }
69
0
}
70
71
/// Create a dense rank window function
72
0
pub fn dense_rank(name: String, data_type: &DataType) -> Rank {
73
0
    Rank {
74
0
        name,
75
0
        rank_type: RankType::Dense,
76
0
        data_type: data_type.clone(),
77
0
    }
78
0
}
79
80
/// Create a percent rank window function
81
0
pub fn percent_rank(name: String, data_type: &DataType) -> Rank {
82
0
    Rank {
83
0
        name,
84
0
        rank_type: RankType::Percent,
85
0
        data_type: data_type.clone(),
86
0
    }
87
0
}
88
89
impl BuiltInWindowFunctionExpr for Rank {
90
    /// Return a reference to Any that can be used for downcasting
91
0
    fn as_any(&self) -> &dyn Any {
92
0
        self
93
0
    }
94
95
0
    fn field(&self) -> Result<Field> {
96
0
        let nullable = false;
97
0
        Ok(Field::new(self.name(), self.data_type.clone(), nullable))
98
0
    }
99
100
0
    fn expressions(&self) -> Vec<Arc<dyn PhysicalExpr>> {
101
0
        vec![]
102
0
    }
103
104
0
    fn name(&self) -> &str {
105
0
        &self.name
106
0
    }
107
108
0
    fn create_evaluator(&self) -> Result<Box<dyn PartitionEvaluator>> {
109
0
        Ok(Box::new(RankEvaluator {
110
0
            state: RankState::default(),
111
0
            rank_type: self.rank_type,
112
0
        }))
113
0
    }
114
115
0
    fn get_result_ordering(&self, schema: &SchemaRef) -> Option<PhysicalSortExpr> {
116
0
        // The built-in RANK window function (in all modes) introduces a new ordering:
117
0
        schema.column_with_name(self.name()).map(|(idx, field)| {
118
0
            let expr = Arc::new(Column::new(field.name(), idx));
119
0
            let options = SortOptions {
120
0
                descending: false,
121
0
                nulls_first: false,
122
0
            }; // ASC, NULLS LAST
123
0
            PhysicalSortExpr { expr, options }
124
0
        })
125
0
    }
126
}
127
128
#[derive(Debug)]
129
pub(crate) struct RankEvaluator {
130
    state: RankState,
131
    rank_type: RankType,
132
}
133
134
impl PartitionEvaluator for RankEvaluator {
135
0
    fn is_causal(&self) -> bool {
136
0
        matches!(self.rank_type, RankType::Basic | RankType::Dense)
137
0
    }
138
139
    /// Evaluates the window function inside the given range.
140
0
    fn evaluate(
141
0
        &mut self,
142
0
        values: &[ArrayRef],
143
0
        range: &Range<usize>,
144
0
    ) -> Result<ScalarValue> {
145
0
        let row_idx = range.start;
146
0
        // There is no argument, values are order by column values (where rank is calculated)
147
0
        let range_columns = values;
148
0
        let last_rank_data = get_row_at_idx(range_columns, row_idx)?;
149
0
        let new_rank_encountered =
150
0
            if let Some(state_last_rank_data) = &self.state.last_rank_data {
151
                // if rank data changes, new rank is encountered
152
0
                state_last_rank_data != &last_rank_data
153
            } else {
154
                // First rank seen
155
0
                true
156
            };
157
0
        if new_rank_encountered {
158
0
            self.state.last_rank_data = Some(last_rank_data);
159
0
            self.state.last_rank_boundary += self.state.current_group_count;
160
0
            self.state.current_group_count = 1;
161
0
            self.state.n_rank += 1;
162
0
        } else {
163
0
            // data is still in the same rank
164
0
            self.state.current_group_count += 1;
165
0
        }
166
0
        match self.rank_type {
167
0
            RankType::Basic => Ok(ScalarValue::UInt64(Some(
168
0
                self.state.last_rank_boundary as u64 + 1,
169
0
            ))),
170
0
            RankType::Dense => Ok(ScalarValue::UInt64(Some(self.state.n_rank as u64))),
171
            RankType::Percent => {
172
0
                exec_err!("Can not execute PERCENT_RANK in a streaming fashion")
173
            }
174
        }
175
0
    }
176
177
0
    fn evaluate_all_with_rank(
178
0
        &self,
179
0
        num_rows: usize,
180
0
        ranks_in_partition: &[Range<usize>],
181
0
    ) -> Result<ArrayRef> {
182
        // see https://www.postgresql.org/docs/current/functions-window.html
183
0
        let result: ArrayRef = match self.rank_type {
184
0
            RankType::Dense => Arc::new(UInt64Array::from_iter_values(
185
0
                ranks_in_partition
186
0
                    .iter()
187
0
                    .zip(1u64..)
188
0
                    .flat_map(|(range, rank)| {
189
0
                        let len = range.end - range.start;
190
0
                        iter::repeat(rank).take(len)
191
0
                    }),
192
0
            )),
193
            RankType::Percent => {
194
                // Returns the relative rank of the current row, that is (rank - 1) / (total partition rows - 1). The value thus ranges from 0 to 1 inclusive.
195
0
                let denominator = num_rows as f64;
196
0
                Arc::new(Float64Array::from_iter_values(
197
0
                    ranks_in_partition
198
0
                        .iter()
199
0
                        .scan(0_u64, |acc, range| {
200
0
                            let len = range.end - range.start;
201
0
                            let value = (*acc as f64) / (denominator - 1.0).max(1.0);
202
0
                            let result = iter::repeat(value).take(len);
203
0
                            *acc += len as u64;
204
0
                            Some(result)
205
0
                        })
206
0
                        .flatten(),
207
0
                ))
208
            }
209
0
            RankType::Basic => Arc::new(UInt64Array::from_iter_values(
210
0
                ranks_in_partition
211
0
                    .iter()
212
0
                    .scan(1_u64, |acc, range| {
213
0
                        let len = range.end - range.start;
214
0
                        let result = iter::repeat(*acc).take(len);
215
0
                        *acc += len as u64;
216
0
                        Some(result)
217
0
                    })
218
0
                    .flatten(),
219
0
            )),
220
        };
221
0
        Ok(result)
222
0
    }
223
224
0
    fn supports_bounded_execution(&self) -> bool {
225
0
        matches!(self.rank_type, RankType::Basic | RankType::Dense)
226
0
    }
227
228
0
    fn include_rank(&self) -> bool {
229
0
        true
230
0
    }
231
}
232
233
#[cfg(test)]
234
mod tests {
235
    use super::*;
236
    use datafusion_common::cast::{as_float64_array, as_uint64_array};
237
238
    fn test_with_rank(expr: &Rank, expected: Vec<u64>) -> Result<()> {
239
        test_i32_result(expr, vec![0..2, 2..3, 3..6, 6..7, 7..8], expected)
240
    }
241
242
    #[allow(clippy::single_range_in_vec_init)]
243
    fn test_without_rank(expr: &Rank, expected: Vec<u64>) -> Result<()> {
244
        test_i32_result(expr, vec![0..8], expected)
245
    }
246
247
    fn test_f64_result(
248
        expr: &Rank,
249
        num_rows: usize,
250
        ranks: Vec<Range<usize>>,
251
        expected: Vec<f64>,
252
    ) -> Result<()> {
253
        let result = expr
254
            .create_evaluator()?
255
            .evaluate_all_with_rank(num_rows, &ranks)?;
256
        let result = as_float64_array(&result)?;
257
        let result = result.values();
258
        assert_eq!(expected, *result);
259
        Ok(())
260
    }
261
262
    fn test_i32_result(
263
        expr: &Rank,
264
        ranks: Vec<Range<usize>>,
265
        expected: Vec<u64>,
266
    ) -> Result<()> {
267
        let result = expr.create_evaluator()?.evaluate_all_with_rank(8, &ranks)?;
268
        let result = as_uint64_array(&result)?;
269
        let result = result.values();
270
        assert_eq!(expected, *result);
271
        Ok(())
272
    }
273
274
    #[test]
275
    fn test_dense_rank() -> Result<()> {
276
        let r = dense_rank("arr".into(), &DataType::UInt64);
277
        test_without_rank(&r, vec![1; 8])?;
278
        test_with_rank(&r, vec![1, 1, 2, 3, 3, 3, 4, 5])?;
279
        Ok(())
280
    }
281
282
    #[test]
283
    fn test_rank() -> Result<()> {
284
        let r = rank("arr".into(), &DataType::UInt64);
285
        test_without_rank(&r, vec![1; 8])?;
286
        test_with_rank(&r, vec![1, 1, 3, 4, 4, 4, 7, 8])?;
287
        Ok(())
288
    }
289
290
    #[test]
291
    #[allow(clippy::single_range_in_vec_init)]
292
    fn test_percent_rank() -> Result<()> {
293
        let r = percent_rank("arr".into(), &DataType::Float64);
294
295
        // empty case
296
        let expected = vec![0.0; 0];
297
        test_f64_result(&r, 0, vec![0..0; 0], expected)?;
298
299
        // singleton case
300
        let expected = vec![0.0];
301
        test_f64_result(&r, 1, vec![0..1], expected)?;
302
303
        // uniform case
304
        let expected = vec![0.0; 7];
305
        test_f64_result(&r, 7, vec![0..7], expected)?;
306
307
        // non-trivial case
308
        let expected = vec![0.0, 0.0, 0.0, 0.5, 0.5, 0.5, 0.5];
309
        test_f64_result(&r, 7, vec![0..3, 3..7], expected)?;
310
311
        Ok(())
312
    }
313
}