datafusion_physical_plan/aggregates/group_values/single_group_by/bytes.rs
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
18use std::mem::size_of;
19
20use crate::aggregates::group_values::GroupValues;
21
22use arrow::array::{Array, ArrayRef, OffsetSizeTrait, RecordBatch};
23use datafusion_common::Result;
24use datafusion_expr::EmitTo;
25use datafusion_physical_expr_common::binary_map::{ArrowBytesMap, OutputType};
26
27/// A [`GroupValues`] storing single column of Utf8/LargeUtf8/Binary/LargeBinary values
28///
29/// This specialization is significantly faster than using the more general
30/// purpose `Row`s format
31pub struct GroupValuesBytes<O: OffsetSizeTrait> {
32 /// Map string/binary values to group index
33 map: ArrowBytesMap<O, usize>,
34 /// The total number of groups so far (used to assign group_index)
35 num_groups: usize,
36}
37
38impl<O: OffsetSizeTrait> GroupValuesBytes<O> {
39 pub fn new(output_type: OutputType) -> Self {
40 Self {
41 map: ArrowBytesMap::new(output_type),
42 num_groups: 0,
43 }
44 }
45}
46
47impl<O: OffsetSizeTrait> GroupValues for GroupValuesBytes<O> {
48 fn intern(&mut self, cols: &[ArrayRef], groups: &mut Vec<usize>) -> Result<()> {
49 assert_eq!(cols.len(), 1);
50
51 // look up / add entries in the table
52 let arr = &cols[0];
53
54 groups.clear();
55 self.map.insert_if_new(
56 arr,
57 // called for each new group
58 |_value| {
59 // assign new group index on each insert
60 let group_idx = self.num_groups;
61 self.num_groups += 1;
62 group_idx
63 },
64 // called for each group
65 |group_idx| {
66 groups.push(group_idx);
67 },
68 );
69
70 // ensure we assigned a group to for each row
71 assert_eq!(groups.len(), arr.len());
72 Ok(())
73 }
74
75 fn size(&self) -> usize {
76 self.map.size() + size_of::<Self>()
77 }
78
79 fn is_empty(&self) -> bool {
80 self.num_groups == 0
81 }
82
83 fn len(&self) -> usize {
84 self.num_groups
85 }
86
87 fn emit(&mut self, emit_to: EmitTo) -> Result<Vec<ArrayRef>> {
88 // Reset the map to default, and convert it into a single array
89 let map_contents = self.map.take().into_state();
90
91 let group_values = match emit_to {
92 EmitTo::All => {
93 self.num_groups -= map_contents.len();
94 map_contents
95 }
96 EmitTo::First(n) if n == self.len() => {
97 self.num_groups -= map_contents.len();
98 map_contents
99 }
100 EmitTo::First(n) => {
101 // if we only wanted to take the first n, insert the rest back
102 // into the map we could potentially avoid this reallocation, at
103 // the expense of much more complex code.
104 // see https://github.com/apache/datafusion/issues/9195
105 let emit_group_values = map_contents.slice(0, n);
106 let remaining_group_values =
107 map_contents.slice(n, map_contents.len() - n);
108
109 self.num_groups = 0;
110 let mut group_indexes = vec![];
111 self.intern(&[remaining_group_values], &mut group_indexes)?;
112
113 // Verify that the group indexes were assigned in the correct order
114 assert_eq!(0, group_indexes[0]);
115
116 emit_group_values
117 }
118 };
119
120 Ok(vec![group_values])
121 }
122
123 fn clear_shrink(&mut self, _batch: &RecordBatch) {
124 // in theory we could potentially avoid this reallocation and clear the
125 // contents of the maps, but for now we just reset the map from the beginning
126 self.map.take();
127 }
128}