datafusion_physical_expr/window/
aggregate.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
18//! Physical exec for aggregate window function expressions.
19
20use std::any::Any;
21use std::ops::Range;
22use std::sync::Arc;
23
24use crate::aggregate::AggregateFunctionExpr;
25use crate::window::standard::add_new_ordering_expr_with_partition_by;
26use crate::window::window_expr::AggregateWindowExpr;
27use crate::window::{
28    PartitionBatches, PartitionWindowAggStates, SlidingAggregateWindowExpr, WindowExpr,
29};
30use crate::{EquivalenceProperties, PhysicalExpr};
31
32use arrow::array::Array;
33use arrow::array::ArrayRef;
34use arrow::datatypes::FieldRef;
35use arrow::record_batch::RecordBatch;
36use datafusion_common::{DataFusionError, Result, ScalarValue};
37use datafusion_expr::{Accumulator, WindowFrame, WindowFrameBound, WindowFrameUnits};
38use datafusion_physical_expr_common::sort_expr::PhysicalSortExpr;
39
40/// A window expr that takes the form of an aggregate function.
41///
42/// See comments on [`WindowExpr`] for more details.
43#[derive(Debug)]
44pub struct PlainAggregateWindowExpr {
45    aggregate: Arc<AggregateFunctionExpr>,
46    partition_by: Vec<Arc<dyn PhysicalExpr>>,
47    order_by: Vec<PhysicalSortExpr>,
48    window_frame: Arc<WindowFrame>,
49    is_constant_in_partition: bool,
50}
51
52impl PlainAggregateWindowExpr {
53    /// Create a new aggregate window function expression
54    pub fn new(
55        aggregate: Arc<AggregateFunctionExpr>,
56        partition_by: &[Arc<dyn PhysicalExpr>],
57        order_by: &[PhysicalSortExpr],
58        window_frame: Arc<WindowFrame>,
59    ) -> Self {
60        let is_constant_in_partition =
61            Self::is_window_constant_in_partition(order_by, &window_frame);
62        Self {
63            aggregate,
64            partition_by: partition_by.to_vec(),
65            order_by: order_by.to_vec(),
66            window_frame,
67            is_constant_in_partition,
68        }
69    }
70
71    /// Get aggregate expr of AggregateWindowExpr
72    pub fn get_aggregate_expr(&self) -> &AggregateFunctionExpr {
73        &self.aggregate
74    }
75
76    pub fn add_equal_orderings(
77        &self,
78        eq_properties: &mut EquivalenceProperties,
79        window_expr_index: usize,
80    ) -> Result<()> {
81        if let Some(expr) = self
82            .get_aggregate_expr()
83            .get_result_ordering(window_expr_index)
84        {
85            add_new_ordering_expr_with_partition_by(
86                eq_properties,
87                expr,
88                &self.partition_by,
89            )?;
90        }
91        Ok(())
92    }
93
94    // Returns true if every row in the partition has the same window frame. This allows
95    // for preventing bound + function calculation for every row due to the values being the
96    // same.
97    //
98    // This occurs when both bounds fall under either condition below:
99    //  1. Bound is unbounded (`Preceding` or `Following`)
100    //  2. Bound is `CurrentRow` while using `Range` units with no order by clause
101    //  This results in an invalid range specification. Following PostgreSQL’s convention,
102    //  we interpret this as the entire partition being used for the current window frame.
103    fn is_window_constant_in_partition(
104        order_by: &[PhysicalSortExpr],
105        window_frame: &WindowFrame,
106    ) -> bool {
107        let is_constant_bound = |bound: &WindowFrameBound| match bound {
108            WindowFrameBound::CurrentRow => {
109                window_frame.units == WindowFrameUnits::Range && order_by.is_empty()
110            }
111            _ => bound.is_unbounded(),
112        };
113
114        is_constant_bound(&window_frame.start_bound)
115            && is_constant_bound(&window_frame.end_bound)
116    }
117}
118
119/// peer based evaluation based on the fact that batch is pre-sorted given the sort columns
120/// and then per partition point we'll evaluate the peer group (e.g. SUM or MAX gives the same
121/// results for peers) and concatenate the results.
122impl WindowExpr for PlainAggregateWindowExpr {
123    /// Return a reference to Any that can be used for downcasting
124    fn as_any(&self) -> &dyn Any {
125        self
126    }
127
128    fn field(&self) -> Result<FieldRef> {
129        Ok(self.aggregate.field())
130    }
131
132    fn name(&self) -> &str {
133        self.aggregate.name()
134    }
135
136    fn expressions(&self) -> Vec<Arc<dyn PhysicalExpr>> {
137        self.aggregate.expressions()
138    }
139
140    fn evaluate(&self, batch: &RecordBatch) -> Result<ArrayRef> {
141        self.aggregate_evaluate(batch)
142    }
143
144    fn evaluate_stateful(
145        &self,
146        partition_batches: &PartitionBatches,
147        window_agg_state: &mut PartitionWindowAggStates,
148    ) -> Result<()> {
149        self.aggregate_evaluate_stateful(partition_batches, window_agg_state)?;
150
151        // Update window frame range for each partition. As we know that
152        // non-sliding aggregations will never call `retract_batch`, this value
153        // can safely increase, and we can remove "old" parts of the state.
154        // This enables us to run queries involving UNBOUNDED PRECEDING frames
155        // using bounded memory for suitable aggregations.
156        for partition_row in partition_batches.keys() {
157            let window_state =
158                window_agg_state.get_mut(partition_row).ok_or_else(|| {
159                    DataFusionError::Execution("Cannot find state".to_string())
160                })?;
161            let state = &mut window_state.state;
162            if self.window_frame.start_bound.is_unbounded() {
163                state.window_frame_range.start =
164                    state.window_frame_range.end.saturating_sub(1);
165            }
166        }
167        Ok(())
168    }
169
170    fn partition_by(&self) -> &[Arc<dyn PhysicalExpr>] {
171        &self.partition_by
172    }
173
174    fn order_by(&self) -> &[PhysicalSortExpr] {
175        &self.order_by
176    }
177
178    fn get_window_frame(&self) -> &Arc<WindowFrame> {
179        &self.window_frame
180    }
181
182    fn get_reverse_expr(&self) -> Option<Arc<dyn WindowExpr>> {
183        self.aggregate.reverse_expr().map(|reverse_expr| {
184            let reverse_window_frame = self.window_frame.reverse();
185            if reverse_window_frame.is_ever_expanding() {
186                Arc::new(PlainAggregateWindowExpr::new(
187                    Arc::new(reverse_expr),
188                    &self.partition_by.clone(),
189                    &self
190                        .order_by
191                        .iter()
192                        .map(|e| e.reverse())
193                        .collect::<Vec<_>>(),
194                    Arc::new(self.window_frame.reverse()),
195                )) as _
196            } else {
197                Arc::new(SlidingAggregateWindowExpr::new(
198                    Arc::new(reverse_expr),
199                    &self.partition_by.clone(),
200                    &self
201                        .order_by
202                        .iter()
203                        .map(|e| e.reverse())
204                        .collect::<Vec<_>>(),
205                    Arc::new(self.window_frame.reverse()),
206                )) as _
207            }
208        })
209    }
210
211    fn uses_bounded_memory(&self) -> bool {
212        !self.window_frame.end_bound.is_unbounded()
213    }
214}
215
216impl AggregateWindowExpr for PlainAggregateWindowExpr {
217    fn get_accumulator(&self) -> Result<Box<dyn Accumulator>> {
218        self.aggregate.create_accumulator()
219    }
220
221    /// For a given range, calculate accumulation result inside the range on
222    /// `value_slice` and update accumulator state.
223    // We assume that `cur_range` contains `last_range` and their start points
224    // are same. In summary if `last_range` is `Range{start: a,end: b}` and
225    // `cur_range` is `Range{start: a1, end: b1}`, it is guaranteed that a1=a and b1>=b.
226    fn get_aggregate_result_inside_range(
227        &self,
228        last_range: &Range<usize>,
229        cur_range: &Range<usize>,
230        value_slice: &[ArrayRef],
231        accumulator: &mut Box<dyn Accumulator>,
232    ) -> Result<ScalarValue> {
233        if cur_range.start == cur_range.end {
234            self.aggregate
235                .default_value(self.aggregate.field().data_type())
236        } else {
237            // Accumulate any new rows that have entered the window:
238            let update_bound = cur_range.end - last_range.end;
239            // A non-sliding aggregation only processes new data, it never
240            // deals with expiring data as its starting point is always the
241            // same point (i.e. the beginning of the table/frame). Hence, we
242            // do not call `retract_batch`.
243            if update_bound > 0 {
244                let update: Vec<ArrayRef> = value_slice
245                    .iter()
246                    .map(|v| v.slice(last_range.end, update_bound))
247                    .collect();
248                accumulator.update_batch(&update)?
249            }
250            accumulator.evaluate()
251        }
252    }
253
254    fn is_constant_in_partition(&self) -> bool {
255        self.is_constant_in_partition
256    }
257}
datafusion_physical_expr/window/aggregate.rs

datafusion_physical_expr/window/
aggregate.rs
