datafusion_physical_expr/

aggregate.rs

// 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.

pub(crate) mod groups_accumulator {
    #[allow(unused_imports)]
    pub(crate) mod accumulate {
        pub use datafusion_functions_aggregate_common::aggregate::groups_accumulator::accumulate::NullState;
    }
    pub use datafusion_functions_aggregate_common::aggregate::groups_accumulator::{
        accumulate::NullState, GroupsAccumulatorAdapter,
    };
}
pub(crate) mod stats {
    pub use datafusion_functions_aggregate_common::stats::StatsType;
}
pub mod utils {
    #[allow(deprecated)] // allow adjust_output_array
    pub use datafusion_functions_aggregate_common::utils::{
        adjust_output_array, get_accum_scalar_values_as_arrays, get_sort_options,
        ordering_fields, DecimalAverager, Hashable,
    };
}

use std::fmt::Debug;
use std::sync::Arc;

use crate::expressions::Column;

use arrow::compute::SortOptions;
use arrow::datatypes::{DataType, FieldRef, Schema, SchemaRef};
use datafusion_common::{internal_err, not_impl_err, Result, ScalarValue};
use datafusion_expr::{AggregateUDF, ReversedUDAF, SetMonotonicity};
use datafusion_expr_common::accumulator::Accumulator;
use datafusion_expr_common::groups_accumulator::GroupsAccumulator;
use datafusion_expr_common::type_coercion::aggregates::check_arg_count;
use datafusion_functions_aggregate_common::accumulator::{
    AccumulatorArgs, StateFieldsArgs,
};
use datafusion_functions_aggregate_common::order::AggregateOrderSensitivity;
use datafusion_physical_expr_common::physical_expr::PhysicalExpr;
use datafusion_physical_expr_common::sort_expr::PhysicalSortExpr;

/// Builder for physical [`AggregateFunctionExpr`]
///
/// `AggregateFunctionExpr` contains the information necessary to call
/// an aggregate expression.
#[derive(Debug, Clone)]
pub struct AggregateExprBuilder {
    fun: Arc<AggregateUDF>,
    /// Physical expressions of the aggregate function
    args: Vec<Arc<dyn PhysicalExpr>>,
    alias: Option<String>,
    /// A human readable name
    human_display: String,
    /// Arrow Schema for the aggregate function
    schema: SchemaRef,
    /// The physical order by expressions
    order_bys: Vec<PhysicalSortExpr>,
    /// Whether to ignore null values
    ignore_nulls: bool,
    /// Whether is distinct aggregate function
    is_distinct: bool,
    /// Whether the expression is reversed
    is_reversed: bool,
}

impl AggregateExprBuilder {
    pub fn new(fun: Arc<AggregateUDF>, args: Vec<Arc<dyn PhysicalExpr>>) -> Self {
        Self {
            fun,
            args,
            alias: None,
            human_display: String::default(),
            schema: Arc::new(Schema::empty()),
            order_bys: vec![],
            ignore_nulls: false,
            is_distinct: false,
            is_reversed: false,
        }
    }

    /// Constructs an `AggregateFunctionExpr` from the builder
    ///
    /// Note that an [`Self::alias`] must be provided before calling this method.
    ///
    /// # Example: Create an [`AggregateUDF`]
    ///
    /// In the following example, [`AggregateFunctionExpr`] will be built using [`AggregateExprBuilder`]
    /// which provides a build function. Full example could be accessed from the source file.
    ///
    /// ```
    /// # use std::any::Any;
    /// # use std::sync::Arc;
    /// # use arrow::datatypes::{DataType, FieldRef};
    /// # use datafusion_common::{Result, ScalarValue};
    /// # use datafusion_expr::{col, ColumnarValue, Documentation, Signature, Volatility, Expr};
    /// # use datafusion_expr::{AggregateUDFImpl, AggregateUDF, Accumulator, function::{AccumulatorArgs, StateFieldsArgs}};
    /// # use arrow::datatypes::Field;
    /// #
    /// # #[derive(Debug, Clone)]
    /// # struct FirstValueUdf {
    /// #     signature: Signature,
    /// # }
    /// #
    /// # impl FirstValueUdf {
    /// #     fn new() -> Self {
    /// #         Self {
    /// #             signature: Signature::any(1, Volatility::Immutable),
    /// #         }
    /// #     }
    /// # }
    /// #
    /// # impl AggregateUDFImpl for FirstValueUdf {
    /// #     fn as_any(&self) -> &dyn Any {
    /// #         unimplemented!()
    /// #     }
    /// #
    /// #     fn name(&self) -> &str {
    /// #         unimplemented!()
    /// #     }
    /// #
    /// #     fn signature(&self) -> &Signature {
    /// #         unimplemented!()
    /// #     }
    /// #
    /// #     fn return_type(&self, args: &[DataType]) -> Result<DataType> {
    /// #         unimplemented!()
    /// #     }
    /// #
    /// #     fn accumulator(&self, acc_args: AccumulatorArgs) -> Result<Box<dyn Accumulator>> {
    /// #         unimplemented!()
    /// #         }
    /// #     
    /// #     fn state_fields(&self, args: StateFieldsArgs) -> Result<Vec<FieldRef>> {
    /// #         unimplemented!()
    /// #     }
    /// #
    /// #     fn documentation(&self) -> Option<&Documentation> {
    /// #         unimplemented!()
    /// #     }
    /// # }
    /// #
    /// # let first_value = AggregateUDF::from(FirstValueUdf::new());
    /// # let expr = first_value.call(vec![col("a")]);
    /// #
    /// # use datafusion_physical_expr::expressions::Column;
    /// # use datafusion_physical_expr_common::physical_expr::PhysicalExpr;
    /// # use datafusion_physical_expr::aggregate::AggregateExprBuilder;
    /// # use datafusion_physical_expr::expressions::PhysicalSortExpr;
    /// # use datafusion_physical_expr::PhysicalSortRequirement;
    /// #
    /// fn build_aggregate_expr() -> Result<()> {
    ///     let args = vec![Arc::new(Column::new("a", 0)) as Arc<dyn PhysicalExpr>];
    ///     let order_by = vec![PhysicalSortExpr {
    ///         expr: Arc::new(Column::new("x", 1)) as Arc<dyn PhysicalExpr>,
    ///         options: Default::default(),
    ///     }];
    ///
    ///     let first_value = AggregateUDF::from(FirstValueUdf::new());
    ///
    ///     let aggregate_expr = AggregateExprBuilder::new(
    ///         Arc::new(first_value),
    ///         args
    ///     )
    ///     .order_by(order_by)
    ///     .alias("first_a_by_x")
    ///     .ignore_nulls()
    ///     .build()?;
    ///
    ///     Ok(())
    /// }
    /// ```
    ///
    /// This creates a physical expression equivalent to SQL:
    /// `first_value(a ORDER BY x) IGNORE NULLS AS first_a_by_x`
    pub fn build(self) -> Result<AggregateFunctionExpr> {
        let Self {
            fun,
            args,
            alias,
            human_display,
            schema,
            order_bys,
            ignore_nulls,
            is_distinct,
            is_reversed,
        } = self;
        if args.is_empty() {
            return internal_err!("args should not be empty");
        }

        let ordering_types = order_bys
            .iter()
            .map(|e| e.expr.data_type(&schema))
            .collect::<Result<Vec<_>>>()?;

        let ordering_fields = utils::ordering_fields(&order_bys, &ordering_types);

        let input_exprs_fields = args
            .iter()
            .map(|arg| arg.return_field(&schema))
            .collect::<Result<Vec<_>>>()?;

        check_arg_count(
            fun.name(),
            &input_exprs_fields,
            &fun.signature().type_signature,
        )?;

        let return_field = fun.return_field(&input_exprs_fields)?;
        let is_nullable = fun.is_nullable();
        let name = match alias {
            None => {
                return internal_err!(
                    "AggregateExprBuilder::alias must be provided prior to calling build"
                )
            }
            Some(alias) => alias,
        };

        Ok(AggregateFunctionExpr {
            fun: Arc::unwrap_or_clone(fun),
            args,
            return_field,
            name,
            human_display,
            schema: Arc::unwrap_or_clone(schema),
            order_bys,
            ignore_nulls,
            ordering_fields,
            is_distinct,
            input_fields: input_exprs_fields,
            is_reversed,
            is_nullable,
        })
    }

    pub fn alias(mut self, alias: impl Into<String>) -> Self {
        self.alias = Some(alias.into());
        self
    }

    pub fn human_display(mut self, name: String) -> Self {
        self.human_display = name;
        self
    }

    pub fn schema(mut self, schema: SchemaRef) -> Self {
        self.schema = schema;
        self
    }

    pub fn order_by(mut self, order_bys: Vec<PhysicalSortExpr>) -> Self {
        self.order_bys = order_bys;
        self
    }

    pub fn reversed(mut self) -> Self {
        self.is_reversed = true;
        self
    }

    pub fn with_reversed(mut self, is_reversed: bool) -> Self {
        self.is_reversed = is_reversed;
        self
    }

    pub fn distinct(mut self) -> Self {
        self.is_distinct = true;
        self
    }

    pub fn with_distinct(mut self, is_distinct: bool) -> Self {
        self.is_distinct = is_distinct;
        self
    }

    pub fn ignore_nulls(mut self) -> Self {
        self.ignore_nulls = true;
        self
    }

    pub fn with_ignore_nulls(mut self, ignore_nulls: bool) -> Self {
        self.ignore_nulls = ignore_nulls;
        self
    }
}

/// Physical aggregate expression of a UDAF.
///
/// Instances are constructed via [`AggregateExprBuilder`].
#[derive(Debug, Clone)]
pub struct AggregateFunctionExpr {
    fun: AggregateUDF,
    args: Vec<Arc<dyn PhysicalExpr>>,
    /// Output / return field of this aggregate
    return_field: FieldRef,
    /// Output column name that this expression creates
    name: String,
    /// Simplified name for `tree` explain.
    human_display: String,
    schema: Schema,
    // The physical order by expressions
    order_bys: Vec<PhysicalSortExpr>,
    // Whether to ignore null values
    ignore_nulls: bool,
    // fields used for order sensitive aggregation functions
    ordering_fields: Vec<FieldRef>,
    is_distinct: bool,
    is_reversed: bool,
    input_fields: Vec<FieldRef>,
    is_nullable: bool,
}

impl AggregateFunctionExpr {
    /// Return the `AggregateUDF` used by this `AggregateFunctionExpr`
    pub fn fun(&self) -> &AggregateUDF {
        &self.fun
    }

    /// expressions that are passed to the Accumulator.
    /// Single-column aggregations such as `sum` return a single value, others (e.g. `cov`) return many.
    pub fn expressions(&self) -> Vec<Arc<dyn PhysicalExpr>> {
        self.args.clone()
    }

    /// Human readable name such as `"MIN(c2)"`.
    pub fn name(&self) -> &str {
        &self.name
    }

    /// Simplified name for `tree` explain.
    pub fn human_display(&self) -> &str {
        &self.human_display
    }

    /// Return if the aggregation is distinct
    pub fn is_distinct(&self) -> bool {
        self.is_distinct
    }

    /// Return if the aggregation ignores nulls
    pub fn ignore_nulls(&self) -> bool {
        self.ignore_nulls
    }

    /// Return if the aggregation is reversed
    pub fn is_reversed(&self) -> bool {
        self.is_reversed
    }

    /// Return if the aggregation is nullable
    pub fn is_nullable(&self) -> bool {
        self.is_nullable
    }

    /// the field of the final result of this aggregation.
    pub fn field(&self) -> FieldRef {
        self.return_field
            .as_ref()
            .clone()
            .with_name(&self.name)
            .into()
    }

    /// the accumulator used to accumulate values from the expressions.
    /// the accumulator expects the same number of arguments as `expressions` and must
    /// return states with the same description as `state_fields`
    pub fn create_accumulator(&self) -> Result<Box<dyn Accumulator>> {
        let acc_args = AccumulatorArgs {
            return_field: Arc::clone(&self.return_field),
            schema: &self.schema,
            ignore_nulls: self.ignore_nulls,
            order_bys: self.order_bys.as_ref(),
            is_distinct: self.is_distinct,
            name: &self.name,
            is_reversed: self.is_reversed,
            exprs: &self.args,
        };

        self.fun.accumulator(acc_args)
    }

    /// the field of the final result of this aggregation.
    pub fn state_fields(&self) -> Result<Vec<FieldRef>> {
        let args = StateFieldsArgs {
            name: &self.name,
            input_fields: &self.input_fields,
            return_field: Arc::clone(&self.return_field),
            ordering_fields: &self.ordering_fields,
            is_distinct: self.is_distinct,
        };

        self.fun.state_fields(args)
    }

    /// Returns the ORDER BY expressions for the aggregate function.
    pub fn order_bys(&self) -> &[PhysicalSortExpr] {
        if self.order_sensitivity().is_insensitive() {
            &[]
        } else {
            &self.order_bys
        }
    }

    /// Indicates whether aggregator can produce the correct result with any
    /// arbitrary input ordering. By default, we assume that aggregate expressions
    /// are order insensitive.
    pub fn order_sensitivity(&self) -> AggregateOrderSensitivity {
        if self.order_bys.is_empty() {
            AggregateOrderSensitivity::Insensitive
        } else {
            // If there is an ORDER BY clause, use the sensitivity of the implementation:
            self.fun.order_sensitivity()
        }
    }

    /// Sets the indicator whether ordering requirements of the aggregator is
    /// satisfied by its input. If this is not the case, aggregators with order
    /// sensitivity `AggregateOrderSensitivity::Beneficial` can still produce
    /// the correct result with possibly more work internally.
    ///
    /// # Returns
    ///
    /// Returns `Ok(Some(updated_expr))` if the process completes successfully.
    /// If the expression can benefit from existing input ordering, but does
    /// not implement the method, returns an error. Order insensitive and hard
    /// requirement aggregators return `Ok(None)`.
    pub fn with_beneficial_ordering(
        self: Arc<Self>,
        beneficial_ordering: bool,
    ) -> Result<Option<AggregateFunctionExpr>> {
        let Some(updated_fn) = self
            .fun
            .clone()
            .with_beneficial_ordering(beneficial_ordering)?
        else {
            return Ok(None);
        };

        AggregateExprBuilder::new(Arc::new(updated_fn), self.args.to_vec())
            .order_by(self.order_bys.clone())
            .schema(Arc::new(self.schema.clone()))
            .alias(self.name().to_string())
            .with_ignore_nulls(self.ignore_nulls)
            .with_distinct(self.is_distinct)
            .with_reversed(self.is_reversed)
            .build()
            .map(Some)
    }

    /// Creates accumulator implementation that supports retract
    pub fn create_sliding_accumulator(&self) -> Result<Box<dyn Accumulator>> {
        let args = AccumulatorArgs {
            return_field: Arc::clone(&self.return_field),
            schema: &self.schema,
            ignore_nulls: self.ignore_nulls,
            order_bys: self.order_bys.as_ref(),
            is_distinct: self.is_distinct,
            name: &self.name,
            is_reversed: self.is_reversed,
            exprs: &self.args,
        };

        let accumulator = self.fun.create_sliding_accumulator(args)?;

        // Accumulators that have window frame startings different
        // than `UNBOUNDED PRECEDING`, such as `1 PRECEDING`, need to
        // implement retract_batch method in order to run correctly
        // currently in DataFusion.
        //
        // If this `retract_batches` is not present, there is no way
        // to calculate result correctly. For example, the query
        //
        // ```sql
        // SELECT
        //  SUM(a) OVER(ORDER BY a ROWS BETWEEN 1 PRECEDING AND 1 FOLLOWING) AS sum_a
        // FROM
        //  t
        // ```
        //
        // 1. First sum value will be the sum of rows between `[0, 1)`,
        //
        // 2. Second sum value will be the sum of rows between `[0, 2)`
        //
        // 3. Third sum value will be the sum of rows between `[1, 3)`, etc.
        //
        // Since the accumulator keeps the running sum:
        //
        // 1. First sum we add to the state sum value between `[0, 1)`
        //
        // 2. Second sum we add to the state sum value between `[1, 2)`
        // (`[0, 1)` is already in the state sum, hence running sum will
        // cover `[0, 2)` range)
        //
        // 3. Third sum we add to the state sum value between `[2, 3)`
        // (`[0, 2)` is already in the state sum).  Also we need to
        // retract values between `[0, 1)` by this way we can obtain sum
        // between [1, 3) which is indeed the appropriate range.
        //
        // When we use `UNBOUNDED PRECEDING` in the query starting
        // index will always be 0 for the desired range, and hence the
        // `retract_batch` method will not be called. In this case
        // having retract_batch is not a requirement.
        //
        // This approach is a a bit different than window function
        // approach. In window function (when they use a window frame)
        // they get all the desired range during evaluation.
        if !accumulator.supports_retract_batch() {
            return not_impl_err!(
                "Aggregate can not be used as a sliding accumulator because \
                     `retract_batch` is not implemented: {}",
                self.name
            );
        }
        Ok(accumulator)
    }

    /// If the aggregate expression has a specialized
    /// [`GroupsAccumulator`] implementation. If this returns true,
    /// `[Self::create_groups_accumulator`] will be called.
    pub fn groups_accumulator_supported(&self) -> bool {
        let args = AccumulatorArgs {
            return_field: Arc::clone(&self.return_field),
            schema: &self.schema,
            ignore_nulls: self.ignore_nulls,
            order_bys: self.order_bys.as_ref(),
            is_distinct: self.is_distinct,
            name: &self.name,
            is_reversed: self.is_reversed,
            exprs: &self.args,
        };
        self.fun.groups_accumulator_supported(args)
    }

    /// Return a specialized [`GroupsAccumulator`] that manages state
    /// for all groups.
    ///
    /// For maximum performance, a [`GroupsAccumulator`] should be
    /// implemented in addition to [`Accumulator`].
    pub fn create_groups_accumulator(&self) -> Result<Box<dyn GroupsAccumulator>> {
        let args = AccumulatorArgs {
            return_field: Arc::clone(&self.return_field),
            schema: &self.schema,
            ignore_nulls: self.ignore_nulls,
            order_bys: self.order_bys.as_ref(),
            is_distinct: self.is_distinct,
            name: &self.name,
            is_reversed: self.is_reversed,
            exprs: &self.args,
        };
        self.fun.create_groups_accumulator(args)
    }

    /// Construct an expression that calculates the aggregate in reverse.
    /// Typically the "reverse" expression is itself (e.g. SUM, COUNT).
    /// For aggregates that do not support calculation in reverse,
    /// returns None (which is the default value).
    pub fn reverse_expr(&self) -> Option<AggregateFunctionExpr> {
        match self.fun.reverse_udf() {
            ReversedUDAF::NotSupported => None,
            ReversedUDAF::Identical => Some(self.clone()),
            ReversedUDAF::Reversed(reverse_udf) => {
                let mut name = self.name().to_string();
                // If the function is changed, we need to reverse order_by clause as well
                // i.e. First(a order by b asc null first) -> Last(a order by b desc null last)
                if self.fun().name() != reverse_udf.name() {
                    replace_order_by_clause(&mut name);
                }
                replace_fn_name_clause(&mut name, self.fun.name(), reverse_udf.name());

                AggregateExprBuilder::new(reverse_udf, self.args.to_vec())
                    .order_by(self.order_bys.iter().map(|e| e.reverse()).collect())
                    .schema(Arc::new(self.schema.clone()))
                    .alias(name)
                    .with_ignore_nulls(self.ignore_nulls)
                    .with_distinct(self.is_distinct)
                    .with_reversed(!self.is_reversed)
                    .build()
                    .ok()
            }
        }
    }

    /// Returns all expressions used in the [`AggregateFunctionExpr`].
    /// These expressions are  (1)function arguments, (2) order by expressions.
    pub fn all_expressions(&self) -> AggregatePhysicalExpressions {
        let args = self.expressions();
        let order_by_exprs = self
            .order_bys()
            .iter()
            .map(|sort_expr| Arc::clone(&sort_expr.expr))
            .collect();
        AggregatePhysicalExpressions {
            args,
            order_by_exprs,
        }
    }

    /// Rewrites [`AggregateFunctionExpr`], with new expressions given. The argument should be consistent
    /// with the return value of the [`AggregateFunctionExpr::all_expressions`] method.
    /// Returns `Some(Arc<dyn AggregateExpr>)` if re-write is supported, otherwise returns `None`.
    pub fn with_new_expressions(
        &self,
        _args: Vec<Arc<dyn PhysicalExpr>>,
        _order_by_exprs: Vec<Arc<dyn PhysicalExpr>>,
    ) -> Option<AggregateFunctionExpr> {
        None
    }

    /// If this function is max, return (output_field, true)
    /// if the function is min, return (output_field, false)
    /// otherwise return None (the default)
    ///
    /// output_field is the name of the column produced by this aggregate
    ///
    /// Note: this is used to use special aggregate implementations in certain conditions
    pub fn get_minmax_desc(&self) -> Option<(FieldRef, bool)> {
        self.fun.is_descending().map(|flag| (self.field(), flag))
    }

    /// Returns default value of the function given the input is Null
    /// Most of the aggregate function return Null if input is Null,
    /// while `count` returns 0 if input is Null
    pub fn default_value(&self, data_type: &DataType) -> Result<ScalarValue> {
        self.fun.default_value(data_type)
    }

    /// Indicates whether the aggregation function is monotonic as a set
    /// function. See [`SetMonotonicity`] for details.
    pub fn set_monotonicity(&self) -> SetMonotonicity {
        let field = self.field();
        let data_type = field.data_type();
        self.fun.inner().set_monotonicity(data_type)
    }

    /// Returns `PhysicalSortExpr` based on the set monotonicity of the function.
    pub fn get_result_ordering(&self, aggr_func_idx: usize) -> Option<PhysicalSortExpr> {
        // If the aggregate expressions are set-monotonic, the output data is
        // naturally ordered with it per group or partition.
        let monotonicity = self.set_monotonicity();
        if monotonicity == SetMonotonicity::NotMonotonic {
            return None;
        }
        let expr = Arc::new(Column::new(self.name(), aggr_func_idx));
        let options =
            SortOptions::new(monotonicity == SetMonotonicity::Decreasing, false);
        Some(PhysicalSortExpr { expr, options })
    }
}

/// Stores the physical expressions used inside the `AggregateExpr`.
pub struct AggregatePhysicalExpressions {
    /// Aggregate function arguments
    pub args: Vec<Arc<dyn PhysicalExpr>>,
    /// Order by expressions
    pub order_by_exprs: Vec<Arc<dyn PhysicalExpr>>,
}

impl PartialEq for AggregateFunctionExpr {
    fn eq(&self, other: &Self) -> bool {
        self.name == other.name
            && self.return_field == other.return_field
            && self.fun == other.fun
            && self.args.len() == other.args.len()
            && self
                .args
                .iter()
                .zip(other.args.iter())
                .all(|(this_arg, other_arg)| this_arg.eq(other_arg))
    }
}

fn replace_order_by_clause(order_by: &mut String) {
    let suffixes = [
        (" DESC NULLS FIRST]", " ASC NULLS LAST]"),
        (" ASC NULLS FIRST]", " DESC NULLS LAST]"),
        (" DESC NULLS LAST]", " ASC NULLS FIRST]"),
        (" ASC NULLS LAST]", " DESC NULLS FIRST]"),
    ];

    if let Some(start) = order_by.find("ORDER BY [") {
        if let Some(end) = order_by[start..].find(']') {
            let order_by_start = start + 9;
            let order_by_end = start + end;

            let column_order = &order_by[order_by_start..=order_by_end];
            for (suffix, replacement) in suffixes {
                if column_order.ends_with(suffix) {
                    let new_order = column_order.replace(suffix, replacement);
                    order_by.replace_range(order_by_start..=order_by_end, &new_order);
                    break;
                }
            }
        }
    }
}

fn replace_fn_name_clause(aggr_name: &mut String, fn_name_old: &str, fn_name_new: &str) {
    *aggr_name = aggr_name.replace(fn_name_old, fn_name_new);
}