datafusion_physical_optimizer/

limit_pushdown.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.

//! [`LimitPushdown`] pushes `LIMIT` down through `ExecutionPlan`s to reduce
//! data transfer as much as possible.
//!
//! # Plan Limit Absorption
//! In addition to pushing down `GlobalLimitExec` and `LocalLimitExec` nodes in
//! the plan, some operators can "absorb" a limit and stop early during
//! execution.
//!
//! ## Background: vectorized volcano execution model
//! DataFusion uses a batched volcano model. For most operators, output is
//! produced in batches of `datafusion.execution.batch_size` (default 8192), so
//! the batch sizes typically look like:
//! ```text
//! 8192, 8192, ..., 8192, 100 (the final batch may be partial)
//! ```
//!
//! ## Example
//! For a join with an expensive, selective predicate:
//! ```text
//! GlobalLimitExec: skip=0, fetch=10
//! -- NestedLoopJoinExec(on=expr_expensive_and_selective)
//! --- DataSourceExec()
//! --- DataSourceExec()
//! ```
//!
//! Under this model, `NestedLoopJoinExec` would keep working until it can emit
//! a full batch (8192 rows), even though the query only needs 10. If the limit
//! cannot be pushed below the join, we can still embed it inside the join so it
//! stops once the limit is satisfied. The transformed plan looks like:
//!
//! ```text
//! NestedLoopJoinExec(on=expr_expensive_and_selective, fetch=10)
//! --- DataSourceExec()
//! --- DataSourceExec()
//! ```
//!
//! ## Implementation
//! The current optimizer rule optionally pushes `fetch` requirements into
//! operators via [`ExecutionPlan::with_fetch`].
//!
//! To support early termination in operators, [`LimitedBatchCoalescer`](https://docs.rs/datafusion/latest/datafusion/physical_plan/coalesce/struct.LimitedBatchCoalescer.html)
//! can help manage the output buffer.
//!
//! Reference implementation in Hash Join: <https://github.com/apache/datafusion/pull/20228>

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

use crate::PhysicalOptimizerRule;

use datafusion_common::config::ConfigOptions;
use datafusion_common::error::Result;
use datafusion_common::stats::Precision;
use datafusion_common::tree_node::{Transformed, TreeNodeRecursion};
use datafusion_common::utils::combine_limit;
use datafusion_physical_plan::coalesce_partitions::CoalescePartitionsExec;
use datafusion_physical_plan::empty::EmptyExec;
use datafusion_physical_plan::limit::{GlobalLimitExec, LocalLimitExec};
use datafusion_physical_plan::placeholder_row::PlaceholderRowExec;
use datafusion_physical_plan::projection::ProjectionExec;
use datafusion_physical_plan::sorts::sort_preserving_merge::SortPreservingMergeExec;
use datafusion_physical_plan::{ExecutionPlan, ExecutionPlanProperties};
/// This rule inspects [`ExecutionPlan`]'s and pushes down the fetch limit from
/// the parent to the child if applicable.
#[derive(Default, Debug)]
pub struct LimitPushdown {}

/// This is a "data class" we use within the [`LimitPushdown`] rule to push
/// down limits in the plan. GlobalRequirements are hold as a rule-wide state
/// and holds the fetch and skip information. The struct also has a field named
/// satisfied which means if the "current" plan is valid in terms of limits or not.
///
/// For example: If the plan is satisfied with current fetch info, we decide to not add a LocalLimit
///
/// [`LimitPushdown`]: crate::limit_pushdown::LimitPushdown
#[derive(Default, Clone, Debug)]
pub struct GlobalRequirements {
    fetch: Option<usize>,
    skip: usize,
    satisfied: bool,
    preserve_order: bool,
}

impl LimitPushdown {
    #[expect(missing_docs)]
    pub fn new() -> Self {
        Self {}
    }
}

impl PhysicalOptimizerRule for LimitPushdown {
    fn optimize(
        &self,
        plan: Arc<dyn ExecutionPlan>,
        _config: &ConfigOptions,
    ) -> Result<Arc<dyn ExecutionPlan>> {
        let global_state = GlobalRequirements {
            fetch: None,
            skip: 0,
            satisfied: false,
            preserve_order: false,
        };
        pushdown_limits(plan, global_state)
    }

    fn name(&self) -> &str {
        "LimitPushdown"
    }

    fn schema_check(&self) -> bool {
        true
    }
}

struct LimitInfo {
    input: Arc<dyn ExecutionPlan>,
    fetch: Option<usize>,
    skip: usize,
    preserve_order: bool,
}

/// This function is the main helper function of the `LimitPushDown` rule.
/// The helper takes an `ExecutionPlan` and a global (algorithm) state which is
/// an instance of `GlobalRequirements` and modifies these parameters while
/// checking if the limits can be pushed down or not.
///
/// If a limit is encountered, a [`TreeNodeRecursion::Stop`] is returned. Otherwise,
/// return a [`TreeNodeRecursion::Continue`].
pub fn pushdown_limit_helper(
    mut pushdown_plan: Arc<dyn ExecutionPlan>,
    mut global_state: GlobalRequirements,
) -> Result<(Transformed<Arc<dyn ExecutionPlan>>, GlobalRequirements)> {
    // Extract limit, if exist, and return child inputs.
    if let Some(limit_info) = extract_limit(&pushdown_plan) {
        // If we have fetch/skip info in the global state already, we need to
        // decide which one to continue with:
        let (skip, fetch) = combine_limit(
            global_state.skip,
            global_state.fetch,
            limit_info.skip,
            limit_info.fetch,
        );
        global_state.skip = skip;
        global_state.fetch = fetch;
        global_state.preserve_order = limit_info.preserve_order;
        global_state.satisfied = false;

        if let Some(fetch) = fetch
            && limit_satisfied_by_input(&limit_info.input, skip, fetch)?
        {
            // The input already produces at most `fetch` rows, so no new limit
            // node is needed. Mark satisfied so downstream won't re-add one,
            // but preserve skip/fetch so any nested limit nodes (e.g. an inner
            // GlobalLimitExec) can still be merged with the outer constraint.
            global_state.satisfied = true;

            return Ok((
                Transformed {
                    data: limit_info.input,
                    transformed: true,
                    tnr: TreeNodeRecursion::Stop,
                },
                global_state,
            ));
        }

        // Now the global state has the most recent information, we can remove
        // the limit node. We will decide later if we should add it again or
        // not.
        return Ok((
            Transformed {
                data: limit_info.input,
                transformed: true,
                tnr: TreeNodeRecursion::Stop,
            },
            global_state,
        ));
    }

    // If we have a non-limit operator with fetch capability, update global
    // state as necessary:
    if pushdown_plan.fetch().is_some() {
        if global_state.skip == 0 {
            global_state.satisfied = true;
        }
        (global_state.skip, global_state.fetch) = combine_limit(
            global_state.skip,
            global_state.fetch,
            0,
            pushdown_plan.fetch(),
        );
    }

    let Some(global_fetch) = global_state.fetch else {
        // There's no valid fetch information, exit early:
        return if global_state.skip > 0 && !global_state.satisfied {
            // There might be a case with only offset, if so add a global limit:
            global_state.satisfied = true;
            Ok((
                Transformed::yes(add_global_limit(
                    pushdown_plan,
                    global_state.skip,
                    None,
                )),
                global_state,
            ))
        } else {
            // There's no info on offset or fetch, nothing to do:
            Ok((Transformed::no(pushdown_plan), global_state))
        };
    };

    let skip_and_fetch = Some(global_fetch + global_state.skip);

    if pushdown_plan.supports_limit_pushdown() {
        if !combines_input_partitions(&pushdown_plan) {
            // We have information in the global state and the plan pushes down,
            // continue:
            Ok((Transformed::no(pushdown_plan), global_state))
        } else if let Some(plan_with_fetch) = pushdown_plan.with_fetch(skip_and_fetch) {
            // This plan is combining input partitions, so we need to add the
            // fetch info to plan if possible. If not, we must add a limit node
            // with the information from the global state.
            let mut new_plan = plan_with_fetch;
            // Execution plans can't (yet) handle skip, so if we have one,
            // we still need to add a global limit
            if global_state.skip > 0 {
                new_plan =
                    add_global_limit(new_plan, global_state.skip, global_state.fetch);
            }
            global_state.fetch = skip_and_fetch;
            global_state.skip = 0;
            global_state.satisfied = true;
            Ok((Transformed::yes(new_plan), global_state))
        } else if global_state.satisfied {
            // If the plan is already satisfied, do not add a limit:
            Ok((Transformed::no(pushdown_plan), global_state))
        } else {
            global_state.satisfied = true;
            Ok((
                Transformed::yes(add_limit(
                    pushdown_plan,
                    global_state.skip,
                    global_fetch,
                )),
                global_state,
            ))
        }
    } else {
        // The plan does not support push down and it is not a limit. We will need
        // to add a limit or a fetch. If the plan is already satisfied, we will try
        // to add the fetch info and return the plan.

        // There's no push down, change fetch & skip to default values:
        let global_skip = global_state.skip;
        global_state.fetch = None;
        global_state.skip = 0;

        let maybe_fetchable = pushdown_plan.with_fetch(skip_and_fetch);
        if global_state.satisfied {
            if let Some(plan_with_fetch) = maybe_fetchable {
                let plan_with_preserve_order = plan_with_fetch
                    .with_preserve_order(global_state.preserve_order)
                    .unwrap_or(plan_with_fetch);
                Ok((Transformed::yes(plan_with_preserve_order), global_state))
            } else {
                Ok((Transformed::no(pushdown_plan), global_state))
            }
        } else {
            global_state.satisfied = true;
            pushdown_plan = if let Some(plan_with_fetch) = maybe_fetchable {
                let plan_with_preserve_order = plan_with_fetch
                    .with_preserve_order(global_state.preserve_order)
                    .unwrap_or(plan_with_fetch);

                if global_skip > 0 {
                    add_global_limit(
                        plan_with_preserve_order,
                        global_skip,
                        Some(global_fetch),
                    )
                } else {
                    plan_with_preserve_order
                }
            } else {
                add_limit(pushdown_plan, global_skip, global_fetch)
            };
            Ok((Transformed::yes(pushdown_plan), global_state))
        }
    }
}

/// Returns true if exact input statistics prove that applying the limit would
/// not remove any rows.
fn limit_satisfied_by_input(
    plan: &Arc<dyn ExecutionPlan>,
    skip: usize,
    fetch: usize,
) -> Result<bool> {
    if skip > 0 {
        return Ok(false);
    }

    if plan.output_partitioning().partition_count() != 1 {
        return Ok(false);
    }

    let Some(num_rows) = limit_eliminable_exact_num_rows(plan)? else {
        return Ok(false);
    };

    Ok(num_rows <= fetch)
}

/// Returns exact row counts only from a conservative whitelist of operators
/// whose row-count guarantees are strong enough to remove a limit.
fn limit_eliminable_exact_num_rows(
    plan: &Arc<dyn ExecutionPlan>,
) -> Result<Option<usize>> {
    // Unwrap any wrapping ProjectionExec layers; projections preserve row count
    // but may derive statistics in ways that are not trustworthy, so we peek
    // through them to the underlying producer.
    let mut current = plan;
    while let Some(projection) = current.downcast_ref::<ProjectionExec>() {
        current = projection.input();
    }

    if current.is::<EmptyExec>() {
        return Ok(Some(0));
    }

    if current.is::<PlaceholderRowExec>() {
        return Ok(Some(1));
    }

    if matches!(
        current.partition_statistics(None)?.num_rows,
        Precision::Exact(0)
    ) {
        return Ok(Some(0));
    }

    Ok(None)
}

/// Pushes down the limit through the plan.
pub(crate) fn pushdown_limits(
    pushdown_plan: Arc<dyn ExecutionPlan>,
    global_state: GlobalRequirements,
) -> Result<Arc<dyn ExecutionPlan>> {
    // Call pushdown_limit_helper.
    // This will either extract the limit node (returning the child), or apply the limit pushdown.
    let (mut new_node, mut global_state) =
        pushdown_limit_helper(pushdown_plan, global_state)?;

    // While limits exist, continue combining the global_state.
    while new_node.tnr == TreeNodeRecursion::Stop {
        (new_node, global_state) = pushdown_limit_helper(new_node.data, global_state)?;
    }

    // Once a limit has been materialized above the current node, child
    // subtrees should not inherit its `skip`. Keep `fetch`, but clear
    // `skip` before recursing so child-local limits are not merged with
    // an `OFFSET` that has already been applied.
    if global_state.satisfied {
        global_state.skip = 0;
    }

    // Apply pushdown limits in children
    let children = new_node.data.children();
    let mut changed = false;
    let new_children = children
        .into_iter()
        .map(|child: &Arc<dyn ExecutionPlan>| {
            let new_child = pushdown_limits(
                Arc::<dyn ExecutionPlan>::clone(child),
                global_state.clone(),
            )?;
            // Tracking if any of the children changed
            changed |= !Arc::ptr_eq(child, &new_child);
            Ok(new_child)
        })
        .collect::<Result<_>>()?;

    if changed {
        new_node.data.with_new_children(new_children)
    } else {
        Ok(new_node.data)
    }
}

/// Extracts limit information from the [`ExecutionPlan`] if it is a
/// [`GlobalLimitExec`] or a [`LocalLimitExec`].
fn extract_limit(plan: &Arc<dyn ExecutionPlan>) -> Option<LimitInfo> {
    if let Some(global_limit) = plan.downcast_ref::<GlobalLimitExec>() {
        Some(LimitInfo {
            input: Arc::clone(global_limit.input()),
            fetch: global_limit.fetch(),
            skip: global_limit.skip(),
            preserve_order: global_limit.required_ordering().is_some(),
        })
    } else {
        plan.downcast_ref::<LocalLimitExec>()
            .map(|local_limit| LimitInfo {
                input: Arc::clone(local_limit.input()),
                fetch: Some(local_limit.fetch()),
                skip: 0,
                preserve_order: local_limit.required_ordering().is_some(),
            })
    }
}

/// Checks if the given plan combines input partitions.
fn combines_input_partitions(plan: &Arc<dyn ExecutionPlan>) -> bool {
    plan.is::<CoalescePartitionsExec>() || plan.is::<SortPreservingMergeExec>()
}

/// Adds a limit to the plan, chooses between global and local limits based on
/// skip value and the number of partitions.
fn add_limit(
    pushdown_plan: Arc<dyn ExecutionPlan>,
    skip: usize,
    fetch: usize,
) -> Arc<dyn ExecutionPlan> {
    if skip > 0 || pushdown_plan.output_partitioning().partition_count() == 1 {
        add_global_limit(pushdown_plan, skip, Some(fetch))
    } else {
        Arc::new(LocalLimitExec::new(pushdown_plan, fetch + skip)) as _
    }
}

/// Adds a global limit to the plan.
fn add_global_limit(
    pushdown_plan: Arc<dyn ExecutionPlan>,
    skip: usize,
    fetch: Option<usize>,
) -> Arc<dyn ExecutionPlan> {
    Arc::new(GlobalLimitExec::new(pushdown_plan, skip, fetch)) as _
}

// See tests in datafusion/core/tests/physical_optimizer