asolimando commented on code in PR #21815: URL: https://github.com/apache/datafusion/pull/21815#discussion_r3291430588
########## datafusion/physical-plan/src/statistics_context.rs: ########## @@ -0,0 +1,213 @@ +// 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. + +//! Context for computing statistics in physical plans. +//! +//! [`StatisticsArgs`] provides external context to +//! [`ExecutionPlan::statistics_with_args`], enabling operators to look up +//! pre-computed child statistics via a shared cache. + +use crate::ExecutionPlan; +use datafusion_common::Result; +use datafusion_common::Statistics; +use std::cell::RefCell; +use std::collections::HashMap; +use std::rc::Rc; +use std::sync::Arc; + +/// Per-call memoization cache for [`compute_statistics`]. +/// +/// Keyed by `(plan node pointer address, partition)`. Created once per +/// top-level [`compute_statistics`] call and shared across all recursive +/// and operator-internal calls via [`StatisticsArgs`]. +/// +/// The pointer-based key is safe within a single synchronous +/// `compute_statistics` call: all `Arc<dyn ExecutionPlan>` nodes are held +/// by the plan tree for the duration of the walk, so addresses cannot be +/// reused. +#[derive(Debug, Default)] +struct StatsCache(HashMap<(usize, Option<usize>), Arc<Statistics>>); + +impl StatsCache { + fn get( + &self, + plan: &dyn ExecutionPlan, + partition: Option<usize>, + ) -> Option<&Arc<Statistics>> { + let key = ( + plan as *const dyn ExecutionPlan as *const () as usize, + partition, + ); + self.0.get(&key) + } + + fn insert( + &mut self, + plan: &dyn ExecutionPlan, + partition: Option<usize>, + stats: Arc<Statistics>, + ) { + let key = ( + plan as *const dyn ExecutionPlan as *const () as usize, + partition, + ); + self.0.insert(key, stats); + } +} + +/// Arguments passed to [`ExecutionPlan::statistics_with_args`] carrying +/// external information that operators can use when computing their +/// statistics. +#[derive(Debug)] +pub struct StatisticsArgs { + partition: Option<usize>, + /// Shared memoization cache for the current `compute_statistics` walk. + /// Children's overall (`None`) stats are pre-populated by + /// `compute_statistics_inner` before calling the operator. + cache: Option<Rc<RefCell<StatsCache>>>, +} + +impl StatisticsArgs { + /// Creates new statistics arguments without a shared cache. + /// Used by callers outside of `compute_statistics` (e.g. benchmarks). + pub fn new(partition: Option<usize>) -> Self { + Self { + partition, + cache: None, + } + } + + pub fn partition(&self) -> Option<usize> { + self.partition + } + + /// Returns the pre-computed overall (`None`) statistics for a child. + /// For per-partition stats, use [`Self::child_stats_for`]. + pub fn child_stats_of(&self, child: &dyn ExecutionPlan) -> Result<Arc<Statistics>> { + self.compute_child_statistics(child, None) + } + + /// Returns statistics for a child, using pre-computed overall stats + /// when `partition` is `None`, or computing per-partition stats on + /// demand when `partition` is `Some`. + pub fn child_stats_for(&self, child: &dyn ExecutionPlan) -> Result<Arc<Statistics>> { + self.compute_child_statistics(child, self.partition) + } + + /// Computes statistics for a child plan, using the shared cache + /// from the current [`compute_statistics`] walk. + pub fn compute_child_statistics( + &self, + plan: &dyn ExecutionPlan, + partition: Option<usize>, + ) -> Result<Arc<Statistics>> { + match &self.cache { + Some(cache) => compute_statistics_inner(plan, partition, cache), + None => compute_statistics(plan, partition), + } + } +} + +/// Computes statistics for a plan node by first recursively computing +/// overall (`None`) statistics for all children, then calling +/// [`ExecutionPlan::statistics_with_args`]. +/// +/// Results are memoized within a single call: operators that call +/// [`StatisticsArgs::child_stats_of`] or [`StatisticsArgs::child_stats_for`] +/// will hit the cache instead of re-walking subtrees. +/// +/// Children are always pre-computed with `partition = None` (overall stats). +/// Operators that need per-partition child stats (e.g., partition-preserving +/// operators when called with `Some(partition)`) should use +/// [`StatisticsArgs::child_stats_for`]. +pub fn compute_statistics( + plan: &dyn ExecutionPlan, + partition: Option<usize>, +) -> Result<Arc<Statistics>> { + let cache = Rc::new(RefCell::new(StatsCache::default())); + compute_statistics_inner(plan, partition, &cache) +} + +fn compute_statistics_inner( + plan: &dyn ExecutionPlan, + partition: Option<usize>, + cache: &Rc<RefCell<StatsCache>>, +) -> Result<Arc<Statistics>> { + if let Some(cached) = cache.borrow().get(plan, partition) { + return Ok(Arc::clone(cached)); + } + + for child in plan.children() { + compute_statistics_inner(child.as_ref(), None, cache)?; Review Comment: I originally misread @xudong963 [comment](https://github.com/apache/datafusion/pull/21815#discussion_r3165497867) as suggesting we pre-compute with `None` (to avoid the discard+recompute), but you were pointing at the broader issue of not eagerly computing stats that operators don't need. I have fixed this by removing the eager pre-computation loop for `None`, child stats are now computed lazily on demand. Only the stats actually needed by each operator are computed. Re-running the filter chain benchmark confirms cached_partition ~= cached_overall (no more double-walk), so this is clearly better, apologies for misreading the original feedback. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
