2010YOUY01 commented on code in PR #17482: URL: https://github.com/apache/datafusion/pull/17482#discussion_r2345889471
########## datafusion/physical-plan/src/joins/piecewise_merge_join/exec.rs: ########## @@ -0,0 +1,729 @@ +// 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. + +use arrow::array::Array; +use arrow::{ + array::{ArrayRef, BooleanBufferBuilder, RecordBatch}, + compute::concat_batches, + util::bit_util, +}; +use arrow_schema::{SchemaRef, SortOptions}; +use datafusion_common::not_impl_err; +use datafusion_common::{internal_err, JoinSide, Result}; +use datafusion_execution::{ + memory_pool::{MemoryConsumer, MemoryReservation}, + SendableRecordBatchStream, +}; +use datafusion_expr::{JoinType, Operator}; +use datafusion_physical_expr::equivalence::join_equivalence_properties; +use datafusion_physical_expr::{ + LexOrdering, OrderingRequirements, PhysicalExpr, PhysicalExprRef, PhysicalSortExpr, +}; +use datafusion_physical_expr_common::physical_expr::fmt_sql; +use futures::TryStreamExt; +use parking_lot::Mutex; +use std::fmt::Formatter; +use std::sync::Arc; + +use crate::execution_plan::{boundedness_from_children, EmissionType}; + +use crate::joins::piecewise_merge_join::classic_join::{ + ClassicPWMJStream, PiecewiseMergeJoinStreamState, +}; +use crate::joins::piecewise_merge_join::utils::{ + build_visited_indices_map, is_existence_join, is_right_existence_join, +}; +use crate::joins::utils::symmetric_join_output_partitioning; +use crate::{ + joins::{ + utils::{build_join_schema, BuildProbeJoinMetrics, OnceAsync, OnceFut}, + SharedBitmapBuilder, + }, + metrics::ExecutionPlanMetricsSet, + spill::get_record_batch_memory_size, + ExecutionPlan, PlanProperties, +}; +use crate::{DisplayAs, DisplayFormatType, ExecutionPlanProperties}; + +/// `PiecewiseMergeJoinExec` is a join execution plan that only evaluates single range filter. +/// +/// The physical planner will choose to evaluate this join when there is only one range predicate. This +/// is a binary expression which contains [`Operator::Lt`], [`Operator::LtEq`], [`Operator::Gt`], and +/// [`Operator::GtEq`].: +/// Examples: +/// - `col0` < `colb`, `col0` <= `colb`, `col0` > `colb`, `col0` >= `colb` +/// +/// Since the join only support range predicates, equijoins are not supported in `PiecewiseMergeJoinExec`, +/// however you can first evaluate another join and run `PiecewiseMergeJoinExec` if left with one range +/// predicate. +/// +/// # Execution Plan Inputs +/// For `PiecewiseMergeJoin` we label all right inputs as the `streamed' side and the left outputs as the +/// 'buffered' side. +/// +/// `PiecewiseMergeJoin` takes a sorted input for the side to be buffered and is able to sort streamed record +/// batches during processing. Sorted input must specifically be ascending/descending based on the operator. +/// +/// # Algorithms +/// Classic joins are processed differently compared to existence joins. +/// +/// ## Classic Joins (Inner, Full, Left, Right) +/// For classic joins we buffer the right side (buffered), and incrementally process the left side (streamed). +/// Every streamed batch is sorted so we can perform a sort merge algorithm. For the buffered side we want to Review Comment: I recommend to break this sort order rationale to two paragraphs: 1. How is input order determined (like left asc, right desc). I think a better way to phrase the words is: it finds a order so that the operator iterates both left and right side from index 0 to end, when it finds the first match, it outputs the buffer side's current index to end index joining the current stream side index. 2. How is input order enforced? - For stream side it should be read a maybe unordered batch, then sort it inside this operator - I don't know how is it done for buffer side, it's worth more explanation here. I vaguely remember it's declaring it in the `requires_input_order` in the plan property, and some optimizer rule will insert a `SortExec` automatically? -- 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]
