milenkovicm commented on code in PR #1752:
URL:
https://github.com/apache/datafusion-ballista/pull/1752#discussion_r3313839243
##########
ballista/scheduler/src/state/aqe/planner.rs:
##########
@@ -99,7 +107,8 @@ impl AdaptivePlanner {
let plan = planner.optimize_physical_plan(plan, &session_state, |_, _|
{})?;
Ok(Self {
- stage_id_generator: 0,
+ stage_id_generator: 0, // FIXME: compatibility issue with static
where stages start from 1
Review Comment:
For follow up once this get merged
##########
ballista/scheduler/src/state/aqe/execution_plan/dynamic_join.rs:
##########
@@ -0,0 +1,537 @@
+// 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.
+
+//! Functional tests for [`CoalescePartitionsRule`]: drive a query through
+//! `AdaptivePlanner`, finalize the upstream stage with synthetic per-partition
+//! byte stats, and snapshot the displayed plan tree so the rule's effect on
+//! the leaf `ExchangeExec` is visible at the `coalesce=K of M` field.
+//!
+//! Each test uses small synthetic byte sizes paired with a small
+//! `coalesce_target_partition_bytes` so the bin-pack outcome is hand-traceable
+//! against `split_size_list_by_target_size`.
+
+use datafusion::{
+ arrow::compute::SortOptions,
+ common::{JoinType, NullEquality, Result, exec_err, internal_err},
+ physical_expr_common::physical_expr::fmt_sql,
+ physical_plan::{
+ DisplayAs, DisplayFormatType, Distribution, ExecutionPlan,
PlanProperties,
+ joins::{
+ HashJoinExec, HashJoinExecBuilder, JoinOn, PartitionMode,
SortMergeJoinExec,
+ utils::JoinFilter,
+ },
+ },
+};
+use std::sync::Arc;
+
+use crate::state::aqe::execution_plan::ExchangeExec;
+
+/// has children of this join been
+/// repartitioned
+#[derive(Debug, Clone, PartialEq, Eq)]
+pub enum ChildrenState {
+ Repartitioned,
+ Unknown,
+}
+
+// SortMergeJoinExec::try_new(left, right, on, filter, join_type,
sort_options, null_equality )
+// HashJoinExec::try_new (left, right, on, filter, join_type, projection,
partition_mode, null_equality, null_aware )
+#[derive(Debug)]
+pub struct DynamicJoinSelectionExec {
+ pub left: Arc<dyn ExecutionPlan>,
+ pub right: Arc<dyn ExecutionPlan>,
+ pub on: JoinOn,
+ pub filter: Option<JoinFilter>,
+ pub join_type: JoinType,
+ pub projection: Option<Vec<usize>>,
+ pub null_equality: NullEquality,
+ pub properties: Arc<PlanProperties>,
+ pub selection_state: ChildrenState,
+}
+
+impl ExecutionPlan for DynamicJoinSelectionExec {
+ fn name(&self) -> &str {
+ "DynamicJoinSelectionExec"
+ }
+
+ fn as_any(&self) -> &dyn std::any::Any {
+ self
+ }
+
+ fn properties(&self) ->
&std::sync::Arc<datafusion::physical_plan::PlanProperties> {
+ &self.properties
+ }
+
+ fn children(&self) -> Vec<&std::sync::Arc<dyn ExecutionPlan>> {
+ vec![&self.left, &self.right]
+ }
+
+ fn with_new_children(
+ self: std::sync::Arc<Self>,
+ children: Vec<std::sync::Arc<dyn ExecutionPlan>>,
+ ) -> datafusion::error::Result<std::sync::Arc<dyn ExecutionPlan>> {
+ if children.len() != 2 {
+ return internal_err!(
+ "DynamicJoinSelectionExec expects 2 children, got {}",
+ children.len()
+ );
+ }
+ Ok(Arc::new(DynamicJoinSelectionExec {
+ left: Arc::clone(&children[0]),
+ right: Arc::clone(&children[1]),
+ on: self.on.clone(),
+ filter: self.filter.clone(),
+ join_type: self.join_type,
+ projection: self.projection.clone(),
+ null_equality: self.null_equality,
+ properties: Arc::clone(&self.properties),
+ selection_state: self.selection_state.clone(),
+ }))
+ }
+
+ fn execute(
+ &self,
+ _partition: usize,
+ _context: std::sync::Arc<datafusion::execution::TaskContext>,
+ ) ->
datafusion::error::Result<datafusion::execution::SendableRecordBatchStream> {
+ exec_err!(
+ "Operator should not be executed; it should have been replaced by
an optimizer rule."
+ )
+ }
+}
+
+impl DisplayAs for DynamicJoinSelectionExec {
+ fn fmt_as(
+ &self,
+ t: datafusion::physical_plan::DisplayFormatType,
+ f: &mut std::fmt::Formatter,
+ ) -> std::fmt::Result {
+ match t {
+ DisplayFormatType::Default | DisplayFormatType::Verbose => {
+ let on = self
+ .on
+ .iter()
+ .map(|(c1, c2)| format!("({c1}, {c2})"))
+ .collect::<Vec<String>>()
+ .join(", ");
+ let display_null_equality =
+ if self.null_equality == NullEquality::NullEqualsNull {
+ ", NullsEqual: true"
+ } else {
+ ""
+ };
+ write!(
+ f,
+ "{}: join_type={:?}, on=[{}]{}{}",
+ Self::static_name(),
+ self.join_type,
+ on,
+ self.filter.as_ref().map_or_else(
+ || "".to_string(),
+ |f| format!(", filter={}", f.expression())
+ ),
+ display_null_equality,
+ )?;
+
+ write!(
+ f,
+ " repartitioned={}",
+ matches!(self.selection_state,
ChildrenState::Repartitioned)
+ )?;
+
+ Ok(())
+ }
+ DisplayFormatType::TreeRender => {
+ let on = self
+ .on
+ .iter()
+ .map(|(c1, c2)| {
+ format!("({} = {})", fmt_sql(c1.as_ref()),
fmt_sql(c2.as_ref()))
+ })
+ .collect::<Vec<String>>()
+ .join(", ");
+
+ if self.join_type != JoinType::Inner {
+ writeln!(f, "join_type={:?}", self.join_type)?;
+ }
+ writeln!(f, "on={on}")?;
+
+ if self.null_equality == NullEquality::NullEqualsNull {
+ writeln!(f, "NullsEqual: true")?;
+ }
+
+ writeln!(
+ f,
+ " repartitioned={}",
+ matches!(self.selection_state,
ChildrenState::Repartitioned)
+ )?;
+
+ Ok(())
+ }
+ }
+ }
+}
+
+pub enum JoinSelectionAction {
+ Repartition(Arc<DynamicJoinSelectionExec>),
+ CollectLeft(Arc<HashJoinExec>),
+ LateCollectLeft(Arc<HashJoinExec>),
Review Comment:
This action gets triggered when both sides get repartitioned, does it make
sense to change from partitioned to collect left join at that point ?
--
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]
