Skip to content

[SPARK-34014][SQL] Ignore Distinct if it is the right child of a left semi or anti join #31045

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
tanelk wants to merge 3 commits into from
Closed

[SPARK-34014][SQL] Ignore Distinct if it is the right child of a left semi or anti join #31045

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
1708e56
Ignore Distinct if it is the right child of a left semi or anti join
tanelk Jan 5, 2021
8e323e5
Merge remote-tracking branch 'upstream/master' into SPARK-34014_ignor…
tanelk Jan 5, 2021
3bb7cd5
Address comments
tanelk Jan 6, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
7 changes: 5 additions & 2 deletions sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -176,7 +176,7 @@ abstract class Optimizer(catalogManager: CatalogManager)
// to enforce idempotence on it and we change this batch from Once to FixedPoint(1).
Batch("Subquery", FixedPoint(1),
OptimizeSubqueries) ::
Batch("Replace Operators", fixedPoint,
Batch("Replace Operators", Once,
RewriteExceptAll,
RewriteIntersectAll,
ReplaceIntersectWithSemiJoin,
Expand Down Expand Up @@ -1612,9 +1612,12 @@ object ConvertToLocalRelation extends Rule[LogicalPlan] {
* {{{
* SELECT DISTINCT f1, f2 FROM t ==> SELECT f1, f2 FROM t GROUP BY f1, f2
* }}}
*
* The [[Distinct]] can be ignored if it is the right child of a left semi or anti join.

@maropu maropu Jan 5, 2021

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: could you move this comment between L1619/L1620?

// The [[Distinct]] can be ignored if it is the right child of a left semi or anti join.
case j @ Join(_, Distinct(right), LeftSemiOrAnti(_), _, _) => j.copy(right = right)

*/
object ReplaceDistinctWithAggregate extends Rule[LogicalPlan] {
def apply(plan: LogicalPlan): LogicalPlan = plan transform {
def apply(plan: LogicalPlan): LogicalPlan = plan transformDown {
case j @ Join(_, Distinct(right), LeftSemiOrAnti(_), _, _) => j.copy(right = right)

@maropu maropu Jan 5, 2021

Copy link
Copy Markdown
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Just out of curiosity, this rule can always make performance better? For example, how about the case where a right table has too many duplicates?

@tanelk tanelk Jan 6, 2021

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I benchmarked a bit and you are right - this can cause a performance regression. For example q87 took a 10% hit on sf100. I think it's best to close this PR and issue.

case Distinct(child) => Aggregate(child.output, child.output, child)
}
}
Expand Down
24 changes: 24 additions & 0 deletions ...atalyst/src/test/scala/org/apache/spark/sql/catalyst/optimizer/ReplaceOperatorSuite.scala
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -191,6 +191,30 @@ class ReplaceOperatorSuite extends PlanTest {
comparePlans(optimized, correctAnswer)
}

test("SPARK-34014: Ignore Distinct if it is the right child of a left semi join") {
val x = LocalRelation('x1.int, 'x2.int)
val y = LocalRelation('y1.int, 'y2.int)

val query = x.join(Distinct(y), LeftSemi, Some('x1 === 'y1))
val optimized = Optimize.execute(query.analyze)

val correctAnswer = x.join(y, LeftSemi, Some('x1 === 'y1)).analyze

comparePlans(optimized, correctAnswer)
}

test("SPARK-34014: Ignore Distinct if it is the right child of a left anti join") {
val x = LocalRelation('x1.int, 'x2.int)
val y = LocalRelation('y1.int, 'y2.int)

val query = x.join(Distinct(y), LeftAnti, Some('x1 === 'y1))
val optimized = Optimize.execute(query.analyze)

val correctAnswer = x.join(y, LeftAnti, Some('x1 === 'y1)).analyze

comparePlans(optimized, correctAnswer)
}

test("replace batch Deduplicate with Aggregate") {
val input = LocalRelation('a.int, 'b.int)
val attrA = input.output(0)
Expand Down
254 changes: 106 additions & 148 deletions sql/core/src/test/resources/tpcds-plan-stability/approved-plans-v1_4/q38.sf100/explain.txt
View file
Open in desktop

Large diffs are not rendered by default.

106 changes: 48 additions & 58 deletions ...core/src/test/resources/tpcds-plan-stability/approved-plans-v1_4/q38.sf100/simplified.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,8 +1,8 @@
WholeStageCodegen (26)
WholeStageCodegen (24)
HashAggregate [count] [count(1),count(1),count]
InputAdapter
Exchange #1
WholeStageCodegen (25)
WholeStageCodegen (23)
HashAggregate [count,count]
HashAggregate [c_last_name,c_first_name,d_date]
HashAggregate [c_last_name,c_first_name,d_date]
Expand All @@ -11,7 +11,7 @@ WholeStageCodegen (26)
HashAggregate [c_last_name,c_first_name,d_date]
InputAdapter
Exchange [c_last_name,c_first_name,d_date] #2
WholeStageCodegen (24)
WholeStageCodegen (22)
HashAggregate [c_last_name,c_first_name,d_date]
InputAdapter
SortMergeJoin [c_last_name,c_first_name,d_date,c_last_name,c_first_name,d_date]
Expand Down Expand Up @@ -53,65 +53,55 @@ WholeStageCodegen (26)
ColumnarToRow
InputAdapter
Scan parquet default.customer [c_customer_sk,c_first_name,c_last_name]
WholeStageCodegen (15)
WholeStageCodegen (14)
Sort [c_last_name,c_first_name,d_date]
InputAdapter
Exchange [c_last_name,c_first_name,d_date] #7
WholeStageCodegen (14)
HashAggregate [c_last_name,c_first_name,d_date]
InputAdapter
Exchange [c_last_name,c_first_name,d_date] #8
WholeStageCodegen (13)
HashAggregate [c_last_name,c_first_name,d_date]
Project [c_last_name,c_first_name,d_date]
SortMergeJoin [cs_bill_customer_sk,c_customer_sk]
InputAdapter
WholeStageCodegen (10)
Sort [cs_bill_customer_sk]
InputAdapter
Exchange [cs_bill_customer_sk] #9
WholeStageCodegen (9)
Project [cs_bill_customer_sk,d_date]
BroadcastHashJoin [cs_sold_date_sk,d_date_sk]
Filter [cs_sold_date_sk,cs_bill_customer_sk]
ColumnarToRow
InputAdapter
Scan parquet default.catalog_sales [cs_sold_date_sk,cs_bill_customer_sk]
InputAdapter
ReusedExchange [d_date_sk,d_date] #5
InputAdapter
WholeStageCodegen (12)
Sort [c_customer_sk]
WholeStageCodegen (13)
Project [c_last_name,c_first_name,d_date]
SortMergeJoin [cs_bill_customer_sk,c_customer_sk]
InputAdapter
WholeStageCodegen (10)
Sort [cs_bill_customer_sk]
InputAdapter
Exchange [cs_bill_customer_sk] #8
WholeStageCodegen (9)
Project [cs_bill_customer_sk,d_date]
BroadcastHashJoin [cs_sold_date_sk,d_date_sk]
Filter [cs_sold_date_sk,cs_bill_customer_sk]
ColumnarToRow
InputAdapter
Scan parquet default.catalog_sales [cs_sold_date_sk,cs_bill_customer_sk]
InputAdapter
ReusedExchange [c_customer_sk,c_first_name,c_last_name] #6
WholeStageCodegen (23)
ReusedExchange [d_date_sk,d_date] #5
InputAdapter
WholeStageCodegen (12)
Sort [c_customer_sk]
InputAdapter
ReusedExchange [c_customer_sk,c_first_name,c_last_name] #6
WholeStageCodegen (21)
Sort [c_last_name,c_first_name,d_date]
InputAdapter
Exchange [c_last_name,c_first_name,d_date] #10
WholeStageCodegen (22)
HashAggregate [c_last_name,c_first_name,d_date]
InputAdapter
Exchange [c_last_name,c_first_name,d_date] #11
WholeStageCodegen (21)
HashAggregate [c_last_name,c_first_name,d_date]
Project [c_last_name,c_first_name,d_date]
SortMergeJoin [ws_bill_customer_sk,c_customer_sk]
InputAdapter
WholeStageCodegen (18)
Sort [ws_bill_customer_sk]
InputAdapter
Exchange [ws_bill_customer_sk] #12
WholeStageCodegen (17)
Project [ws_bill_customer_sk,d_date]
BroadcastHashJoin [ws_sold_date_sk,d_date_sk]
Filter [ws_sold_date_sk,ws_bill_customer_sk]
ColumnarToRow
InputAdapter
Scan parquet default.web_sales [ws_sold_date_sk,ws_bill_customer_sk]
InputAdapter
ReusedExchange [d_date_sk,d_date] #5
InputAdapter
WholeStageCodegen (20)
Sort [c_customer_sk]
Exchange [c_last_name,c_first_name,d_date] #9
WholeStageCodegen (20)
Project [c_last_name,c_first_name,d_date]
SortMergeJoin [ws_bill_customer_sk,c_customer_sk]
InputAdapter
WholeStageCodegen (17)
Sort [ws_bill_customer_sk]
InputAdapter
Exchange [ws_bill_customer_sk] #10
WholeStageCodegen (16)
Project [ws_bill_customer_sk,d_date]
BroadcastHashJoin [ws_sold_date_sk,d_date_sk]
Filter [ws_sold_date_sk,ws_bill_customer_sk]
ColumnarToRow
InputAdapter
Scan parquet default.web_sales [ws_sold_date_sk,ws_bill_customer_sk]
InputAdapter
ReusedExchange [c_customer_sk,c_first_name,c_last_name] #6
ReusedExchange [d_date_sk,d_date] #5
InputAdapter
WholeStageCodegen (19)
Sort [c_customer_sk]
InputAdapter
ReusedExchange [c_customer_sk,c_first_name,c_last_name] #6
Loading