[
https://issues.apache.org/jira/browse/SPARK-60135?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Hussein Awala updated SPARK-60135:
----------------------------------
Description:
DeduplicateRelations.deduplicate unconditionally calls
plan.transformExpressions with a partial function that matches only
SubqueryExpression. This also walks operator expressions that cannot contain a
subquery.
QueryPlan propagates expression tree-pattern bits, and SubqueryExpression
inherits the PLAN_EXPRESSION marker. A narrowly scoped optimization can skip
this expression scan when that marker is absent from the plan, and prune
expression subtrees without the marker when the scan is needed.
The relation-renewal traversal, attribute remapping, subquery recursion and
processing order must remain unchanged. The expression transform must retain
UnknownRuleId because its closure reads and updates traversal state;
ineffective-rule memoization would be inappropriate here.
Suggested validation uses generic Catalyst plans: projects containing
arithmetic/literal expressions, subqueries wrapped in aliases or other
expressions, nested/correlated subqueries, and the same subquery object reused
in multiple positions.
This is narrower than SPARK-49881 (conditional rule-wide skipping), and
distinct from SPARK-59452 (wide UNION branch comparisons and related synthetic
self-join overhead).
was:
DeduplicateRelations.deduplicate unconditionally calls
plan.transformExpressions with a partial function that matches only
SubqueryExpression. This also walks operator expressions that cannot contain a
subquery.
QueryPlan propagates expression tree-pattern bits, and SubqueryExpression
inherits the PLAN_EXPRESSION marker. A narrowly scoped optimization can skip
this expression scan when that marker is absent from the plan, and prune
expression subtrees without the marker when the scan is needed.
The relation-renewal traversal, attribute remapping, subquery recursion and
processing order must remain unchanged. The expression transform must retain
UnknownRuleId because its closure reads and updates traversal state;
ineffective-rule memoization would be inappropriate here.
Suggested validation uses generic Catalyst plans: projects containing
arithmetic/literal expressions, subqueries wrapped in aliases or other
expressions, nested/correlated subqueries, and the same subquery object reused
in multiple positions.
This is narrower than SPARK-49881 (conditional rule-wide skipping), and
distinct from SPARK-59452 (wide UNION branch comparisons and related synthetic
self-join overhead).
Generated-by: Pi Coding Agent 1.1.0
> Prune unnecessary subquery-expression traversal in DeduplicateRelations
> -----------------------------------------------------------------------
>
> Key: SPARK-60135
> URL: https://issues.apache.org/jira/browse/SPARK-60135
> Project: Spark
> Issue Type: Improvement
> Components: SQL
> Affects Versions: 5.0.0
> Reporter: Hussein Awala
> Priority: Major
>
> DeduplicateRelations.deduplicate unconditionally calls
> plan.transformExpressions with a partial function that matches only
> SubqueryExpression. This also walks operator expressions that cannot contain
> a subquery.
> QueryPlan propagates expression tree-pattern bits, and SubqueryExpression
> inherits the PLAN_EXPRESSION marker. A narrowly scoped optimization can skip
> this expression scan when that marker is absent from the plan, and prune
> expression subtrees without the marker when the scan is needed.
> The relation-renewal traversal, attribute remapping, subquery recursion and
> processing order must remain unchanged. The expression transform must retain
> UnknownRuleId because its closure reads and updates traversal state;
> ineffective-rule memoization would be inappropriate here.
> Suggested validation uses generic Catalyst plans: projects containing
> arithmetic/literal expressions, subqueries wrapped in aliases or other
> expressions, nested/correlated subqueries, and the same subquery object
> reused in multiple positions.
> This is narrower than SPARK-49881 (conditional rule-wide skipping), and
> distinct from SPARK-59452 (wide UNION branch comparisons and related
> synthetic self-join overhead).
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
