github-actions[bot] commented on code in PR #70480:
URL: https://github.com/apache/doris/pull/70480#discussion_r4655974149
##########
fe/fe-core/src/main/java/org/apache/doris/nereids/processor/post/RuntimeFilterPruner.java:
##########
@@ -95,6 +99,18 @@ public PhysicalSetOperation
visitPhysicalSetOperation(PhysicalSetOperation setOp
return setOperation;
}
+ @Override
+ public PhysicalUnion visitPhysicalUnion(PhysicalUnion union,
CascadesContext context) {
+ visitPhysicalSetOperation(union, context);
+ RuntimeFilterContext rfContext = context.getRuntimeFilterContext();
+ if (union.children().stream()
Review Comment:
[P2] Preserve changelog effectiveness through nested-loop outputs. The new
CHANGELOG marker is recorded on the stream child, but
visitPhysicalNestedLoopJoin only propagates join.child(0) and drops a marker
from the right child. A non-equi nested-loop with the stream on the right can
feed an outer hash-join build side, where the missing marker lets unknown or
non-selective stats prune the probe runtime filter. Propagate the right-side
source when incremental rows can reach the output and add a nested-loop build
regression case.
##########
fe/fe-core/src/main/java/org/apache/doris/nereids/processor/post/RuntimeFilterPruner.java:
##########
@@ -95,6 +99,18 @@ public PhysicalSetOperation
visitPhysicalSetOperation(PhysicalSetOperation setOp
return setOperation;
}
+ @Override
+ public PhysicalUnion visitPhysicalUnion(PhysicalUnion union,
CascadesContext context) {
+ visitPhysicalSetOperation(union, context);
+ RuntimeFilterContext rfContext = context.getRuntimeFilterContext();
+ if (union.children().stream()
Review Comment:
[P2] Propagate changelog effectiveness across materialized CTEs. The new
marker is recorded on the mixed stream UNION inside a PhysicalCTEProducer, but
visitPhysicalCTEAnchor visits the producer and each PhysicalCTEConsumer
independently, and the consumer is a leaf with no link to that marker. For a
materialized multi-use CTE whose stream consumer is the hash-join build side,
visitPhysicalHashJoin therefore sees no effective source and can prune the
probe runtime filter under unknown or non-selective stats. Carry the source
classification through the CTE producer/consumer mapping and add a
materialized-CTE regression case.
##########
fe/fe-core/src/main/java/org/apache/doris/nereids/processor/post/RuntimeFilterPruner.java:
##########
@@ -95,6 +99,18 @@ public PhysicalSetOperation
visitPhysicalSetOperation(PhysicalSetOperation setOp
return setOperation;
}
Review Comment:
[P2] Preserve changelog effectiveness through EXCEPT builds.
visitPhysicalUnion now promotes CHANGELOG, but PhysicalExcept still uses the
generic set-operation visitor, which only visits its children and leaves the
EXCEPT node unmarked. A stream branch in EXCEPT child 0 used as a hash-join
build input therefore loses the marker, and unknown or non-selective build
stats can make visitPhysicalHashJoin remove the probe runtime filter. Propagate
the strongest child source through PhysicalExcept and add a build-side
set-operation case.
##########
fe/fe-core/src/main/java/org/apache/doris/nereids/processor/post/RuntimeFilterPruner.java:
##########
@@ -95,6 +99,18 @@ public PhysicalSetOperation
visitPhysicalSetOperation(PhysicalSetOperation setOp
return setOperation;
}
+ @Override
+ public PhysicalUnion visitPhysicalUnion(PhysicalUnion union,
CascadesContext context) {
+ visitPhysicalSetOperation(union, context);
+ RuntimeFilterContext rfContext = context.getRuntimeFilterContext();
+ if (union.children().stream()
Review Comment:
[P2] Preserve changelog effectiveness through recursive unions.
visitPhysicalRecursiveUnion visits the two terms but never promotes a child
CHANGELOG source to the recursive-union output. A stream scan in a recursive
CTE anchor can therefore feed an unmarked build subtree, allowing unknown or
non-selective stats to remove the outer probe runtime filter. Propagate the
strongest child source through recursive unions and add a recursive-stream
build case.
--
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]