englefly commented on code in PR #67796:
URL: https://github.com/apache/doris/pull/67796#discussion_r4004034213
##########
fe/fe-core/src/main/java/org/apache/doris/nereids/rules/rewrite/PushDownTopNDistinctThroughJoin.java:
##########
@@ -197,6 +201,31 @@ private List<OrderKey> getPushedOrderKeys(Set<Slot>
groupBySlots, Set<Slot> join
notFound = true;
}
}
- return pushedOrderKeys.build();
+ List<OrderKey> pushedOrderKeyList = pushedOrderKeys.build();
+ if (pushedOrderKeyList.size() == orderKeys.size()
+ || isOrderKeyPrefixUniqueAfterDistinct(joinChild,
pushedOrderKeyList)) {
+ return pushedOrderKeyList;
+ }
+ return ImmutableList.of();
+ }
+
+ /**
+ * A partial order-key prefix is safe for a hard limit only when it
uniquely orders the rows produced by
+ * {@link PlanUtils#distinct(Plan)}. This is true when a leading part of
the prefix either is already a
+ * non-null unique key, covers every child output, or functionally
determines every remaining child output.
+ */
+ private boolean isOrderKeyPrefixUniqueAfterDistinct(Plan joinChild,
List<OrderKey> orderKeyPrefix) {
+ Set<Slot> childOutput = joinChild.getOutputSet();
+ DataTrait childTrait = joinChild.getLogicalProperties().getTrait();
+ Set<Slot> prefixSlots = new HashSet<>();
+ for (OrderKey orderKey : orderKeyPrefix) {
+ prefixSlots.add((Slot) orderKey.getExpr());
+ if (prefixSlots.containsAll(childOutput) ||
childTrait.isUniqueAndNotNull(prefixSlots)
+ || childOutput.stream().allMatch(slot ->
prefixSlots.contains(slot)
+ || childTrait.isDependent(prefixSlots,
ImmutableSet.of(slot)))) {
Review Comment:
这里可能漏掉优化
比如对project(a, b, a+1 as c)
有fd: a->c, 但没有 {a, b}->c
所以这里 childTrait.isDependent( {a, b}, {c} ) 返回false.
--
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]
