tanelk commented on a change in pull request #30018:
URL: https://github.com/apache/spark/pull/30018#discussion_r509010483
##########
File path:
sql/catalyst/src/main/scala/org/apache/spark/sql/catalyst/optimizer/Optimizer.scala
##########
@@ -475,6 +476,58 @@ object RemoveRedundantAliases extends Rule[LogicalPlan] {
def apply(plan: LogicalPlan): LogicalPlan = removeRedundantAliases(plan,
AttributeSet.empty)
}
+/**
+ * Remove redundant aggregates from a query plan. A redundant aggregate is an
aggregate whose
+ * only goal is to keep distinct values, while its parent aggregate would
ignore duplicate values.
+ */
+object RemoveRedundantAggregates extends Rule[LogicalPlan] with AliasHelper {
+ def apply(plan: LogicalPlan): LogicalPlan = plan transformUp {
+ case upper @ Aggregate(_, _, lower: Aggregate) if lowerIsRedundant(upper,
lower) =>
+ val aliasMap = getAliasMap(lower)
+ upper.copy(
+ child = lower.child,
+ groupingExpressions = upper.groupingExpressions.map(replaceAlias(_,
aliasMap)),
+ aggregateExpressions = upper.aggregateExpressions.map(
+ replaceAliasButKeepOuter(_, aliasMap))
+ )
+ }
+
+ private def lowerIsRedundant(upper: Aggregate, lower: Aggregate): Boolean = {
+ val isDeterministic = upper.aggregateExpressions.forall(_.deterministic) &&
+ lower.aggregateExpressions.forall(_.deterministic)
Review comment:
I was thinking about some stateful UDF. As a dummy example - one that
counts the number of times it has been called and returns that count. Even if
it is not referenced, evaluating it changes future results.
Or perhaps I'm overthinking it and we can ignore it?
Taking it a step further - perhaps only upper has to be deterministic?
----------------------------------------------------------------
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.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
