maropu commented on a change in pull request #28804:
URL: https://github.com/apache/spark/pull/28804#discussion_r447512443
##########
File path:
sql/catalyst/src/main/scala/org/apache/spark/sql/internal/SQLConf.scala
##########
@@ -2196,6 +2196,13 @@ object SQLConf {
.checkValue(bit => bit >= 10 && bit <= 30, "The bit value must be in
[10, 30].")
.createWithDefault(16)
+ val SKIP_PARTIAL_AGGREGATE_ENABLED =
+ buildConf("spark.sql.aggregate.partialaggregate.skip.enabled")
+ .internal()
+ .doc("Avoid sort/spill to disk during partial aggregation")
+ .booleanConf
+ .createWithDefault(true)
Review comment:
> They turn off map side aggregate (i.e., partial aggregate will be
pass through) in Physical operator (i.e., Group-By operator) if map-side
aggregation reduce the entries by at least half and they look at 100000 rows to
do that
I think whether that approach improves performance depends on IO
performance, but the idea looks interesting to me. WDYT? @cloud-fan
----------------------------------------------------------------
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]
