maropu commented on a change in pull request #29869:
URL: https://github.com/apache/spark/pull/29869#discussion_r495871556
##########
File path: core/src/main/scala/org/apache/spark/scheduler/DAGScheduler.scala
##########
@@ -274,8 +274,36 @@ private[spark] class DAGScheduler(
accumUpdates: Seq[AccumulatorV2[_, _]],
metricPeaks: Array[Long],
taskInfo: TaskInfo): Unit = {
+ // It may cause heavy full GC problem if some heavy external accumulators
keep in memory.
+ // We update these heavy accumulators before they entering into Spark
listener event loop.
+ val (heavyAccumUpdates, otherAccumUpdates) = accumUpdates.partition { acc
=>
Review comment:
"Heavy" means that accumulated values are too big? If so, we cannot just
check the size here and pre-merge it if the size goes over a threshold? IMHO it
is not a good idea for developers to control the accumulator behavior
explicitly via the name prefix.
----------------------------------------------------------------
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]
