venkata91 commented on a change in pull request #30164:
URL: https://github.com/apache/spark/pull/30164#discussion_r520959277
##########
File path:
core/src/main/scala/org/apache/spark/storage/BlockManagerMasterEndpoint.scala
##########
@@ -657,6 +681,28 @@ class BlockManagerMasterEndpoint(
}
}
+ private def getShufflePushMergerLocations(
+ numMergersNeeded: Int,
+ hostsToFilter: Set[String]): Seq[BlockManagerId] = {
+ val activeBlockManagers = blockManagerIdByExecutor.groupBy(_._2.host)
+ .mapValues(_.head).values.map(_._2).toSet
+ val filteredActiveBlockManagers = activeBlockManagers
+ .filterNot(x => hostsToFilter.contains(x.host))
+ val filteredActiveMergers = filteredActiveBlockManagers.map(
+ x => BlockManagerId(x.executorId, x.host,
StorageUtils.externalShuffleServicePort(conf)))
+
+ // Enough mergers are available as part of active executors list
+ if (filteredActiveMergers.size >= numMergersNeeded) {
+ filteredActiveMergers.toSeq
+ } else {
+ // Delta mergers added from inactive mergers list to the active mergers
list
+ val filteredDeadMergers = shuffleMergerLocations.values
+ .filterNot(mergerHost => filteredActiveMergers.exists(x => x.host ==
mergerHost.host))
+ filteredActiveMergers.toSeq ++
+ filteredDeadMergers.toSeq.take(numMergersNeeded -
filteredActiveMergers.size)
Review comment:
Yes, though this is not intentional. It seems like a side-effect. Do you
think we should prefer `active executors` locations which are also part of
`shuffleMergerLocations` before other `active executors` locations?
----------------------------------------------------------------
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]
