Ngone51 commented on a change in pull request #34156:
URL: https://github.com/apache/spark/pull/34156#discussion_r719968691



##########
File path: core/src/main/scala/org/apache/spark/MapOutputTracker.scala
##########
@@ -1448,7 +1448,7 @@ private[spark] object MapOutputTracker extends Logging {
     // TODO: improve push based shuffle to read partial merged blocks 
satisfying the start/end
     // TODO: map indexes
     if (mergeStatuses.exists(_.nonEmpty) && startMapIndex == 0
-      && endMapIndex == mapStatuses.length) {
+      && endMapIndex == mapStatuses.length && endPartition - startPartition == 
1) {

Review comment:
       > No behavior change : other than disabling block fetch.
   
   @mridulm But this is the major concern from me now by disabling batch fetch 
for all the cases when PBS is enabled. Because the case like 
`PartialMapperPartitionSpec` can no longer use the batch fetch when PBS is 
enabled (which could have perf issue), while it could use in other ways:
    1) `endPartition - startPartition == 1`
    2) only disable batch fetch for the case of partition coalesce rather than 
all the cases
     And sound like @Victsm wants to take the way 2) as he said:
   
   > We are evaluating passing this additional information to 
BlockStoreShuffleReader as well so that we can properly disable batch fetch 
only for cases of partition coalesce but not for local shuffle read.
   
   For 2), just in case you may touch`ShuffleManager`: there're users who use 
`ShuffleManager` as API to implement their own manager, though it's private 
API. But I still don't want to see changes in this interface to cause 
compatibility issues in 3.2 at this monent.




-- 
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]

Reply via email to