DDDominik commented on code in PR #1731:
URL: https://github.com/apache/auron/pull/1731#discussion_r2609315028
##########
thirdparty/auron-celeborn-0.6/src/main/scala/org/apache/spark/sql/execution/auron/shuffle/celeborn/AuronCelebornShuffleWriter.scala:
##########
@@ -76,4 +76,19 @@ class AuronCelebornShuffleWriter[K, V](
Shims.get
.getMapStatus(blockManagerId,
celebornPartitionWriter.getPartitionLengthMap, mapId))
}
+
+ // Override stop to use partition length map directly instead of rssStop's
mapStatus
+ // because celeborn writer doesn't populate partition sizes correctly when
using native writer
Review Comment:
1. The code changes[ in this
patch](https://github.com/apache/auron/commit/fe59e570af32b944b8ac4d80f2f3804cd874d76c#diff-3ee59754498ad02636be38e9d1057df9c7d9ce65ba93cbf686bf2418a7d8d19a)
prioritize using theMapStatus provided by the wrapped shuffle writer. However,
when integrating Celeborn support, we directly used partition writer instead of
the shuffle writer. As a result, the MapStatus values are all zeros.
2. In BlazeCelebornShuffleWriter.scala, overriding the stop method from the
base class with the original implementation will correctly populate partition
the size.
--
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]
