Ngone51 commented on a change in pull request #34304:
URL: https://github.com/apache/spark/pull/34304#discussion_r730939578
##########
File path:
core/src/main/scala/org/apache/spark/storage/ShuffleBlockFetcherIterator.scala
##########
@@ -623,18 +623,8 @@ final class ShuffleBlockFetcherIterator(
logDebug(s"Asynchronous fetching host-local blocks without cached
executors' dir: " +
s"${hostLocalBlocksWithMissingDirs.mkString(", ")}")
- // If the external shuffle service is enabled, we'll fetch the local
directories for
- // multiple executors from the external shuffle service, which located
at the same host
- // with the executors, in once. Otherwise, we'll fetch the local
directories from those
- // executors directly one by one. The fetch requests won't be too much
since one host is
- // almost impossible to have many executors at the same time practically.
- val dirFetchRequests = if (blockManager.externalShuffleServiceEnabled) {
- val host = blockManager.blockManagerId.host
- val port = blockManager.externalShuffleServicePort
- Seq((host, port, hostLocalBlocksWithMissingDirs.keys.toArray))
- } else {
- hostLocalBlocksWithMissingDirs.keys.map(bmId => (bmId.host, bmId.port,
Array(bmId))).toSeq
- }
+ val dirFetchRequests = hostLocalBlocksWithMissingDirs.keys.map(bmId =>
+ (bmId.host, bmId.port, Array(bmId))).toSeq
Review comment:
I think there's a behavior difference after the change - previously,
there would be only one call on `getHostLocalDirs` below. After this change, it
becomes the number of `hostLocalBlocksWithMissingDirs`, which could introduce
more RPC calls.
--
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]