attilapiros commented on a change in pull request #25299: [SPARK-27651][Core]
Avoid the network when shuffle blocks are fetched from the same host
URL: https://github.com/apache/spark/pull/25299#discussion_r317614716
##########
File path: core/src/main/scala/org/apache/spark/storage/BlockManager.scala
##########
@@ -995,6 +1003,19 @@ private[spark] class BlockManager(
None
}
+ private[spark] def getHostLocalDirs(executorIds: Array[String])
+ : scala.collection.Map[String, Array[String]] = {
+ val cachedItems =
executorIdToLocalDirsCache.filterKeys(executorIds.contains(_))
+ if (cachedItems.size < executorIds.length) {
+ val notCachedItems = master
+
.getHostLocalDirs(executorIds.filterNot(executorIdToLocalDirsCache.contains))
+ executorIdToLocalDirsCache ++= notCachedItems
Review comment:
Yes, it is unbounded but this case is the very same what we have at the
external shuffle service: here only those executors are stored which is/was
running on the same hosts. I assume this number is quite small but we can use
the same limit here if you think it still worth to do that.
----------------------------------------------------------------
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]
With regards,
Apache Git Services
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
