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_r347023415
##########
File path:
common/network-shuffle/src/main/java/org/apache/spark/network/shuffle/ExternalShuffleBlockResolver.java
##########
@@ -369,6 +371,12 @@ public int removeBlocks(String appId, String execId,
String[] blockIds) {
return numRemovedBlocks;
}
+ public Map<String, String[]> getLocalDirs(String appId, String[] execIds) {
+ return Arrays.stream(execIds)
+ .map(exec -> Pair.of(exec, executors.get(new AppExecId(appId,
exec)).localDirs))
Review comment:
Here as much data as possible.
But if we have a case where we cannot give back a directories for an
executor then **a fallback at the block content loading have to be provided**
too. So your https://github.com/apache/spark/pull/25299#discussion_r346979781
idea is much more appealing (although at first it seams to me more complex than
the current solution).
----------------------------------------------------------------
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]
