zuston commented on code in PR #2653:
URL: https://github.com/apache/uniffle/pull/2653#discussion_r2522195598
##########
client-spark/spark3/src/main/java/org/apache/spark/shuffle/writer/RssShuffleWriter.java:
##########
@@ -933,14 +949,12 @@ public Option<MapStatus> stop(boolean success) {
bitmapSplitNum,
reportDuration);
shuffleWriteMetrics.incWriteTime(TimeUnit.MILLISECONDS.toNanos(reportDuration));
+
// todo: we can replace the dummy host and port with the real shuffle
server which we prefer
// to read
final BlockManagerId blockManagerId =
BlockManagerId.apply(
- appId + "_" + taskId,
- DUMMY_HOST,
- DUMMY_PORT,
- Option.apply(Long.toString(taskAttemptId)));
+ appId + "_" + taskId, DUMMY_HOST, DUMMY_PORT,
Option.apply(createShuffleInfo()));
Review Comment:
> Would it be better to send this data to ShuffleServer? This can easily
cause an OOM of the Driver.
I have to say you are right. for the huge stage like 100000 tasks with 10000
partitions, the driver will be oom. And so, I have to make this design possible
by the shuffle-server store
--
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]
