You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@celeborn.apache.org by GitBox <gi...@apache.org> on 2022/11/30 12:54:30 UTC
[GitHub] [incubator-celeborn] RexXiong opened a new pull request, #1034: [CELEBORN-76][FOLLOWUP] fix inFlightCommitRequest counting problem
RexXiong opened a new pull request, #1034:
URL: https://github.com/apache/incubator-celeborn/pull/1034
# [BUG]/[FEATURE] title
### What changes were proposed in this pull request?
### Why are the changes needed?
<!--
Please clarify why the changes are needed. For instance,
1. If you propose a new API, clarify the use case for a new API.
2. If you fix a bug, you can clarify why it is a bug.
-->
### What are the items that need reviewer attention?
### Related issues.
### Related pull requests.
### How was this patch tested?
/cc @related-reviewer
/assign @AngersZhuuuu
--
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: dev-unsubscribe@celeborn.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [incubator-celeborn] waitinfuture commented on pull request #1034: [CELEBORN-76][FOLLOWUP] fix inFlightCommitRequest counting problem
Posted by GitBox <gi...@apache.org>.
waitinfuture commented on PR #1034:
URL: https://github.com/apache/incubator-celeborn/pull/1034#issuecomment-1333301794
> > I don't quite get the point this pr fixes, could you explain more? IMO the inFlightCommitRequest stores the number of commitFiles requests, each request can commit files to multiple workers.
>
> because it may only be partially success, as stageEnd process waits on shuffleCommittedInfo.inFlightCommitRequest, CommitThreads in batchHandleCommitPartitionExecutors may not even execute, but inFlightCommitRequest may be equal to 0.
Previously we increment by 1 before requestCommitFilesWithRetry and decrements by 1 after returns, I think it's consistent.
--
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: dev-unsubscribe@celeborn.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [incubator-celeborn] RexXiong commented on pull request #1034: [CELEBORN-76][FOLLOWUP] fix inFlightCommitRequest counting problem
Posted by GitBox <gi...@apache.org>.
RexXiong commented on PR #1034:
URL: https://github.com/apache/incubator-celeborn/pull/1034#issuecomment-1333038122
> I don't quite get the point this pr fixes, could you explain more? IMO the inFlightCommitRequest stores the number of commitFiles requests, each request can commit files to multiple workers.
because it may only be partially success, as stageEnd process waits on shuffleCommittedInfo.inFlightCommitRequest, CommitThreads in batchHandleCommitPartitionExecutors may not even execute, but inFlightCommitRequest may be equal to 0.
--
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: dev-unsubscribe@celeborn.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [incubator-celeborn] RexXiong commented on pull request #1034: [CELEBORN-76][FOLLOWUP] fix inFlightCommitRequest counting problem
Posted by GitBox <gi...@apache.org>.
RexXiong commented on PR #1034:
URL: https://github.com/apache/incubator-celeborn/pull/1034#issuecomment-1333305946
> > > I don't quite get the point this pr fixes, could you explain more? IMO the inFlightCommitRequest stores the number of commitFiles requests, each request can commit files to multiple workers.
> >
> >
> > because it may only be partially success, as stageEnd process waits on shuffleCommittedInfo.inFlightCommitRequest, CommitThreads in batchHandleCommitPartitionExecutors may not even execute, but inFlightCommitRequest may be equal to 0.
>
> Previously we increment by 1 before requestCommitFilesWithRetry and decrements by 1 after returns, I think it's consistent.
the inflight count will be consistent finally. but stage end will lost some commit files in those uncommitted threads
--
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: dev-unsubscribe@celeborn.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [incubator-celeborn] waitinfuture commented on pull request #1034: [CELEBORN-76][FOLLOWUP] fix inFlightCommitRequest counting problem
Posted by GitBox <gi...@apache.org>.
waitinfuture commented on PR #1034:
URL: https://github.com/apache/incubator-celeborn/pull/1034#issuecomment-1332299588
I don't quite get the point this pr fixes, could you explain more? IMO the inFlightCommitRequest stores the number of commitFiles requests, each request can commit files to multiple workers.
--
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: dev-unsubscribe@celeborn.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [incubator-celeborn] AngersZhuuuu merged pull request #1034: [CELEBORN-76][FOLLOWUP] fix inFlightCommitRequest counting problem
Posted by GitBox <gi...@apache.org>.
AngersZhuuuu merged PR #1034:
URL: https://github.com/apache/incubator-celeborn/pull/1034
--
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: dev-unsubscribe@celeborn.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [incubator-celeborn] AngersZhuuuu commented on pull request #1034: [CELEBORN-76][FOLLOWUP] fix inFlightCommitRequest counting problem
Posted by GitBox <gi...@apache.org>.
AngersZhuuuu commented on PR #1034:
URL: https://github.com/apache/incubator-celeborn/pull/1034#issuecomment-1333318744
> > > I don't quite get the point this pr fixes, could you explain more? IMO the inFlightCommitRequest stores the number of commitFiles requests, each request can commit files to multiple workers.
> >
> >
> > because it may only be partially success, as stageEnd process waits on shuffleCommittedInfo.inFlightCommitRequest, CommitThreads in batchHandleCommitPartitionExecutors may not even execute, but inFlightCommitRequest may be equal to 0.
>
> Previously we increment by 1 before requestCommitFilesWithRetry and decrements by 1 after returns, I think it's consistent.
We are calculating the completion of this batch. In the previous code, we can only guarantee that the request of a certain worker in this batch is completed, but it is possible that the request of other workers has not started yet, so it enters the calculation datalost
--
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: dev-unsubscribe@celeborn.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org
[GitHub] [incubator-celeborn] waitinfuture commented on a diff in pull request #1034: [CELEBORN-76][FOLLOWUP] fix inFlightCommitRequest counting problem
Posted by GitBox <gi...@apache.org>.
waitinfuture commented on code in PR #1034:
URL: https://github.com/apache/incubator-celeborn/pull/1034#discussion_r1036062236
##########
client/src/main/scala/org/apache/celeborn/client/LifecycleManager.scala:
##########
@@ -293,57 +294,67 @@ class LifecycleManager(appId: String, val conf: CelebornConf) extends RpcEndpoin
.add(commitPartitionRequest.partition.getPeer)
}
}
- currentBatch
+
+ if (currentBatch.nonEmpty) {
+ logWarning(s"Commit current batch HARD_SPLIT partitions for $shuffleId: " +
+ s"${currentBatch.map(_.partition.getUniqueId).mkString("[", ",", "]")}")
+ val workerToRequests = currentBatch.flatMap { request =>
+ if (request.partition.getPeer != null) {
+ Seq(request.partition, request.partition.getPeer)
+ } else {
+ Seq(request.partition)
+ }
+ }.groupBy(_.getWorker)
+ shuffleCommittedInfo.inFlightCommitRequest.addAndGet(
+ workerToRequests.size)
+ return workerToRequests
Review Comment:
prefer not to use return, just add a else block surrounding Map.empty
--
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: dev-unsubscribe@celeborn.apache.org
For queries about this service, please contact Infrastructure at:
users@infra.apache.org