You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@spark.apache.org by "Wang Shuo (JIRA)" <ji...@apache.org> on 2018/12/20 08:39:00 UTC
[jira] [Comment Edited] (SPARK-26418) Only OpenBlocks without any
ChunkFetch for one stream will cause memory leak in ExternalShuffleService
[ https://issues.apache.org/jira/browse/SPARK-26418?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16725649#comment-16725649 ]
Wang Shuo edited comment on SPARK-26418 at 12/20/18 8:38 AM:
-------------------------------------------------------------
We use YarnShuffleService as aux service of NodeManager in our cluster. Full GC happens in some NodeManagers. We dump the heap memory, found that the Map in OneForOneStreamManager is 3G, almost 80% of heap size. Some applications have finished, but the StreamState is still in StreamManager.
Two reasons will cause this:
# OpenBlocks request is received and StreamState is initialized in server side. Then Client lost and no ChunkFetch request is sent to server for the stream.
# OpenBlocks request is received and StreamState is initialized in server side. ChunkFetch request for the stream is sent to server. But server is under heavy pressure and not able to handle the ChunkFetch request before timeout. Then client close connection.
Could we call StreamManager#registerChannel to associate the channel with StreamState when server handles OpenBlocks request?
was (Author: wang-shuo):
We use YarnShuffleService as aux service of NodeManager in our cluster. Full GC happens in some NodeManagers. We dump the heap memory, found that the Map in OneForOneStreamManager is 3G, almost 80% of heap size. Some applications have finished, but the StreamState is still in StreamManager.
Two reasons will cause this:
# OpenBlocks request is receivedand StreamState is initialized in server side. Then Client lost and no ChunkFetch request is sent to server for the stream.
# OpenBlocks request is received and StreamState is initialized in server side. ChunkFetch request for the stream is sent to server. But server is under heavy pressure and not able to handle the ChunkFetch request before timeout. Then client close connection.
Could we call StreamManager#registerChannel to associate the channel with StreamState when server handles OpenBlocks request?
> Only OpenBlocks without any ChunkFetch for one stream will cause memory leak in ExternalShuffleService
> ------------------------------------------------------------------------------------------------------
>
> Key: SPARK-26418
> URL: https://issues.apache.org/jira/browse/SPARK-26418
> Project: Spark
> Issue Type: Bug
> Components: Shuffle
> Affects Versions: 2.4.0
> Reporter: Wang Shuo
> Priority: Major
>
> In current code path, OneForOneStreamManager holds StreamState in a Map named streams.
> A StreamState is initialized and put into streams when OpenBlocks request received.
> One specific StreamState is removed from streams in two scenarios below:
> # The last chunk of a stream is fetched
> # The connection of ChunkFetch is closed
> StreamState will never be clean up, if OpenBlocks request is received without and following ChunkFetch request. This will cause memory leak in server side, which is harmful for long running service such as ExternalShuffleService.
>
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org