[
https://issues.apache.org/jira/browse/FLINK-23724?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17401550#comment-17401550
]
Yuan Mei commented on FLINK-23724:
----------------------------------
*IMPORTANT* for reference (based on offline discussion)
Why pulling {{closeBufferPool}} into the {{fail}} method and why not directly
calling BufferWritingResultPartition#close() method in task cancelation (same
reason why not put bufferbuilder#close in the
BufferWritingResultPartition#releaseInternal method).
Because of potential race condition:
Task calculation thread and task thread may access {{bufferbuilder}} at the
same time. In the current solution, bufferbuilder release is done in the
task#doRun#finally only (task thread only).
> Network buffer leak when ResultPartition is released (failover)
> ---------------------------------------------------------------
>
> Key: FLINK-23724
> URL: https://issues.apache.org/jira/browse/FLINK-23724
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Network
> Affects Versions: 1.14.0, 1.13.2
> Reporter: Yingjie Cao
> Assignee: Yingjie Cao
> Priority: Blocker
> Labels: pull-request-available
> Fix For: 1.14.0, 1.13.3
>
>
> The BufferBuilders in BufferWritingResultPartition are not properly released
> when ResultPartition is released.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
