[
https://issues.apache.org/jira/browse/HDFS-3041?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Todd Lipcon updated HDFS-3041:
------------------------------
Attachment: test.txt
Here's a test modification which shows the problem.
It's not trivial to fix... will work on this in the coming weeks.
> DFSOutputStream.close doesn't properly handle interruption
> ----------------------------------------------------------
>
> Key: HDFS-3041
> URL: https://issues.apache.org/jira/browse/HDFS-3041
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: hdfs client
> Affects Versions: 0.23.0, 0.24.0
> Reporter: Todd Lipcon
> Assignee: Todd Lipcon
> Attachments: test.txt
>
>
> TestHFlush.testHFlushInterrupted can fail occasionally due to a race: if a
> thread is interrupted while calling close(), then the {{finally}} clause of
> the {{close}} function sets {{closed = true}}. At this point it has enqueued
> the "end of block" packet to the DNs, but hasn't called {{completeFile}}.
> Then, if {{close}} is called again (as in the test case), it will be
> short-circuited since {{closed}} is already true. Thus {{completeFile}} never
> ends up getting called. This also means that the test can fail if the
> pipeline is running slowly, since the assertion that the file is the correct
> length won't see the last packet or two.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
