[
https://issues.apache.org/jira/browse/HADOOP-2863?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Raghu Angadi updated HADOOP-2863:
---------------------------------
Description:
Why does FSDataOutputStream.close() call flush()? This stream itself does not
store any data that it needs to flush. It is a wrapper and it should just
invoke its outputstream's close().
For. e.g one bad side effect is that, in the case of DFSOutputStream which
extends FSOutputSummer, flush() inside close sends the current data even though
FSOutputSummer might have some data.. this left over data will be sent in side
close() (so it sends data in two different packets instead of one). Other
filesystems might have similar side effects.
I will submit a patch.
was:
Why does FSDataOutputStream.close() call flush()? This stream itself does not
store any data that it needs to flush. It is a wrapper and it should just
invoke its outputstream's close().
For. e.g one bad side effect is that, in the case of DFSOutputStream which
extends FSOutputSummer, flush() inside close sends the current data even though
FSOutputSummer might have some data.. this left over data will be sent in side
close() (so it sends data in two different packets instead of one). Other
filesystems might have similar side effects.
I will submit a patch.
Fix Version/s: (was: 0.17.0)
bq. For. e.g one bad side effect is that, in the case of DFSOutputStream which
extends FSOutputSummer, flush() inside close sends the current data even though
FSOutputSummer might have some data.. this left over data will be sent in side
close() (so it sends data in two different packets instead of one). Other
filesystems might have similar side effects.
I am fixing the above as part of HADOOP-1702 by renaming
{{DFSOutputStream.flush()}} to {{flushInternal()}} (because it is not a normal
flush() and needs to be called only inside close()).
> FSDataOutputStream should not flush() inside close().
> -----------------------------------------------------
>
> Key: HADOOP-2863
> URL: https://issues.apache.org/jira/browse/HADOOP-2863
> Project: Hadoop Core
> Issue Type: Bug
> Components: fs
> Reporter: Raghu Angadi
> Assignee: Raghu Angadi
> Attachments: HADOOP-2863.patch, HADOOP-2863.patch
>
>
> Why does FSDataOutputStream.close() call flush()? This stream itself does not
> store any data that it needs to flush. It is a wrapper and it should just
> invoke its outputstream's close().
> For. e.g one bad side effect is that, in the case of DFSOutputStream which
> extends FSOutputSummer, flush() inside close sends the current data even
> though FSOutputSummer might have some data.. this left over data will be sent
> in side close() (so it sends data in two different packets instead of one).
> Other filesystems might have similar side effects.
> I will submit a patch.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
