[
https://issues.apache.org/jira/browse/MAPREDUCE-971?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Aaron Kimball updated MAPREDUCE-971:
------------------------------------
Attachment: MAPREDUCE-971.patch
This patch fixes the problem by explcitly creating the temp directory. File
open operations in, e.g., hdfs, will auto-create the tmpdir. But in s3n, which
expects an object with the name {{_somename_$folder$}}, this won't happen. As a
result, the {{fullyDelete()}} call fails (silently) because the folder doesn't
exist, even though there are objects with the tmpdir prefix in their object
names.
I tested this patch manually by verifying temp dir creation during a distcp to
s3n, and verifying that the temp dir object was removed at the end of the
transfer.
> distcp does not always remove distcp.tmp.dir
> --------------------------------------------
>
> Key: MAPREDUCE-971
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-971
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Components: distcp
> Reporter: Aaron Kimball
> Assignee: Aaron Kimball
> Attachments: MAPREDUCE-971.patch
>
>
> Sometimes distcp leaves behind its tmpdir when the target filesystem is s3n.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
