[
https://issues.apache.org/jira/browse/HADOOP-15577?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Steve Loughran resolved HADOOP-15577.
-------------------------------------
Resolution: Duplicate
We don't need the 0-rename stuff, because distcp, except in the --atomic mode,
isn't trying to do atomic operations.
What we do need, is for distcp to not upload to a temp file and rename each one
into place: remove that and for non-atomic uploads you eliminate the O(data)
delay after each upload.
Closing as a duplicate of that. *as that JIRA has no code/tests, I would
support anyone who sat down to do implement the feature*
There's also lots of work going on with HDFS to have an explicit multipart
upload mechanism for filesystems, which can be used for a block-by-block upload
to S3, this would improve distcp upload perf on files in HDFS > 1 block, as the
blocks could be uploaded in parallel with locality. Keep an eye on that
> Update distcp to use zero-rename s3 committers
> ----------------------------------------------
>
> Key: HADOOP-15577
> URL: https://issues.apache.org/jira/browse/HADOOP-15577
> Project: Hadoop Common
> Issue Type: Improvement
> Components: fs/s3, tools/distcp
> Affects Versions: 3.1.0
> Reporter: Tim Sammut
> Priority: Major
>
> Hello!
> distcp through 3.1.0 appears to copy files and then rename them into their
> final/destination filename.
> https://issues.apache.org/jira/browse/HADOOP-13786 added support for more
> efficient S3 committers that do not use renames.
> Please update distcp to use these efficient committers and no renames.
> Thanks!
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: common-dev-unsubscr...@hadoop.apache.org
For additional commands, e-mail: common-dev-h...@hadoop.apache.org
