[
https://issues.apache.org/jira/browse/HADOOP-1226?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12497951
]
Doug Cutting commented on HADOOP-1226:
--------------------------------------
> Please comment on whether keeping the task output path as a fullyQualified
> path [...]
Sigh. That code shouldn't be there in the first place. I was asleep when
HADOOP-1127 was committed. We should try to keep the mapred kernel from
touching input or output files, leaving that to InputFormats and OutputFormats.
It's a church/state thing. (Also, taskOutputPath stuff is reduce specific, so
if anywhere, it belongs in ReduceTask, no?)
File renaming for speculative reduce belongs in OutputFormatBase (which should
really be called FileOutputFormat). We'll need to add another method to the
OutputFormat interface, something like complete(JobConf, String tempName,
String finalName), that's called for each task to rename its temporary output
when speculative reduces are enabled, where tempName is what was passed to
getRecordWriter(). I'll file an issue for this.
Back to your question: I don't know why relative paths are used there. Perhaps
Arun does, since he wrote that code as a part of HADOOP-1127.
> makeQualified should return an instance of a DfsPath when passed a DfsPath
> ----------------------------------------------------------------------------
>
> Key: HADOOP-1226
> URL: https://issues.apache.org/jira/browse/HADOOP-1226
> Project: Hadoop
> Issue Type: Bug
> Components: dfs
> Reporter: Koji Noguchi
> Assigned To: dhruba borthakur
> Fix For: 0.14.0
>
> Attachments: makeQualified2.patch
>
>
> Currently, it returns a instance of a Path.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
