[
https://issues.apache.org/jira/browse/MAPREDUCE-5911?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14176509#comment-14176509
]
Gera Shegalov commented on MAPREDUCE-5911:
------------------------------------------
[~ivanmi], This fix does not work if the default TotalOrderPartitioner is used
instead of of SimplePartitioner. It will always fail because the partion file
will have been written into the output dir by the the time checkOutputSpec is
called. You should have taken my fix from MAPREDUCE-4879.
> Terasort TeraOutputFormat does not check for output directory existance
> -----------------------------------------------------------------------
>
> Key: MAPREDUCE-5911
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-5911
> Project: Hadoop Map/Reduce
> Issue Type: Bug
> Components: examples
> Reporter: Ivan Mitic
> Assignee: Bruno P. Kinoshita
> Priority: Minor
> Fix For: 2.6.0
>
> Attachments: HADOOP-5911.patch
>
>
> The enforcement that the directory must not yet exist is implemented in
> {{FileOutputFormat#checkOutputSpecs}} by throwing
> {{FileAlreadyExistsException}}. However, terasort uses a specialized output
> format, {{TeraOutputFormat}}, which is a subclass of {{FileOutputFormat}}.
> The subclass overrides {{checkOutputSpecs}}, but does not re-implement the
> existence check and throw {{FileAlreadyExistsException}}.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
