[
https://issues.apache.org/jira/browse/MAPREDUCE-2293?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Harsh J Chouraria updated MAPREDUCE-2293:
-----------------------------------------
Attachment: mapreduce.mo.removecheck.r1.diff
For extensions to be applied, I think one needs to make a change in the
OutputFormat class chosen. MultipleOutputs can only handle the part before the
'-XXXXX' (partition) numbering in Map/Reduce outputs, not after it.
I've posted a patch that removes the check (from both old and new API). I can't
post the result of an {{ant test-patch}} since that isn't working for me right
now (Mumak build is failing for some reason, in MR trunk). I'll post that when
I get it working.
This should be marked as an Incompatible Change in my opinion, as it is a
removal of a strong validation. People may also be relying on the
MultiName_OutputName-Partition syntax via string splits, etc. in the Stable API
MO class.
Also, I'm curious to see if allowing any character to go in is a good idea
'Path' wise. Does HDFS have any restrictions on Filenames? I've not seen a
documentation on it (although I think it is pretty POSIX compliant), but
HDFS-13 points out that there may be some trouble, any thoughts on that?
> Enhance MultipleOutputs to allow additional characters in the named output
> name
> -------------------------------------------------------------------------------
>
> Key: MAPREDUCE-2293
> URL: https://issues.apache.org/jira/browse/MAPREDUCE-2293
> Project: Hadoop Map/Reduce
> Issue Type: Improvement
> Affects Versions: 0.21.0
> Reporter: David Rosenstrauch
> Assignee: Harsh J Chouraria
> Priority: Minor
> Attachments: mapreduce.mo.removecheck.r1.diff
>
>
> Currently you are only allowed to use alpha-numeric characters in a named
> output name in the MultipleOutputs class. This is a bit of an onerous
> restriction, as it would be extremely convenient to be able to use non
> alpha-numerics in the name too. (E.g., a '.' character would be very
> helpful, so that you can use the named output name for holding a file
> name/extension. Perhaps '-' and a '_' characters as well.)
> The restriction seems to be somewhat arbitrary - it appears to be only
> enforced in the checkTokenName method. (Though I don't know if there's any
> downstream impact by loosening this restriction.)
> Would be extremely helpful/useful to have this fixed though!
--
This message is automatically generated by JIRA.
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
