hadoop-mapreduce-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "nicu marasoiu (JIRA)" <j...@apache.org>
Subject [jira] [Updated] (MAPREDUCE-5287) Create a generic InputFormat wrapping any other InputFormat, to control the number of map tasks
Date Thu, 13 Jun 2013 11:51:20 GMT

     [ https://issues.apache.org/jira/browse/MAPREDUCE-5287?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]

nicu marasoiu updated MAPREDUCE-5287:
-------------------------------------

    Attachment: CompositeRecordReader.java
                CompositeInputSplit.java
                AggregatingInputFormat.java

implementation of the split-merging inputformat which is optimal in both distribution and
data locality only when single locations are used as in hbase. for multiple locations another
algorithm I can implement quite easy if interest exists
                
> Create a generic InputFormat wrapping any other InputFormat, to control the number of
map tasks
> -----------------------------------------------------------------------------------------------
>
>                 Key: MAPREDUCE-5287
>                 URL: https://issues.apache.org/jira/browse/MAPREDUCE-5287
>             Project: Hadoop Map/Reduce
>          Issue Type: Improvement
>          Components: mrv1, performance
>            Reporter: nicu marasoiu
>            Assignee: nicu marasoiu
>            Priority: Minor
>         Attachments: AggregatingInputFormat.java, CompositeInputSplit.java, CompositeRecordReader.java
>
>   Original Estimate: 4h
>  Remaining Estimate: 4h
>
> I wrote a generic InputFormat that wraps any other InputFormat, and creates CompositeInputSplits
to reduce the number of map tasks in a controllable manner while preserving data locality.
A correspondent CompositeRecordReader is written to iterate through underlying RecordReaders
as created by the underlying InputFormat for each underlying raw split.
> An application to this is to group TableSplits when the raw splits are coming from multiple
regions and are filtered with key ranges. We use this to shard/distribute a time based incremental
access to an hbase table.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Mime
View raw message