Mailing-List: contact crunch-dev-help@incubator.apache.org; run by ezmlm
Precedence: bulk
Reply-To: crunch-dev@incubator.apache.org
Date: Mon, 17 Sep 2012 21:05:07 +1100 (NCT)
From: "Gabriel Reid (JIRA)" <jira@apache.org>
To: crunch-dev@incubator.apache.org
Message-ID: <2008861694.87094.1347876307589.JavaMail.jiratomcat@arcas>
In-Reply-To: <1961614037.5257.1345923120163.JavaMail.jiratomcat@arcas>
Subject: [jira] [Commented] (CRUNCH-51) PCollection#sort relies on using a
 single reducer for total order sorting
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit


    [ https://issues.apache.org/jira/browse/CRUNCH-51?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13456920#comment-13456920 ] 

Gabriel Reid commented on CRUNCH-51:
------------------------------------

I was just doing some general testing with this, and ran into a couple of issues:

- it seems that this patch doesn't work with descending (i.e. reverse-order) sorting. When I tried reverse-order sorting strings, I got an IOException with "Split points are out of order"
- sorting reflection-based Avro or generic records in Avro doesn't seem to work -- in both cases I got a message of "Number of keys to write in partion file:0" and then a materialize crashes due to the file to be materialized not existing

Apart from these issues, I think it's looking pretty good. The Avro reflection sorting issue seems pretty minor to me (i.e. not a very common use case), but this is something that does work right now with the single-reducer solution.
                
> PCollection#sort relies on using a single reducer for total order sorting
> -------------------------------------------------------------------------
>
>                 Key: CRUNCH-51
>                 URL: https://issues.apache.org/jira/browse/CRUNCH-51
>             Project: Crunch
>          Issue Type: Improvement
>    Affects Versions: 0.3.0
>            Reporter: Gabriel Reid
>         Attachments: 0001-CRUNCH-51-Total-Order-Sort.patch, CRUNCH-51.patch
>
>
> The total-order sorting provided by the Sort class (and therefore PCollection#sort) relies on using a single reducer in order to provide total-order sorting. This is very inefficient for large datasets, and should be replaced with a total order partitioner instead.
> For more information, see CRUNCH-23 (and possibly also MAPREDUCE-4574).

--
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