[
https://issues.apache.org/jira/browse/PIG-196?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12590558#action_12590558
]
Benjamin Reed commented on PIG-196:
-----------------------------------
This patch looks like it assumes that splitSpec is not null if there isn't a
groupSpec. It would be nice to add a comment stating that assumption and
justifying it. Looking back at MapReduceLauncher doesn't hint that this is the
case. I think it is important so that later changes do not break the
justification for this assumption. (Or at least it will give the person
debugging a hint at where to look.)
> Pig should use '-reducer NONE' for map-only jobs
> ------------------------------------------------
>
> Key: PIG-196
> URL: https://issues.apache.org/jira/browse/PIG-196
> Project: Pig
> Issue Type: Bug
> Reporter: Arun C Murthy
> Assignee: Arun C Murthy
> Attachments: PIG-196_0_20080412.patch
>
>
> Currently, for map-only jobs, Pig writes map-outputs directly to HDFS and
> then sends zero data to reducers. The problem with this is two fold:
> * Reduce slots are unnecessarily wasted on the cluster
> * Reduces write empty files to HDFS putting pressure on the Namenode
> Both these can we very easily avoided by just calling:
> {noformat}
> job.setNumReduces(0);
> {noformat}
> and letting Hadoop Map-Reduce take care of writing map-outputs directly to
> HDFS.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
