[jira] [Commented] (PIG-4802) Autoparallelism should estimate less when there is combiner

["Daniel Dai (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/PIG-4802?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15141476#comment-15141476 ] 

Daniel Dai commented on PIG-4802:
---------------------------------

Want to add some clarity to the code:
1. factor *= DEFAULT_DISTINCT_FACTOR; => factor = DEFAULT_DISTINCT_FACTOR;
factor *= DEFAULT_AGGREGATION_FACTOR; => factor = DEFAULT_AGGREGATION_FACTOR;

2. Can we retain the simple version: getParallelismFactor(), the added getParallelismFactor(TezOperator successor) only be used by TezParallelismFactorVisitor. Also add a comment why the parameter is needed.

> Autoparallelism should estimate less when there is combiner
> -----------------------------------------------------------
>
>                 Key: PIG-4802
>                 URL: https://issues.apache.org/jira/browse/PIG-4802
>             Project: Pig
>          Issue Type: Improvement
>            Reporter: Rohini Palaniswamy
>            Assignee: Rohini Palaniswamy
>             Fix For: 0.16.0
>
>         Attachments: PIG-4802-1.patch
>
>
>   When there is a combiner, it reduces records by a lot. Auto-parallelism should take that into account. 
>   Also currently we multiply by a factor of 10 if there is any FLATTEN. Users usually have FLATTEN(group) when the group by key is a compound key and it ends up estimating high. Only FLATTEN of bag should be considered.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)