[jira] [Commented] (PIG-4449) Optimize the case of Order by + Limit in nested foreach

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

    [ https://issues.apache.org/jira/browse/PIG-4449?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14349465#comment-14349465 ] 

Daniel Dai commented on PIG-4449:
---------------------------------

Good optimization. POSort already capture "limit" semantics, we just need to make use of it in POSort and enable it with an optimization rule.

> Optimize the case of Order by + Limit in nested foreach
> -------------------------------------------------------
>
>                 Key: PIG-4449
>                 URL: https://issues.apache.org/jira/browse/PIG-4449
>             Project: Pig
>          Issue Type: Improvement
>            Reporter: Rohini Palaniswamy
>
> This is one of the very frequently used patterns
> {code}
> grouped_data_set = group data_set by id;
> capped_data_set = foreach grouped_data_set
> {
>   ordered = order joined_data_set by timestamp desc;
>   capped = limit ordered $num;
>  generate flatten(capped);
> };
> {code}
> But this performs very poorly when there are millions of rows for a key in the groupby with lot of spills.  This can be easily optimized by pushing the limit into the InternalSortedBag and maintain only $num records any time and avoid memory pressure.



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