[jira] [Updated] (PIG-4767) Partition filter not pushed down when filter clause references variable from another load path

["Koji Noguchi (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/PIG-4767?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Koji Noguchi updated PIG-4767:
------------------------------
    Attachment: pig-4767-v01.patch

Skipping of optimization with filter expression containing scalar was done at PIG-1669.  However, I don't think we need to apply this to {{PartitionFilterOptimizer}} and {{PredicatePushdownOptimizer}} since {{FilterExtractor}} that both optimizer depend on would keep the filter expression that contains {{ScalarExpression}}. Original LOFilter will be preserved with updated filter expression containing this scalar.

Reverted the changes in both optimizers and added a test for PartitionFilter with scalar.
(I looked at Predicate/Orc testing but didn't find an easy way to add this test.)

[~daijy], appreciate if you could take a look.


> Partition filter not pushed down when filter clause references variable from another load path
> ----------------------------------------------------------------------------------------------
>
>                 Key: PIG-4767
>                 URL: https://issues.apache.org/jira/browse/PIG-4767
>             Project: Pig
>          Issue Type: Bug
>    Affects Versions: 0.15.0
>            Reporter: Anthony Hsu
>            Assignee: Koji Noguchi
>             Fix For: 0.18.0
>
>         Attachments: pig-4767-v01.patch
>
>
> To reproduce:
> {noformat:title=test.pig}
> a = load 'a.txt';
> a_group = group a all;
> a_count = foreach a_group generate COUNT(a) as count;
> b = load 'mytable' using org.apache.hcatalog.pig.HCatLoader();
> b = filter b by datepartition == '2015-09-01-00' and foo == a_count.count;
> dump b;
> {noformat}
> The above query ends up reading all the table partitions. If you remove the {{foo == a_count.count}} clause or replace {{a_count.count}} with a constant, then partition filtering happens properly.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)