[PIG-2747] Support more predicate pushdown to a data source by pulling up multiple predicates from branches using the same data source

• Type: Improvement
• Status: Resolved
• Priority: Minor
• Resolution: Duplicate
• Assignee: Unassigned
• Reporter: Yu Xu
• Created: 11/Jun/12 18:16
• Updated: 19/Jan/14 02:00
• Resolved: 19/Jan/14 02:00

Description

consider the following example:

T = load ... ;
T1 = filter T by col == 'hello';
T2 = filter T by col =='world';

currently Pig optimizer does not combine the two predicates and cannot push down the predicates to the data sources (via LoadMetadata). Thus the data source cannot do any filtering. A full table/file scan is required.

A current more efficient workaround (by hand) is to rewrite the above script to the following equivalent one:

T = load ...;
T = filter T by col == 'hello' or col == 'world' ;
T1 = filter T by col == 'hello';
T2 = filter T by col == 'world';

the above script enables Pig to push down the predicate (col == 'hello' or col == 'world') to the data source to use available partitions/indexes for potentially much more efficient processing.

This JIRA is created to request PIG optimizer to perform the above type of optimization automatically.