[PIG-4646] PushUpFilter should not push before nested projection with FILTER operators

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 0.11.1
• Fix Version/s: 0.12.0
• Assignee: Daniel Dai
• Reporter: Haishan Liu
• Created: 30/Jul/15 06:32
• Updated: 04/Aug/15 00:06
• Resolved: 04/Aug/15 00:06

Description

Verified the problem in 0.11.1. In short, filter should not be pushed before a nested foreach in which another filter operator is present. See the following minimum example:

cat data;

(1, {(1000, 'a'), (1001, 'b')})
(2, {(2000, 'a'), (2001, 'b'), (2002, 'c')})

A = load 'data' as (id:int, hits:{(score:int, name:chararray)});
B = foreach A {
  filtered = filter hits by score > 2000;
  generate id, filtered;
};

dump B;

(1,{})
(2,{(2001,'b'),(2002,'c')})

C = filter B by SIZE(filtered) > 0;

dump C;

(1,{})
(2,{(2001,'b'),(2002,'c')})

The desired result can be achieved with either '-optimizer_off PushUpFilter' when invoking Pig, or using the following convoluted way:

C = foreach B generate SIZE(filtered) as size, id, filtered;
D = filter C by size > 0;
E = foreach D generate id, filtered;

dump E;

(2,{(2001,'b'),(2002,'c')})