[IMPALA-1123] Combining fragments with compatible data partitions can lead to incorrect results due to type incompatibilities (missing casts).

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: Impala 1.3.1, Impala 1.4
• Fix Version/s: Impala 2.0
• Labels: correctness
• Assignee: Alexander Behm
• Reporter: Alexander Behm
• Created: 30/Jul/14 19:53
• Updated: 12/Sep/15 17:21
• Resolved: 01/Aug/14 04:32

Description

The following query returns incorrect results but the the equivalent [broadcast] version returns correct results.
The reason is that we hoist up the compatible agg child fragment into the partitioned join fragment, but the hoisted-up agg fragment has an exchange that hashes on "c" as a int type. Even though the "c" columns are value equivalent their hashes are only equal if hashed as the same type.
One way to fix this could be to add a cast to the data exchange of the agg fragment.

create table foo (c int);
insert into foo values(1),(2),(3),(4),(5),(6);

create table bar (c bigint);
insert into bar values(1),(2),(3),(4),(5),(6);

select f.c, b.c from
  (select c from foo union select c from foo) f
inner join [shuffle]
  (select c from bar union select c from bar) b
on f.c = b.c