[HIVE-11576] Data loss in MapJoin

• Type: Bug
• Status: Reopened
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 1.2.0
• Assignee: Matt McCline
• Reporter: Ted Xu
• Created: 17/Aug/15 00:59
• Updated: 12/Jan/23 11:01

Description

In query (TPC-H query4)

query4.sql

create table q4_result as 
select 
o_orderpriority, 
count(*) as order_count 
from 
orders o 
join 
( 
select 
distinct l_orderkey 
from 
( 
select 
* 
from 
lineitem 
where 
l_commitdate < l_receiptdate 
) tab1 
) tab2 
on tab2.l_orderkey = o.o_orderkey 
where 
o.o_orderdate >= '1993-07-01' and o.o_orderdate < '1993-10-01' 
group by 
o_orderpriority 
order by 
o_orderpriority;

The query will cause data-loss if MapJoin is enabled. Both side of join have expected output but some data can't be joined together here. After disabling auto convert join, the problem is gone.

Context:
l_orderkey & o_orderkey are bigint.
vectorized execution enabled.
execution engine is tez.