[jira] [Created] (HIVE-15327) Outerjoin might produce wrong result depending on joinEmitInterval value

["Jesus Camacho Rodriguez (JIRA)" <ji...@apache.org>]

Jesus Camacho Rodriguez created HIVE-15327:
----------------------------------------------

             Summary: Outerjoin might produce wrong result depending on joinEmitInterval value
                 Key: HIVE-15327
                 URL: https://issues.apache.org/jira/browse/HIVE-15327
             Project: Hive
          Issue Type: Bug
          Components: Query Processor
    Affects Versions: 1.3.0, 2.2.0
            Reporter: Jesus Camacho Rodriguez
            Assignee: Jesus Camacho Rodriguez
            Priority: Critical


If joinEmitInterval is smaller than the group size, outerjoins might produce records with NULL appended values multiple times (once per group).

HIVE-4689 targeted the same problem. However, the fix does not seem to cover all cases (in particular, it will not apply to left outer joins with filter conditions on the left input). The solution in HIVE-4689 was to disable (override) joinEmitInterval value for those cases. This fix follows the same approach.

To reproduce the problem:
{code}
set hive.strict.checks.cartesian.product=false;
set hive.join.emit.interval=1;

CREATE TABLE test1 (key INT, value INT, col_1 STRING);
INSERT INTO test1 VALUES (99, 0, 'Alice');
INSERT INTO test1 VALUES (99, 2, 'Mat');
INSERT INTO test1 VALUES (100, 1, 'Bob');
INSERT INTO test1 VALUES (101, 2, 'Car');

CREATE TABLE test2 (key INT, value INT, col_2 STRING);
INSERT INTO test2 VALUES (102, 2, 'Del');
INSERT INTO test2 VALUES (103, 2, 'Ema');
INSERT INTO test2 VALUES (104, 3, 'Fli');


-- Equi-condition and condition on one input (left outer join)
SELECT *
FROM test1 LEFT OUTER JOIN test2
ON (test1.value=test2.value AND test1.key between 100 and 102)
LIMIT 10;

-- Condition on one input (left outer join)
SELECT *
FROM test1 LEFT OUTER JOIN test2
ON (test1.key between 100 and 102)
LIMIT 10;
{code}


For the *first* query, current (incorrect) result is:
{noformat}
 99     0       Alice   NULL    NULL    NULL
 100    1       Bob     NULL    NULL    NULL
 101    2       Car     103     2       Ema
 99     2       Mat     NULL    NULL    NULL
 101    2       Car     102     2       Del
 99     2       Mat     NULL    NULL    NULL
{noformat}

Expected (correct) result is:
{noformat}
 99     0       Alice   NULL    NULL    NULL
 100    1       Bob     NULL    NULL    NULL
 101    2       Car     103     2       Ema
 101    2       Car     102     2       Del
 99     2       Mat     NULL    NULL    NULL
{noformat}


For the *second* query, current (incorrect) result is:
{noformat}
 101    2       Car     104     3       Fli
 100    1       Bob     104     3       Fli
 99     2       Mat     NULL    NULL    NULL
 99     0       Alice   NULL    NULL    NULL
 101    2       Car     103     2       Ema
 100    1       Bob     103     2       Ema
 99     2       Mat     NULL    NULL    NULL
 99     0       Alice   NULL    NULL    NULL
 101    2       Car     102     2       Del
 100    1       Bob     102     2       Del
{noformat}

Expected (correct) result is:
{noformat}
 101    2       Car     104     3       Fli
 101    2       Car     103     2       Ema
 101    2       Car     102     2       Del
 100    1       Bob     104     3       Fli
 100    1       Bob     103     2       Ema
 100    1       Bob     102     2       Del
 99     2       Mat     NULL    NULL    NULL
 99     0       Alice   NULL    NULL    NULL
{noformat}



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)