[
https://issues.apache.org/jira/browse/HIVE-27324?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17720894#comment-17720894
]
Seonggon Namgung commented on HIVE-27324:
-----------------------------------------
I think the first issue is intended behavior while the second issue is bug.
1. Both parent query and subquery contain null.
According to the comment in QBSubQuery.java, Not-In query returns empty result
if its subquery contains null.
In Hive (and other database systems), the result of something <> NULL is NULL,
which is treated as false in conditional clause.
So "a not in (NULL, a, b, c, ...)" is always evaluated to false, and the result
of parent query becomes empty.
(You can also refer to [this
document|https://issues.apache.org/jira/secure/attachment/12614003/SubQuerySpec.pdf].)
2. Only parent query contains null.
Because NULL not in (a, b, c, ...) is always false too, the correct result
should be 5 rows.
(I also got 5 rows when I ran the second query in PostgreSQL.)
So we can conclude that Hive generates wrong query plan for Not-In query when
CBO is disabled.
I compared the query plan for the second query with and without CBO, and I
found that condition "t3.age is not null" is missing in non-CBO plan, which
seems to be the root cause of this issue.
> Hive query with NOT IN condition is giving incorrect results when the sub
> query table contains the null value.
> --------------------------------------------------------------------------------------------------------------
>
> Key: HIVE-27324
> URL: https://issues.apache.org/jira/browse/HIVE-27324
> Project: Hive
> Issue Type: Bug
> Components: Hive
> Affects Versions: All Versions, 3.1.0
> Reporter: Shobika Selvaraj
> Priority: Major
>
> Hive query giving empty results when the sub query table contains the null
> value.
> We encountered two issues here.
> 1) The query - "select * from t3 where age not in (select distinct(age) age
> from t1);" is giving empty results when the table t1 contains a null value.
> Disabling cbo didn't helped here.
> 2) Let's consider the table t3 has null value and table t1 doesn't have any
> null values. Now if we run the above query it is returning other data's but
> not the null value from t3. If we disable the cbo then it's giving result
> with null value.
>
> *REPRO STEPS WITH DETAILED EXPLANATION AS BELOW:*
> *FIRST ISSUE:*
> Create two tables and insert data with null values as below:
> -------
> create table t3 (id int,name string, age int);
>
> insert into t3
> values(1,'Sagar',23),(2,'Sultan',NULL),(3,'Surya',23),(4,'Raman',45),(5,'Scott',23),(6,'Ramya',5),(7,'',23),(8,'',23),(9,'ron',3),(10,'Sam',22),(11,'nick',19),(12,'fed',18),(13,'kong',13),(14,'hela',45);
>
> create table t1 (id int,name string, age int);
> insert into t1
> values(1,'Sagar',23),(2,'Sultan',NULL),(3,'Surya',23),(4,'Raman',45),(5,'Scott',23),(6,'Ramya',5),(7,'',23),(8,'',23);
> -------
>
> Then executed the below query:
> ------
> select * from t3
> where age not in (select distinct(age) age from t1);
> ------
>
> The result should be as below:
> {code:java}
> +--------+----------+---------+
> | t3.id | t3.name | t3.age |
> +--------+----------+---------+
> | 9 | ron | 3 |
> | 10 | Sam | 22 |
> | 11 | nick | 19 |
> | 12 | fed | 18 |
> | 13 | kong | 13 |
> +--------+----------+---------+
> 5 rows selected (35.897 seconds) {code}
> But when we run the above query it is giving zero records:
> {code:java}
> 0: jdbc:hive2://zk0-shobia.lhlexkfu3vfebcezzj> select * from t3
> . . . . . . . . . . . . . . . . . . . . . . .> where age not in (select
> distinct(age) age from t1);
> INFO : Compiling
> command(queryId=hive_20230427164202_e25b671a-f3bd-41e4-b364-844466305d96):
> select * from t3
> where age not in (select distinct(age) age from t1)
> INFO : Warning: Map Join MAPJOIN[37][bigTable=?] in task 'Map 1' is a cross
> product
> ......
> .....
> INFO : Completed executing
> command(queryId=hive_20230427164202_e25b671a-f3bd-41e4-b364-844466305d96);
> Time taken: 10.191 seconds
> INFO : OK
> +--------+----------+---------+
> | t3.id | t3.name | t3.age |
> +--------+----------+---------+
> +--------+----------+---------+
> No rows selected (12.17 seconds) {code}
> The query works fine when we use nvl function or not null condition.
>
> So as a workaround we can use nvl function for both main and sub query as
> below:
> {code:java}
> select * from t3 where nvl(age,'-') not in (select distinct(nvl(age,'-')) age
> from t1); {code}
>
> *SECOND ISSUE:*
> Also while testing multiple scenario's i found one more issue as well.
> When the sub query table (t1) doesn't contain any null values then the query
> is giving result but it is ignoring the null values of the main table(t3) .
>
> For example: Created another table t4 and inserted the data's without any
> null values:
> create table t4 (id int,name string, age int);
> insert into t4
> values(1,'Sagar',23),(3,'Surya',23),(4,'Raman',45),(5,'Scott',23),(6,'Ramya',5),(7,'',23),(8,'',23);
> Now i tested with the below query and it gives 5 records. The count should be
> six and it omitted the null value of the table t3:
> {code:java}
> 0: jdbc:hive2://zk0-shobia.lhlexkfu3vfebcezzj> select * from t3
> . . . . . . . . . . . . . . . . . . . . . . .> where age not in (select
> distinct(age) age from t4);
> INFO : Compiling
> command(queryId=hive_20230427164745_f20f47ce-614d-493d-8910-99a118de089c):
> select * from t3
> where age not in (select distinct(age) age from t4)
> INFO : Warning: Map Join MAPJOIN[37][bigTable=?] in task 'Map 1' is a cross
> product
> ......
> ......
> INFO : Completed executing
> command(queryId=hive_20230427164745_f20f47ce-614d-493d-8910-99a118de089c);
> Time taken: 17.724 seconds
> INFO : OK
> +--------+----------+---------+
> | t3.id | t3.name | t3.age |
> +--------+----------+---------+
> | 9 | ron | 3 |
> | 10 | Sam | 22 |
> | 11 | nick | 19 |
> | 12 | fed | 18 |
> | 13 | kong | 13 |
> +--------+----------+---------+
> 5 rows selected (19.674 seconds) {code}
> But disabling CBO fixed this issue:
> {code:java}
> 0: jdbc:hive2://zk0-shobia.lhlexkfu3vfebcezzj> set hive.cbo.enable=false;
> No rows affected (0.02 seconds)
> 0: jdbc:hive2://zk0-shobia.lhlexkfu3vfebcezzj> select * from t3
> . . . . . . . . . . . . . . . . . . . . . . .> where age not in (select
> distinct(age) age from t4);
> INFO : Compiling
> command(queryId=hive_20230427165029_6c3167b0-c706-4230-a32f-2e4bc528be8f):
> select * from t3
> where age not in (select distinct(age) age from t4)
> INFO : Warning: Map Join MAPJOIN[41][bigTable=t3] in task 'Map 1' is a cross
> product
> INFO : Semantic Analysis Completed (retrial = false)
> .......
> .......
> INFO : Completed executing
> command(queryId=hive_20230427165029_6c3167b0-c706-4230-a32f-2e4bc528be8f);
> Time taken: 13.738 seconds
> INFO : OK
> +--------+----------+---------+
> | t3.id | t3.name | t3.age |
> +--------+----------+---------+
> | 2 | Sultan | NULL |
> | 9 | ron | 3 |
> | 10 | Sam | 22 |
> | 11 | nick | 19 |
> | 12 | fed | 18 |
> | 13 | kong | 13 |
> +--------+----------+---------+
> 6 rows selected (14.526 seconds) {code}
> I am not sure this how hive works or not by omitting the null values. Even if
> it omits the null values then it should give the other records{*}(FOR ISSUE
> 1){*} for the sub query table which contains null values.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
