[ 
https://issues.apache.org/jira/browse/HIVE-8448?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Yongzhi Chen updated HIVE-8448:
-------------------------------
    Affects Version/s: 0.13.1
               Status: Patch Available  (was: Open)

when SemanticAnalyzer check union plan, it use getCommonClassForUnionAll, but 
when evaluate union operator it uses getCommonClass. The inconsistency cause 
some queries with multiple union all on date type column can pass the analyzer 
but fail with HiveException: Incompatible types at execute time.
Fixed by use new updateForUnionAll method which use getCommonClassForUnionAll 
to update column for union operator.

> Union All might not work due to the type conversion issue
> ---------------------------------------------------------
>
>                 Key: HIVE-8448
>                 URL: https://issues.apache.org/jira/browse/HIVE-8448
>             Project: Hive
>          Issue Type: Bug
>    Affects Versions: 0.13.1
>            Reporter: Chaoyu Tang
>            Assignee: Yongzhi Chen
>            Priority: Minor
>         Attachments: HIVE-8448.1.patch
>
>
> create table t1 (val date);
> insert overwrite table t1 select '2014-10-10' from src limit 1;
> create table t2 (val varchar(10));
> insert overwrite table t2 select '2014-10-10' from src limit 1; 
> ==
> Query:
> select t.val from
> (select val from t1
> union all
> select val from t1
> union all
> select val from t2
> union all
> select val from t1) t;
> ==
> Will throw exception: 
> {code}
> Caused by: org.apache.hadoop.hive.ql.metadata.HiveException: Incompatible 
> types for union operator
>       at 
> org.apache.hadoop.hive.ql.exec.UnionOperator.initializeOp(UnionOperator.java:86)
>       at org.apache.hadoop.hive.ql.exec.Operator.initialize(Operator.java:380)
>       at org.apache.hadoop.hive.ql.exec.Operator.initialize(Operator.java:464)
>       at 
> org.apache.hadoop.hive.ql.exec.Operator.initializeChildren(Operator.java:420)
>       at 
> org.apache.hadoop.hive.ql.exec.SelectOperator.initializeOp(SelectOperator.java:65)
>       at org.apache.hadoop.hive.ql.exec.Operator.initialize(Operator.java:380)
>       at org.apache.hadoop.hive.ql.exec.Operator.initialize(Operator.java:464)
>       at 
> org.apache.hadoop.hive.ql.exec.Operator.initializeChildren(Operator.java:420)
>       at 
> org.apache.hadoop.hive.ql.exec.TableScanOperator.initializeOp(TableScanOperator.java:193)
>       at org.apache.hadoop.hive.ql.exec.Operator.initialize(Operator.java:380)
>       at 
> org.apache.hadoop.hive.ql.exec.MapOperator.initializeOp(MapOperator.java:443)
>       at org.apache.hadoop.hive.ql.exec.Operator.initialize(Operator.java:380)
>       at 
> org.apache.hadoop.hive.ql.exec.mr.ExecMapper.configure(ExecMapper.java:133)
>       ... 22 more
> {code}
> It was because at this query parse step, getCommonClassForUnionAll is used, 
> but at execution getCommonClass is used. They are not used consistently in 
> union. The later one does not support the implicit conversion from date to 
> string, which is the problem cause.
> The change might be simple to fix this particular union issue but I noticed 
> that there are three versions of getCommonClass: getCommonClass, 
> getCommonClassForComparison, getCommonClassForUnionAll, and wonder if they 
> need to be cleaned and refactored.



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

Reply via email to