You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@spark.apache.org by "Hyukjin Kwon (JIRA)" <ji...@apache.org> on 2018/12/24 05:00:00 UTC

[jira] [Commented] (SPARK-26407) For an external non-partitioned table, if add a directory named with k=v to the table path, select result will be wrong

    [ https://issues.apache.org/jira/browse/SPARK-26407?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16728142#comment-16728142 ] 

Hyukjin Kwon commented on SPARK-26407:
--------------------------------------

Why don't you just avoid the directory names like part=1 or empty strings? It doesn't looks a good practice to allow.

> For an external non-partitioned table, if add a directory named with k=v to the table path, select result will be wrong
> -----------------------------------------------------------------------------------------------------------------------
>
>                 Key: SPARK-26407
>                 URL: https://issues.apache.org/jira/browse/SPARK-26407
>             Project: Spark
>          Issue Type: Bug
>          Components: SQL
>    Affects Versions: 2.4.0
>            Reporter: Bao Yunz
>            Priority: Major
>              Labels: usability
>
> Scenario 1
> Create an external non-partitioned table, in which location directory has a directory named with "part=1" and its schema is (id, name), for example. And there is some data in the "part=1" directory. Then desc the table, we will find the "part" is added in table scehma as table column. when insert into the table with two columns data, will throw a exception that  target table has 3 columns but the inserted data has 2 columns. 
> Scenario 2
> Create an external non-partitioned table, which location path is empty and its scema is (id, name), for example. After several times insert operation, we add a directory named with "part=1" in the table location directory.  And there is some data in the "part=1" directory.  Then do insert and select operation, we will find the scan path is changed to "tablePath/part=1",so that we will get a wrong result.
>  The right logic should be that if a table is a non-partitioned table, adding a partition-like folder under tablePath should not change its schema and select result.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org