spark-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Sean Owen (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (SPARK-23370) Spark receives a size of 0 for an Oracle Number field and defaults the field type to be BigDecimal(30,10) instead of the actual precision and scale
Date Fri, 09 Feb 2018 22:19:00 GMT

    [ https://issues.apache.org/jira/browse/SPARK-23370?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16359047#comment-16359047
] 

Sean Owen commented on SPARK-23370:
-----------------------------------

It's possible to implement that just in the JDBC dialect for Oracle I suppose. Is it extra
overhead? that is I wonder about leaving in the workaround that impacts all Oracle users for
a long time.

> Spark receives a size of 0 for an Oracle Number field and defaults the field type to
be BigDecimal(30,10) instead of the actual precision and scale
> ---------------------------------------------------------------------------------------------------------------------------------------------------
>
>                 Key: SPARK-23370
>                 URL: https://issues.apache.org/jira/browse/SPARK-23370
>             Project: Spark
>          Issue Type: Improvement
>          Components: SQL
>    Affects Versions: 2.2.1
>         Environment: Spark 2.2
> Oracle 11g
> JDBC ojdbc6.jar
>            Reporter: Harleen Singh Mann
>            Priority: Minor
>         Attachments: Oracle KB Document 1266785.pdf
>
>
> Currently, on jdbc read spark obtains the schema of a table from using {color:#654982}
resultSet.getMetaData.getColumnType{color}
> This works 99.99% of the times except when the column of Number type is added on an Oracle
table using the alter statement. This is essentially an Oracle DB + JDBC bug that has been
documented on Oracle KB and patches exist. [oracle KB|https://support.oracle.com/knowledge/Oracle%20Database%20Products/1266785_1.html]
> {color:#ff0000}As a result of the above mentioned issue, Spark receives a size of 0 for
the field and defaults the field type to be BigDecimal(30,10) instead of what it actually
should be. This is done in OracleDialect.scala. This may cause issues in the downstream application
where relevant information may be missed to the changed precision and scale.{color}
> _The versions that are affected are:_ 
>  _JDBC - Version: 11.2.0.1 and later   [Release: 11.2 and later ]_
>  _Oracle Server - Enterprise Edition - Version: 11.1.0.6 to 11.2.0.1_  _[Release:
11.1 to 11.2]_ 
> +Proposed approach:+
> There is another way of fetching the schema information in Oracle: Which is through
the all_tab_columns table. If we use this table to fetch the precision and scale of Number
time, the above issue is mitigated.
>  
> {color:#14892c}{color:#f6c342}I can implement the changes, but require some inputs on
the approach from the gatekeepers here{color}.{color}
>  {color:#14892c}PS. This is also my first Jira issue and my first fork for Spark, so
I will need some guidance along the way. (yes, I am a newbee to this) Thanks...{color}



--
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


Mime
View raw message