drill-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "ASF GitHub Bot (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (DRILL-6063) Set correct ThreadContext ClassLoader before using Hadoop Configuration class in DrillClient
Date Tue, 09 Jan 2018 01:31:00 GMT

    [ https://issues.apache.org/jira/browse/DRILL-6063?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16317504#comment-16317504
] 

ASF GitHub Bot commented on DRILL-6063:
---------------------------------------

Github user asfgit closed the pull request at:

    https://github.com/apache/drill/pull/1079


> Set correct ThreadContext ClassLoader before using Hadoop Configuration class in DrillClient
> --------------------------------------------------------------------------------------------
>
>                 Key: DRILL-6063
>                 URL: https://issues.apache.org/jira/browse/DRILL-6063
>             Project: Apache Drill
>          Issue Type: Improvement
>          Components: Client - Java
>    Affects Versions: 1.11.0
>            Reporter: Sorabh Hamirwasia
>            Assignee: Sorabh Hamirwasia
>              Labels: ready-to-commit
>             Fix For: 1.13.0
>
>
> Drill Client uses Hadoop Configuration class in security mechanism factories. Configuration
class internally uses ThreadContext class loader to get the default resources for configurations.
When Drill JDBC driver is used with 3rd party tool like Squirrel then it creates a class loader
with all Drill Driver related dependencies. This class loader is used to instantiate an object
of Driver class in a separate thread. But the thread context class loader has dependencies
related to Squirrel only. As a result of this when Hadoop Configuration class tries to load
the resources with default values it doesn't find them in class path.
> If we specify hadoop dependencies in Squirrel's classpath instead of Driver dependencies
class path then it will work since Thread Context class loader will have Hadoop libraries
loaded too and will find the resources. But this is not good as it is tying up one Squirrel
instance with one version of Hadoop which might not be true in all the cases like here when
same Squirrel instance can be used to load multiple versions of Drill Driver having different
hadoop version dependencies.
> To deal with this issue a simple fix is to set the Thread Context class loader to be
same as Drill Driver dependencies class loader before accessing Hadoop Configuration class
in Drill Client and later restore it back.
> Some pointers in code:
> * Place where JDBC driver is loaded as part of separate thread in Squirrel: [OpenConnectionCommand.java|https://sourceforge.net/p/squirrel-sql/git/ci/master/tree/sql12/core/src/net/sourceforge/squirrel_sql/client/mainframe/action/OpenConnectionCommand.java#l167]
> * Place where JDBC driver class is instantiated in Squirrel using Driver dependencies
class loader: [SQLDriverManager.java|https://sourceforge.net/p/squirrel-sql/git/ci/master/tree/sql12/core/src/net/sourceforge/squirrel_sql/fw/sql/SQLDriverManager.java#l74]
> * Place where Hadoop Configuration class looks into Thread Context Class Loader: [Configuration.java|https://github.com/apache/hadoop-common/blob/trunk/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/conf/Configuration.java#L615]



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

Mime
View raw message