[
https://issues.apache.org/jira/browse/FLINK-3228?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15141650#comment-15141650
]
Stephan Ewen commented on FLINK-3228:
-------------------------------------
[~ndimiduk] Are you using Flink's regular {{JDBCOutputFormat}}? That one
actually uses Gyula's trick already.
One thing you can try is change the {{JDBCOutputFormat}} to bind the class to
the UserCodeClassloader explicitly. Replacing the following line:
{code}
Class.forName(drivername);
{code}
with this
{code}
Class.forName(drivername, true, getRuntimeContext().getUserCodeClassLoader());
{code}
> Cannot submit multiple streaming involving JDBC drivers
> -------------------------------------------------------
>
> Key: FLINK-3228
> URL: https://issues.apache.org/jira/browse/FLINK-3228
> Project: Flink
> Issue Type: Bug
> Components: Streaming
> Affects Versions: 0.10.1
> Reporter: Nick Dimiduk
>
> I have a streaming pipeline that writes to JDBC in its SinkFunction. When
> testing in local mode, I cannot submit the same job multiple times without
> restarting flink. Failing to restart flink results in
> {{java.sql.SQLException: No suitable driver found}}. I think this has to do
> with how flink loads user classes and how the {{DriverManager}} resolves the
> JDBC Connection class.
> I'm not sure if this is only when running in local mode, or in general. I've
> been using YARN for cluster jobs, where I have 1:1 cluster to jobs. I suspect
> it will impact any cluster mode where processes are recycled between jobs.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
