[
https://issues.apache.org/jira/browse/FLINK-21138?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17273026#comment-17273026
]
Till Rohrmann commented on FLINK-21138:
---------------------------------------
Thanks for posting this example. I am wondering whether the problem isn't that
we are using the {{Thread.currentThread().getContextClassLoader()}}
[here|https://github.com/apache/flink/blob/master/flink-core/src/main/java/org/apache/flink/api/java/typeutils/runtime/kryo/KryoSerializer.java#L626]
instead of {{original.getClass().getClassLoader()}}. I think this should fix
the problem.
Test-wise you could take a look at the
{{PojoSerializerUpgradeTest.testPojoSerializerUpgrade}} which creates a new
class and instantiates an {{URLClassLoader}} to load it. If you then use the
test thread, it should not know about this new class.
> KvStateServerHandler is not invoked with user code classloader
> --------------------------------------------------------------
>
> Key: FLINK-21138
> URL: https://issues.apache.org/jira/browse/FLINK-21138
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Queryable State
> Affects Versions: 1.11.2
> Reporter: Maciej Prochniak
> Priority: Major
> Attachments: TestJob.java, stacktrace
>
>
> When using e.g. custom Kryo serializers user code classloader has to be set
> as context classloader during invocation of methods such as
> TypeSerializer.duplicat()
> KvStateServerHandler does not do this, which leads to exceptions like
> ClassNotFound etc.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
