Wildfly cluster: session de-serialization fails

[mbolcina]

We have problem in Wildfly Cluster environment. It fails to deserialize web session.

We have web application with Vaadin web framework. Vaadin is stateful framework and has components on session.

This are steps to repreduce:

* extract testWindow.zip

* run: mvn wildfly:deploy

* open http://localhost:8080/testWindow-1.0-SNAPSHOT/

* open window (click "Click Me" button)

* trigger redeploy with: mvn wildfly:deploy

* try to close vaadin window (small popup in web page)

When redeploy happened session was serialized successfully. But when deserialzation happens it fails with error (stacktrace attached).

Since web session was serialized successfully but failed when de-serializing we suspect that this is not Vaadin related problem, but that there might be some problems with how wildfly is un-marshalling. Maybe there is some class loading problem? Is this type of error some known error?

[rhusar]

What version of WildFly is this?

[pferraro]

It's a little hard to tell what's happening here.  I'm not at all familiar with vaadin, but given the exception, it looks to me like the VaadinSession class descriptor isn't getting serialized properly (i.e. it is missing the module identifier expected by ModularClassResolver).

[mbolcina]

Problem was with serialization of this class. If array is length 0, then deserialization fails. If i put something in this array, then deserialization is successful.

public class WindowState extends PanelState {

    public Connector[] contentDescription = new Connector[0];

    ...

}

For now i use workaround to always set this field, so that array is not empty. But shouldn't empty array work with jboss marshalling, or is this some know issue?

[jaikiran]

M Bol wrote:

Problem was with serialization of this class. If array is length 0, then deserialization fails. If i put something in this array, then deserialization is successful.

This sounds similar to the other issue being discussed here https://developer.jboss.org/thread/249485

[mbolcina]

WildFly version 8.1.0.Final

[jaikiran]

This looks more like a JBoss Marshalling thing which perhaps dmlloyd might know more of.

[dmlloyd]

I don't see anything in JBoss Marshalling which explains this behavior.  The treatment of empty and non-empty arrays is essentially the same.  I agree with Paul though, something funny is happening with the class resolver.

[pferraro]

FYI - David just committed a fix for this issue in JBoss Marshalling:

[JBMAR-172] Reserve instance cache entries in same order they are writte... · 0da3740 · jboss-remoting/jboss-marshallin…

[rhn-it-pulgupta]

Hi Paul,

Is there a way we can find out in which version of EAP this fix was released.

Regards,

Pulkit

[pferraro]

This was fixed in JBoss Marshalling 2.0.0.Beta3, which was part of wildfly-core 3.0.0.Alpha15, and thus was fixed as of EAP 7.1.0.DR10.