[jira] Updated: (AMQ-1760) Client internal exceptions occurring during async message processing should not be propagated to Connection's ExceptionListener

["Kai Hudalla (JIRA)" <ji...@apache.org>]

     [ https://issues.apache.org/activemq/browse/AMQ-1760?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Kai Hudalla updated AMQ-1760:
-----------------------------

    Attachment: ExceptionListener.patch

Patch adding the functionality discussed in the issue to the ActiveMQConnection and ActiveMQSession

> Client internal exceptions occurring during async message processing should not be propagated to Connection's ExceptionListener
> -------------------------------------------------------------------------------------------------------------------------------
>
>                 Key: AMQ-1760
>                 URL: https://issues.apache.org/activemq/browse/AMQ-1760
>             Project: ActiveMQ
>          Issue Type: Improvement
>    Affects Versions: 5.1.0
>            Reporter: Kai Hudalla
>            Priority: Minor
>         Attachments: ExceptionListener.patch
>
>
> I have stumbled across the following code in the ActiveMQSession.run() method:
>            try {
>                 messageListener.onMessage(message);
>             } catch (Throwable e) {
>                 // TODO: figure out proper way to handle error.
>                 LOG.error("error dispatching message: ", e);
>                 connection.onAsyncException(e);
>             }
> As the TODO points out, the way how to handle exceptions properly needs some discussion. I have come across this while working on the Resource Adapter in conjunction with GlassFish v2. When I use batching when delivering messages to an MDB and the MDB marks the current TX as rollback only (e.g. if it cannot access a database), any subsequent invocation of the MDB's onMessage() method (actually the invocation of the wrapper around that method provided by the GlassFish app server) in the same TX will throw a javax.ejb.TransactionRolledbackLocalException in order to indicate that it is futile to invoke the bean since the TX will be rolled back anyway. This RuntimeException will now lead to the catch block being executed, i.e. the connection's ExceptionListener will be notified eventually which in this case is the listener that the RA has registered on the connection which in turn will tear down the connection and reconnect to the broker. However, the connection never failed in the first place, i.e. reconnecting to the broker is not necessary at all.
> After some discussion with Rob Davies we are proposing to 
> add another method on Connection - e.g.  
> setClientInternalExceptionListener() - that registers an exception  
> listener of internal exceptions only - like in this  
> ActiveMQSession.run() case.
> We then only pass exceptions to the ExceptionListener registered by  
> setExceptionListener() that directly affect the connection. 
> The ActiveMQSession.run() method will then notify the ClientInternalExceptionListener of the runtime problem within the container.
> I have attached a proposed patch that adds this functionality to ActiveMQConnection and ActiveMQSession.
> Kai

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.