This thread is created when transaction subsystem starts and is active through the life cycle of the application server. If failure happens in any of the applications, periodic recovery finds the name of the compensation handler in tx-object-store. That handler is a CDI bean implemented by the application and has to be invoked in order to redo the work of the application. We use BeanManager to instantiate that handler and currently use JNDI to get it. Also, when compensate method is invoked it is very likely that implementation of it will need access to JNDI as well to get UserTransaction for example.

Additionally, before invoking the compensation handler we need to enable application's class loader (I still need to figure out the best way to store them, but I'm thinking of providing them to periodic recovery during the deployment of each application). Would that class loader help to solve JNDI issue or is there extra context I should save maybe?

Gytis

On 23/06/15 16:34, Stuart Douglas wrote:
I don't know if ManagedThreadFactory is a great option, as then you have a hard dependency on this being available,  which may not be the case in some cut down configurations. 

Does this thread run in the context of a particular deployment, otherwise this does not make a great deal of sense, as the JNDI context is a per deployment construct. 

If this is a per deployment thing then you can manually set up the per deployment contexts using the actions under org.jboss.as.server.deployment.Attachments#SETUP_ACTIONS.

Stuart

On Tue, 23 Jun 2015 at 16:21 Gytis Trikleris <gytis@redhat.com> wrote:
Hello,

I’ve got an issue with my subsystem (https://github.com/wildfly/wildfly/tree/master/transactions) where I cannot access BeanManager or UserTransaction from JNDI (I can however lookup datasources). This error is currently happening during periodic recovery i.e. on PeriodicRecovery thread. My assumption is that the CDI and EE JNDI environment are not initialized because PeriodicRecovery extends Thread and was created with the “new” operator rather than ManagedThreadFactory and I know that EE apps are meant to use this approach to ensure their environments are initialized correctly - does this restriction apply to subsystems too? One issue I am coming up against during prototyping is that I’m not sure how to get the ManagedThreadFactory during our subsystems boot time as it does not appear in JNDI.

I would like to know if it is possible to inject the ManagedThreadFactory into my subsystem so I am not reliant upon its availability in JNDI.

Thanks,
Gytis
_______________________________________________
wildfly-dev mailing list
wildfly-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/wildfly-dev