Client Connect to INADDR_ANY is a no-no.

[alrubinger]

Lately we've been bugged by discrepancies between Hudson environments running the EJB Integration TestSuite.

I found DML on #jboss this evening and asked him if a client connect to 0.0.0.0 was bad (some reading I found hinted at this) and he and Jason agree "yes, don't do it".

(09:54:26 PM) Nihility: yeah it just picks an interface
(09:54:30 PM) Nihility: like playing slots
(09:54:42 PM) dmlloyd: connecting to 0.0.0.0 *finds* the service running on only 127.0.0.1
(09:54:47 PM) dmlloyd: that's very weird
(09:55:56 PM) dmlloyd: http://twistedmatrix.com/trac/ticket/1472
(09:56:07 PM) dmlloyd: doesn't work on winders
(09:59:40 PM) dmlloyd: the little bit I can find implies that it's not a great idea

Therefore:

https://jira.jboss.org/jira/browse/EJBTHREE-1610

S,
ALR

[alrubinger]

This begs the question of what to do with a grammar like:

@RemoteBinding(clientBindUrl="socket://0.0.0.0:3873", jndiBinding="Stateful3873"),

It'd be better to have an EL available so we could do:

@RemoteBinding(clientBindUrl="socket://${jboss.bind.address}:3873", jndiBinding="Stateful3873"),

Worthy of a JIRA?

S,
ALR

[dmlloyd]

Not for that reason. Binding to 0.0.0.0 is just fine. It's connecting to 0.0.0.0 that is a strange thing to do.

Though probably you want to bind to jboss.bind.address in any case...?

[alrubinger]

The "clientBindUrl" doesn't define a binding, it tells the client where to connect (InvokerLocator URL). It's a *-jboss-beans.xml that'll make the server binding (Remoting Connector). Short answer: we agree.

S,
ALR

[dmlloyd]

Errr. Really? That's kinda.... errr.

[slimamar]

Is it the same problem :
https://jira.jboss.org/jira/browse/EJBTHREE-1650

[anil.saldhana]

Running inside JBAS, it is better to honor the -b or jboss.bind.address

[alrubinger]

I've changed the ticket to provide for EL within clientBindUrl so that connect URLs do not have to be hardcoded.

S,
ALR