Hi!
> Carlos Pita wrote:
> Supose I have, for example, 2 bean-managed persistence entity bean
> instances in a pool. A client create a new entity and one of them is
> used. While it is being created, the bean instance gets a data source
> and opens a connection.
Ok so far.
> In some way (perhaps storing it in another
> bean) the entity saves the connection.
Nope, don't. Why would you? In any case, this will break things.
> Then the second bean instance
> in the pool is activated; before creating a new connection it checks
> to see if there is a saved connection (in the place where the other
> instance saved it).
Beans get connections from a DataSource. Period. If you want to mess
with things, prepare to feel pain.
> The connection is already there, so the second
> bean uses it.
==pain.
> Supose then that both instances where used in different
> transactional contexts. When the first got the connection the
> corresponding XAResource was enlisted with the first transaction. When
> the second reused the connection it was already enlisted with the
> first transaction (I can't see a place where a hook to stop/start the
> XAResource could be).
In Connection.close. See, told ya, shouldn't have hold on to that
connection. Bad voodoo, man.
> The JTA specification says:
>
> "Different Java threads may be using the same connection resource
> to access the resource manager if the connection spans multiple method
> invocation [...] it's possible for different threads to be invoking
> the XAResource start and end methods [...] If the applcation server
> allows multiple threads to use a singe XAResource object and the
> associated connection to the resource manager, it is the
> responsability of the application server to ensure that there is only
> one transaction context asociated with the resource at any point of
> time"
Exactly.
> The EJB specification says:
>
> "Because this resource factories allow the Container to affect
> resource management, the connections acquired through the resource
> manager connection factory are called managed resources (these
> resource factories allow the Container to implement connection pooing
> and !!!automatic enlistment of the connection with a transaction!!!)."
Indeed. Correct.
> From this I couldn't not conclude that providing a special factory
> is the only way for a Container to asociate a XAResource with a
> transaction. But it says almost that, doesn't it?. So:
Yes, and this is what Aaron has implemented: a DataSource implementation
that does this piece of the trick.
> 1) The scenario I described above is wrong -I mean, something in it is
> not possible- (I'm new with EJB).
Yes, stick with the normal way of using connections: get it, use it,
close it.
> 2) EJB does't support "multiple threads to use a singe XAResource" so
> I should avoid the scenario.
Exactly.
> 3) The scenario is possible and EJB implementations have more tricks
> than those resource factories to reasociate (stop/start) XAResource
> with transactions.
Yes, it does start on pool.get and stop on connection.close.
> (I could be a little off-topic because the question doesn't
> directly refer to jBoss. Sorry if this is the case: don't answer me.
> Or answer me if jBoss supports multiple threads with... :) )
No problemo.
R U with me on the above?
/Rickard
--
Rickard �berg
Email: [EMAIL PROTECTED]
http://www.telkel.com
http://www.jboss.org
http://www.dreambean.com
