[
https://issues.apache.org/activemq/browse/AMQ-2346?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=53918#action_53918
]
Gary Tully commented on AMQ-2346:
---------------------------------
David, not sure what the best plan of action is here.
I worked through the test case and the problem is that there are two sessions
and two XAResources and only one of those sessions is associated with the
active connection so only one of those sessions will behave as expected when a
transaction is rolled back.
The test holds two connections and two XAResources and alternates the
commit/rollback between them. I find that the second iteration of the test
always fails.
There is an additional problem in that the second session/connection has an
outstanding transaction when it is closed, as a result of the outstanding
transaction, the close is deferred till the transaction commits, which never
happens, thus we see the problems with "javax.jms.InvalidClientIDException:
Broker: localhost - Client: org.apache.activemq.test.JmsResourceProvider
already connected from vm://localhost#4 on subsequent tests.
My first though was that these connections should be presenting the same
XAResource (somethng with isSameRM == true), Essentially an XAResoruce for the
Broker. But it looks like the intention is that an XAResource maps to a
session. Maybe this is the crux of the problem.
When dealing with the two sessions and XAResources, my thinking was that the
test should be driving 2PC, but this does not work because the broker only sees
one transaction branch and fails on a second prepare.
So I am thinking that the correct solution is to have a shared XAResource for
the ResourceAdapter and have sessions just track the association. Not sure yet
what all the ramifications are. What do you think?
> in managed environment, 2 connections/session, UserTransaction, transaction
> management gets confused
> ----------------------------------------------------------------------------------------------------
>
> Key: AMQ-2346
> URL: https://issues.apache.org/activemq/browse/AMQ-2346
> Project: ActiveMQ
> Issue Type: Bug
> Components: JCA Container
> Affects Versions: 5.3.0
> Reporter: David Jencks
> Assignee: Gary Tully
> Fix For: 5.3.0
>
>
> In geronimo we're seeing an intermittent tck problem that seems to relate to
> a situation where there are:
> -- a BMT stateless session ejb
> -- two connections (I think these get shared to the same ManagedConnection)
> -- two sessions
> -- tx control through UT
> -- send a message in one tx
> -- receive message in a tx that is rolled back
> -- receive message in a tx that is committed
> In a public test case I'm seeing that the session proxies get confused about
> whether there is an active transaction. This doesn't happen all the time but
> with a few repititions it happens consistently.
> See GERONIMO-4784
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
