Hi Martin,

I've just been reading JBoss in Action as well and was also stumped by what they said.

Like yourself I thought XA was only required for multiple data sources (or other branches in the transaction e.g JMS).

Fellow colleagues of mine insist this is what it means.

So I can't confirm 100%, but I'm sure (and so are my colleagues) the book is wrong

Regards

Dean

At the time that I wrote this text I checked this with the JBoss engineers and that is what they stated - the XA data source is required only when multiple JBoss AS instances are involved, or when you have JBoss AS and some external service all managed by a transaction manager. Two-phase commit within a single JBoss AS instance is handled by the local data source.

Martin is correct. The book unfortunately is not. Hope the JBoss engineer in question wasn't me on a bad day :-)

Jonathan Halliday

JBoss Transactions dev team lead.

I think that some of the discussion took place in the JBoss Transactions forum, but that was 3 or 4 years ago (wow, was it really that log ago when I started working on the book..) but I can't find the thread. Though there were some teleconferences - my team at the time was contemplating some possible work with the transaction service, I'm pretty sure that Mark Little was involved (or maybe I just remember that from an even earlier meeting with Arjuna), not sure who else was involved. Of course, it is very possible that I misinterpreted what was said.

But it looks like an errata update is warranted - Martin, I hope you don't mind if I plagiarize your text.

Thanks guys. Your input has been much appreciated.

I'd just also like to add, thank's to Peter for writing JBoss in Action.

There are some additional things to mention here.

JBoss AS does seem to do the so-called "Last-Resource Gambit" (term coined by Mike Spille, see http://www.jroller.com/pyrasun/category/XA), aka Last Resource Commit Optimization (see http://community.jboss.org/wiki/Multiple1PC).

This means that when dealing with multi datasources within a single JTA transaction, one of them is allowed to be a non-XA datasource. I explicitly tested this on JBoss AS 5.1 with one XA JDBC datasource and one local TX JDBC datasource, as well as with one XA JMS connection factory (the standard java:/JmsXA one) and local TX JDBC datasource.

In both these situations, JBoss AS happily processed the transaction without complaining.

As soon as you use 2 local TX JDBC datasources however, an exception like the following is thrown when a connection from the second data source is attempted to be obtained:

17:58:01,069 WARN  [loggerI18N] [com.arjuna.ats.internal.jta.transaction.arjunacore.lastResource.disallow] [com.arjuna.ats.internal.jta.transaction.arjunacore.lastResource.disallow] Adding multiple last resources is disallowed. Current resource is org.jboss.resource.connectionmanager.TxConnectionManager$LocalXAResource@5fd8aca1
17:58:01,070 INFO  [STDOUT] Interceptor.invoke(StatelessInstanceInterceptor.java:68)
at org.jboss.aop.joinpoint.MethodInvocation.invokeNext(MethodInvocation.java:102)
at org.jboss.aspects.tx.TxPolicy.invokeInOurTx(TxPolicy.java:79)
[...]

org.jboss.util.NestedSQLException: Could not enlist in transaction on entering meta-aware object!; - nested throwable: (javax.transaction.SystemException: java.lang.Throwable: Unabled to enlist resource, see the previous warnings. tx=TransactionImple < ac, BasicAction: a0000bd:c381:4cbb1bf2:ae status: ActionStatus.ABORT_ONLY >)
at org.jboss.resource.adapter.jdbc.WrapperDataSource.getConnection(WrapperDataSource.java:95)

One thing I wonder though, how efficient is the one-phase optimization in JBoss AS, especially in combination with the local-tx-datasource.

Suppose you have an application where 95% of time only a single JDBC datasource participates in a transaction and in 5% of the cases multiple resources participate.

Would it then be more efficient to use a local-tx-datasource for the resource that is used 95% of the time and rely on Last Resource Commit Optimization, or would JBoss AS be smart enough to do an efficient one-phase optimization in those cases where there is only one participant in the transaction?

If the latter is true, why is there even an local-tx-datasource and not just only an xa and no-tx datasource?

good that you've found this mistake in the book testrot! it's not a minor issue!

testrot wrote:

 

I did some tests and I am pretty sure, that JBoss uses one-phase commit optimization for transactions with only one participant. Except it is disabled by configuration.

What configuration would that be? I tried to do some searching for "one-phase" and "1pc" in the JBoss documentation, but nothing really came up.