Integration issue with the ejb caching

[starksm64]

Brian raised the following issue about differences in behavior between the legacy in memory tm and JBossTS. I'm posting here in case not everyone watches all forums.

http://www.jboss.com/index.html?module=bb&op=viewtopic&p=3986745

[kconner]

This has been discussed previously with Bela and Manik, they should be using an XAResource to handle this.

There was a discussion on one of the forums but I can't find it at the moment, perhaps Bela or Manik can point to it.

[marklittle]

Thanks. Just posted something there.

[kconner]

ditto :-)

[marklittle]

Yeah, but mine got in before yours so I win this round ;-) TOO SLOW :-)

[kconner]

lol, I was distracted for a minute :-)

[brian.stansberry]

LOL. I never thought for a second that fixing this would involve changing the synchronization handling behavior of JBossTS, although my post didn't say that. But it's why I posted on the EJB3 forum and not here. :-)

JBoss Cache and Hibernate need to integrate properly with any transaction manager; relying on the implementation details of one or two obviously doesn't cut it.

[marklittle]

No problem. JBossTS does support ordering of invocations if you want, but it's not part of the specs so you'd be no better off than you are now really, at least as far as non-JBoss TMs are concerned.

[brian.stansberry]

Is this some kind of a externally configurable switch, or does it require coding to a JBossTS API?

[kconner]

I would also be interested in this answer as I believe the sequence is hashed. :-)

How do you specify the order?

[marklittle]

It's a JBossTS specific API. You'd need to user AbstractRecords and either use one of the pre-defined types or add your own type. Full details are in the Core Programmers Guide.

However, I WOULD NOT RECOMMEND doing it this way, if portability is something you want to achieve.

[brian.stansberry]

No, we definitely wouldn't code to a JBossTS specific API. If it was something externally configurable, I would have considered using it short term in test fixtures, etc., so I could test other aspects of EJB3/JBC integration without this issue getting in the way.

Thanks for the info.