[OJB-developers] WG: [Fwd: Re: Re: JBOSS/OJB code]

[Mahler T. <tho...@it...>]

> -----Urspr=FCngliche Nachricht-----
> Von: Thomas Mahler [mailto:tho...@ho...]
> Gesendet: Dienstag, 19. M=E4rz 2002 07:07
> An: tom
> Betreff: [Fwd: Re: Re: JBOSS/OJB code]
>=20
>=20
>=20
>=20
> -------- Original Message --------
> Subject: Re: Re: JBOSS/OJB code
> Date: Mon, Mar 18 2002 23:37:00 GMT+0100
> From: jl...@pi... (Jesper Ladegaard)
> To: Thomas Mahler <tho...@ho...>
>=20
> Hi again Thomas
>=20
> [snip]
>  > There is one issue I found: The J2EE spec tells
>  > how a client can obtain a user Transaction (by
>  > EJBContext.getUserTransaction()).
>  >
>  > But I did not find any specification how a "JTA
>  > provider" like OJB can obtain a TransactionManager
>  > instance to obtain external transactions.
>  >
>  > IMO we will need this in the PBF to associate
>  > new PB instances with external transactions.
>=20
> This is where JCA comes into the picture. The EJB server will=20
> attach/detach tx-context to connection/broker resource on=20
> demand. It's=20
> important there's no fixed 1-1 relationship between transaction and=20
> broker. For example, it must be possible for any broker in=20
> the pool to=20
> commit any active transaction. The transaction manager of the=20
> EJB server=20
> only guarantees that only one transaction is active on a single=20
> connection at any time.
>=20
>  >
>  > Armin uses a JNDI lookup to obtain a TransactionManager
>  > instance. But we are not sure if there is an "official"
>  > way to do this.
>=20
> It's possible to obtain the TransactionManager and
> register a javax.transaction.Synchronization with the current=20
> transaction. However, this is mostly used for non-XA-aware programs=20
> (i.e. client/user code) that are interrested in a callback when the=20
> transaction completes.
>=20
> The official way is to enlist a XAResource with the transaction.=20
> However, supporting the JCA connection contract and the connection=20
> manager of the application server will take care of enlisting the=20
> XAResouce for us, which is the official way of doing this in J2EE =
1.3.
>=20
>  > If there is no standardized way for this it will be
>  > difficult to integrate OJB into different EJB
>  > Containers.
>  >
>=20
> Exactly! And this is the very reason we must support JCA, which was=20
> created to solve this exact problem, plus some others also.
>=20
> [snip...]
>=20
>  > I think we can learn a lot from the stuff the
>  > CASTOR and Tyrex developers did in this area.
>=20
> What did they do?
>=20
> I've looked at CASTOR's code once, however it was a long time=20
> ago. Their=20
> code was not nice then, but I don't know how their transaction stuff=20
> works today?
>=20
>  > Providing JUnit test cases is definitely a must.
>  > But it will also very difficult to support the
>  > large variety of scenarios you mention.
>=20
> yes, and I don't think we'll get everything right the first time.=20
> However, from looking at JBoss code and their JDBC and JMS adapter, I =

> can see they don't support many of the senarios either.
>=20
>  >
>  > You will need a comple test lab to do this job properly !
>=20
> I hope not :-)
>=20
>  > I found that it is even difficult to provide support
>  > for the most popular RDBMS.
>  > I'm running Junit tests against HSQLDB, MS-Access,
>  > Oracle and DB2.
>  > But I don't have an Informix DB and I don't have
>  > any MySQL Know-How.
>  > You'll definitely need a team of test engineers,
>  > a server farm and several licences for RDBMS
>  > products to do this job professionally.
>=20
> I can imagine that.
>=20
>=20
>  > I think we'll need a lot of creativity to setup
>  > testscenarios that give us feedback on the quality
>  > of the JTA/JCA Integration!
>=20
> Really!
> I don't think this is such a big problem.
>=20
> IMO, the biggest task is for us to create a robust design for=20
> integrating standard-based transaction support into OJB.
>=20
> For example, some people uses the "OBJ mode" and don't care=20
> about ODMG.=20
> Others (like me) want's to use ODMG. The tx stuff should=20
> intergrate and=20
> work the same for both methods. I mean, there seems to be a lot of=20
> confusion here. For example both JBOSS integrations you've=20
> send me only=20
> works in OJB mode.
>=20
> Regards
> Jesper
>=20
>=20
>=20
>=20
>=20