Re: [Springframework-developer] WLS 7 and TX suspend/resume - continue.

[Eugene K. <eu...@pl...>]

jürgen höller [werk3AT] wrote:
> Eugene,
> 
> We have such autodetection in a variety of places, for example
> HibernateTransactionManager and JdoPersistenceManager which
> autodetect an underlying DataSource, to be able to expose Hibernate
> respectively JDO transactions as JDBC transactions. As of 1.1 final,
> those autodetection mechanisms can be turned off too, but we've
> explicitly been asked to autodetect by default.
> 
> While I do see that autodetection of the JTA TransactionManager might
> not be desirable on WebLogic 7.0, I don't think that this a big deal:
> Simply turn it off via the "autodetectTransactionManager" flag. On a
> variety servers, autodetection eases configuration, as there is no
> need for server-specific configuration in that case. Frankly, if that
> autodetection on WebLogic 7 is the biggest issue that we have to
> worry about, I'm more than happy.

   Currently autodetect in Spring's jta proxy is very confusing. 
Developer *have* to pass user transaction and proxy "autodetect" that it 
is not just user transaction has been passed. Doesn't make sencefor me.

   I do understand how it came to the current state, but it is 
fundamentally wrong and really confusing, especially for new Spring 
users or those who not much familiar with JTA.

> Regarding auto-adapting to the J2EE server version and choosing an
> appropriate lookup strategy: We don't do that in other places either,
> currently: for example, NativeJdbcExtractors for a specific
> connection pool. We do have a simple strategy that works on many
> connection pool there: SimpleNativeJdbcExtractor, fetching the native
> connection via con.getMetaData().getConnection() - but that one still
> has to be explicitly specified.
> 
> So in some respect, Spring is currently not designed to autodetect
> everything for you, but rather to make it easy for you to plug in
> appropriate strategies. You can easily factor the
> JtaTransactionManager definition out into its own XML bean definition
> file, or define a variety of combinations that are all marked as
> lazy-init, plugging a specific one in via a ${...} placeholder for
> the target reference of transaction proxies. There's a variety of
> options, actually.

   Unfortunately, currently it is not just different properties, but 
entirely different XML fragments (1 or 2 spring beans, depends from server).

> Auto-adapting to the J2EE server is not that trivial because it
> incurs a check for specific server versions: On WebLogic 8.1, use a
> default WebLogicJtaTransactionManager; on WebLogic 7.0, use a
> WebLogicJtaTransactionManager with
> WebLogicServerTransactionManagerFactoryBean. It's certainly
> *possible*, but it has to be discussed whether it's *desirable* as
> part of Spring.

   As far as I know you alredy doing version detection for Websphere, so 
I don't see any reason to generalize it and make JTA-related access the 
same for any container. This way you will really simplify usage of tx 
manager, but you'll have to take responsibility that it will work in any 
container.

> As a side note, Hibernate doesn't autodetect the J2EE server either,
> but rather requires you to explicitly specify an appropriate
> TransactionManagerLookup as Hibernate property. Kodo JDO on the other
> hand autodetects, if I'm not mistaken. Don't get me wrong: We might
> introduce such auto-adaption at a later point of time. It's just not
> that crucial, IMHO; in particular, it's no reason to defer 1.1 final.

   regards,
   Eugene


> Juergen
> 
> 
> ________________________________
> 
> From: spr...@li... on behalf
> of Eugene Kuleshov Sent: Sun 05/09/2004 16:24 To:
> spr...@li... Subject: Re:
> [Springframework-developer] WLS 7 and TX suspend/resume - continue.
> 
> 
> 
> jürgen höller [werk3AT] wrote:
> 
> 
>> I agree that it might not be following the intent of every J2EE
>> vendor out there. However, it is still not too far-fetched to
>> consider the JTA TransactionManager as a kind of extension to the
>> UserTransaction interface: If a vendor's UserTransaction object
>> implements the TransactionManager interface too, it usually should
>> be fully usable, else the vendor should just provide a facade that
>> solely implements UserTransaction...
> 
> 
> That is not my point. The point is if user is passing property A, why
> it should affect property B anyway? If he need to do something about
> B, why not to use that B at the first place (e.g. since it is all
> reflection, you can pass UserTransaction instance to
> TansactionManager setter, which will be more straightforward).
> 
> 
>> Anyway, we can't easily turn off that autodetection by default, as
>> some Spring users might rely on it. It also works nicely for a
>> number of containers, for example Resin, Orion (OC4J), JOnAS
>> (JOTM). And you can turn off autodetection now, via the
>> "autodetectTransactionManager" flag, although the only case I know
>> of where that is appropriate is the default JtaTransactionManager
>> on WebLogic.
> 
> 
> It doesn't matter where it works. It looks like ugly hack anyway. :-)
> 
> 
> 
>> Regarding full autodetection of the J2EE server: While that
>> certainly is achievable, it's not as straightforward as it may seem
>> at first glance.
> 
> 
> Why is that?
> 
> 
>> Migration between J2EE servers shouldn't be too hard anyway, as it
>> just affects the JtaTransactionManager definition: The rest of the
>> application can stay exactly the same. We currently leave it to the
>> user to configure the JtaTransactionManager appropriately, and just
>> include support classes for all sorts of cases.
> 
> 
> I didn't say it is hard, it just lead to multiple spring contexts -
> one per each application server and so far it is the only place where
> you have to do this.
> 
> regards, Eugene
> 
> 
> ________________________________
> 
>> Von: spr...@li... im
>> Auftrag von Eugene Kuleshov Gesendet: So 05.09.2004 05:06 An: 
>> spr...@li... Betreff: Re: 
>> [Springframework-developer] WLS 7 and TX suspend/resume - continue.
>> 
>> 
>> 
>> 
>> jürgen höller [werk3AT] wrote:
>> 
>> 
>> 
>>> Thanks, Thomas - it's great to have this for 1.1 final!
>>> 
>>> Did you have a chance to check whether WebLogic 8.1's
>>> UserTransaction object found at "java:comp/UserTransaction"
>>> implements the javax.transaction.TransactionManager interface
>>> too? It does so on WebLogic 7.0. The benefit of such a scenario
>>> is that Spring's JtaTransactionManager will autodetect the
>>> TransactionManager, without any special setup.
>> 
>> 
>> This autodetect feature looks like a half baked solution (or dirty
>> hack). You souldn't really rely on the fact that vendor choose to
>> implement both interfaces in publically available instance.
>> 
>> If you want to have smart tx proxy, it should autodetect j2ee
>> container vendor and version and then choose right way to retrieve
>> tx proxy. Of course it has to be tested carefuly for each
>> environment, but then users can really trust Spring's tx management
>> and can use it without struggling with configuration for each
>> environment type. That will also simplify moving application from
>> one container to another (e.g. from Weblogic to Websphere).
>> 
>> 
>> 
>>> I've documented the basic JTA setup options for popular J2EE
>>> servers in JtaTransactionManager's javadoc. It would be good to
>>> verify whether my autodetection comment just applies to WebLogic
>>> 7.0 or also to WebLogic 8.1.
>> 
>> 
>> 
>> 
>> regards, Eugene
>> 
>> 
>> 
>> 
>> 
>> 
>>> ________________________________
>>> 
>>> Von: spr...@li... im
>>> Auftrag von Thomas Risberg Gesendet: Sa 04.09.2004 17:42 An: 
>>> spr...@li... Betreff: Re: 
>>> [Springframework-developer] WLS 7 and TX suspend/resume -
>>> continue.
>>> 
>>> 
>>> 
>>> Juergen, Eugene,  Dimitry, all WLS 70 users,
>>> 
>>> We now have a new WebLogicServerTransactionManagerFactoryBean.
>>> This bean factory looks up the ServerTransactionManagerImpl to
>>> get around the NPE that was thrown during a "forceResume".  I
>>> based the code on the WebSphere variant.  This has been tested on
>>> WebLogic 7.0 SP5 and checked in.  Here is a snippet from the
>>> configuration I tested with:
>>> 
>>> <!--  transaction manager --> <bean id="wls7tm"
>>> 
>>> class="org.springframework.transaction.jta.WebLogicServerTransactionManagerFactoryBean"/>
>>>  <bean id="transactionManager"
>>> 
>>> class="org.springframework.transaction.jta.WebLogicJtaTransactionManager">
>>>  <property name="transactionManager"> <ref local="wls7tm"/>
>>> </property> </bean>
>>> 
>>> Thomas
>>> 
>>> 
>>> Thomas Risberg wrote:
>>> 
>>> 
>>> 
>>> 
>>>> Juergen,
>>>> 
>>>> I'll try to implement and test a 
>>>> WebLogicServerTransactionManagerFactoryBean today.  I'll keep
>>>> you posted.
>>>> 
>>>> 
>>>> For WLS 8.1 the TxHelper.getTransactionManager method is marked
>>>> as deprecated even though it is not documented as such in the
>>>> JavaDocs. There is also a brand new method
>>>> getClientTransactionManager for 8.1 that is starting out marked
>>>> as deprecated.  Go figure.  We are probably better off
>>>> recommending the JNDI lookups for 8.1.
>>>> 
>>>> Thomas
>>>> 
>>>> 
>>>> 
>>>> jürgen höller [werk3AT] wrote:
>>>> 
>>>> 
>>>> 
>>>> 
>>>>> Thanks, Thomas! This means that there *is* a way to make it
>>>>> work on WebLogic 7! Eugene, Dmitri, your pain finally has an
>>>>> end! :-)
>>>>> 
>>>>> Regarding obtaining the TransactionManager via the TxHelper
>>>>> on WebLogic 7: That can easily be factored out into a custom 
>>>>> FactoryBean, like we did for WebSphere (see 
>>>>> WebSphereTransactionManagerFactoryBean), to be passed into 
>>>>> JtaTransactionManager's respectively
>>>>> WebLogicJtaTransactionManager's "transactionManager"
>>>>> property. So we don't need a further JtaTransactionManager
>>>>> subclass for this, just a further TransactionManager lookup
>>>>> strategy.
>>>>> 
>>>>> BTW, Dmitri told me that the WebLogic UserTransaction object
>>>>> at "java:comp/UserTransaction" implements the
>>>>> TransactionManager interface too, but that obviously
>>>>> forceResume doesn't work there... I assume the
>>>>> ClientTransactionManagerImpl is returned there, not the 
>>>>> ServerTransactionManagerImpl.
>>>>> 
>>>>> If we can double-check that this works better than the 
>>>>> TransactionManager fetched from JNDI, I suggest to add a 
>>>>> WebLogicServerTransactionManagerFactoryBean (or a similar
>>>>> name), clearly documenting that this is meant as an
>>>>> alternative to the standard JNDI TransactionManager reference
>>>>> that WebLogic provides, mainly for WebLogic 7. I wouldn't
>>>>> mind to do this even for the 1.1 final release tomorrow...
>>>>> 
>>>>> Is the TxHelper lookup available on WebLogic 8.1 too? Of
>>>>> course, it shouldn't be necessary there, as the standard JNDI
>>>>> TransactionManager reference will work too. I just wonder
>>>>> whether it's available and would work in principle: People
>>>>> could then configure a 
>>>>> WebLogicServerTransactionManagerFactoryBean to be able to
>>>>> deploy on both WebLogic 7 and WebLogic 8, without adapting
>>>>> their Spring configuration.
>>>>> 
>>>>> Of course, such a WebLogicServerTransactionManagerFactoryBean
>>>>> would have to be coded via reflection to avoid compile-time
>>>>> dependencies on WebLogic, analogous to
>>>>> WebSphereTransactionManagerFactoryBean. Do you have the
>>>>> chance to do this before tomorrow, Thomas?
>>>>> 
>>>>> Juergen
>>>>> 
>>>>> 
>>>>> ________________________________
>>>>> 
>>>>> Von: spr...@li... im
>>>>> Auftrag von Thomas Risberg Gesendet: Sa 04.09.2004 14:30 An: 
>>>>> spr...@li... Betreff: Re: 
>>>>> [Springframework-developer] WLS 7 and TX suspend/resume -
>>>>> continue.
>>>>> 
>>>>> 
>>>>> 
>>>>> I have done some additional testing for this issue and this
>>>>> is what I have found.
>>>>> 
>>>>> WebLogic exposes a client TransactionManager implementation
>>>>> via JNDI 
>>>>> (weblogic.transaction.internal.ClientTransactionManagerImpl)
>>>>> and in WLS 7.0 it is throwing an NPE when you call
>>>>> forceResume (with or without a suspended transaction that has
>>>>> been marked for RollbackOnly).
>>>>> 
>>>>> They also have a different server implementation 
>>>>> (weblogic.transaction.internal.ServerTransactionManagerImpl)
>>>>> that is returnd by a call to
>>>>> TxHelper.getTransactionManager().  With this TM the force
>>>>> resume works as expected.
>>>>> 
>>>>> So the workaround for now is to do the suspend/resume using
>>>>> the TM returned by TxHelper before calling the Spring managed
>>>>> code. Alternatively we could provide a
>>>>> WebLogic7JtaTransactionManager that uses this method of
>>>>> obtaining the TM.
>>>>> 
>>>>> I doubt that BEA will fix this issue in WLS 7.0.  They will
>>>>> probably suggest to upgrade to 8.1 or use the TxHelper to
>>>>> obtain the TM.
>>>>> 
>>>>> Thomas
>>>>> 
>>>>> 
>>>>> jürgen höller [werk3AT] wrote:
>>>>> 
>>>>> 
>>>>> 
>>>>> 
>>>>> 
>>>>> 
>>>>>> Ideally, WebLogic should work properly with standard JTA 
>>>>>> TransactionManager resume in all cases, not needing the
>>>>>> proprietary forceResume call. However, the least we can
>>>>>> expect is that forceResume works properly: You could send a
>>>>>> corresponding bug report to BEA, for WebLogic 7.0. However,
>>>>>> they might tell you that they fixed the issue in WebLogic
>>>>>> 8.1...
>>>>>> 
>>>>>> Juergen
>>>>>> 
>>>>>> 
>>>>>> ________________________________
>>>>>> 
>>>>>> Von: spr...@li...
>>>>>> im Auftrag von jürgen höller [werk3AT] Gesendet: Fr
>>>>>> 03.09.2004 23:31 An:
>>>>>> spr...@li... Betreff:
>>>>>> Re: [Springframework-developer] WLS 7 and TX suspend/resume
>>>>>> - continue.
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> As I already wrote in my JIRA comments: Many thanks for
>>>>>> your efforts, Dmitri!
>>>>>> 
>>>>>> I assume that with JtaTransactionManager, you'll get an 
>>>>>> IllegalStateException ("cannot resume rollback-only
>>>>>> transaction" or the like), and with
>>>>>> WebLogicJtaTransactionManager, you'll get the NPE in
>>>>>> WebLogic code?
>>>>>> 
>>>>>> This really seems to be a bug in WebLogic 7.0... At least
>>>>>> it should work properly as long as not suspending a
>>>>>> transaction that has been marked as rollback-only.
>>>>>> 
>>>>>> Juergen
>>>>>> 
>>>>>> 
>>>>>> ________________________________
>>>>>> 
>>>>>> Von: spr...@li...
>>>>>> im Auftrag von Dmitri Maximovich Gesendet: Fr 03.09.2004
>>>>>> 23:07 An: spr...@li...
>>>>>> Betreff: [Springframework-developer] WLS 7 and TX
>>>>>> suspend/resume - continue.
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> I just tried WLS 7 with Spring 1.1rc2 and BMT transaction 
>>>>>> demaraction. It doesn't work with standard
>>>>>> JtaTransactionManager or with
>>>>>> WeblogicJtaTransactionManager, so something fundamentally 
>>>>>> broken in WLS7 (same NPE as described in JIRA SPR-251).
>>>>>> Next I'll try to use WLS TransactionManager (exposed in
>>>>>> JNDI) to suspend/resume transaction without Spring to see
>>>>>> it it's going to work.