Re: TransactionManagerFactory and WAS (was: [VOTE] publish openjpa 0.9.5-incubating podling release)
Kevin- On Oct 18, 2006, at 2:43 PM, Kevin Sutter wrote: -0 Although it looks like you already have the three +1 votes to publish the 0.9.5 release, I'm hesitant with this publish since the current OpenJPA implementation is using internal WebSphere methods. I knew about the problem of not using the TransactionSynchronizationRegistry interface (OPENJPA-61), but I didn't realize the implications of using internal WebSphere methods to get around this issue. Specifically, OpenJPA is using the following method: com.ibm.ws.Transaction.TransactionManagerFactory.getTransactionManager Is this a problem because you would like to see OpenJPA using more modern methods of getting at the TM, or because there are other serious consequences to calling this method? Note that Kodo has been using this method fine for years, and it looks like a number of other popular frameworks (Spring, Castor, and Hibernate, after some quick Googling on the method name) also use this method to get the WAS TM, so it doesn't seem uncommon. I would like to see this get resolved before we publish the 0.9.5 release. The OPENJPA-61 report has two aspects to it. One is to use the new JTA interface for Java EE 5 compliant environments. That's one problem. But, the other, more immediate, problem is to remove the usage of internal WebSphere methods for existing WebSphere environments. We will attempt resolve this immediate problem first. And, then worry about the TransactionSynchronizationRegistry. If it is just a matter of using a more modern method to get the same TM functionality, then we can pretty quickly implement this by adding a new WASManagedRuntime that gets the TM in whatever way we want. However, if it doesn't work with older WAS versions, we should keep the old method around as well, since otherwise people won't have any way of integrating with this version. Ideally, of course, everything would move towards the TransactionSynchronizationRegistry, but as I mentioned in OPENJPA-61, we currently rely internally on having a javax.transaction.Transaction instance for both managed and stand- along transactions. The quickest route to getting this to work would be to make some TransactionSynchronizationRegistryTransactionManagerFacade that returns a TransactionFacade implementation of javax.transaction.Transaction whose begin()/commit() methods are just no-ops or throw exceptions (since I don't think the Broker should ever be calling those methods when the openjpa.TransactionMode is set to managed). However, I haven't yet experimented with a container that supports the TransactionSynchronizationRegistry, so that implementation work would be best done by someone who has experience with one of those containers (hint :) Anyway, in conclusion, I'm happy to see an update to the methods through which the transaction integration is performed provided we don't break backwards compatibility with older versions. I'd also rather not hold up 0.9.5 just for this ... we can always cut a new release pretty quickly once we get the updated integration working and tested, but in the near term, it'd be nice to get something out there beyond the nightly snapshots that people can start relying on. Thanks, Kevin On 10/18/06, Abe White [EMAIL PROTECTED] wrote: +1 _ __ Notice: This email message, together with any attachments, may contain information of BEA Systems, Inc., its subsidiaries and affiliated entities, that may be confidential, proprietary, copyrighted and/or legally privileged, and is intended solely for the use of the individual or entity named in this message. If you are not the intended recipient, and have received this message in error, please immediately return this by email and then delete it.
Re: TransactionManagerFactory and WAS (was: [VOTE] publish openjpa 0.9.5-incubating podling release)
I think that there are a few issues being discussed here. 1. Best practice for OpenJPA when running with servers/containers implementing the TransactionSynchronizationRegistry is to use this interface to register for transaction events of interest. There is a well-known JNDI name for the instance that implements the interface when running in a standard Java EE 5 environment. So this should be done in preference to other proprietary mechanisms. But this interface is not guaranteed to be available if running in older servers. In particular, AFAIK, Websphere does not yet ship a product that implements this interface. So, 2. For pre-TSR servers, what do we do? Best practice here is to use the proprietary mechanisms of the servers, but do so in a way that does not require the implementation to be dependent on the proprietary interfaces unless running in that environment. 3. I don't see an issue using proprietary (public, documented) interfaces as an alternative to TSR. This is a time-honored way to get what you need given what is available. I've just looked at the implementation of *ManagedRuntime in the org.apache.openjpa.ee package and found best practice 2 to be implemented. Nice job, Marc (attribution in code). Using TSR will require a non-trivial change to the classes in the package. In particular, there will need to be a definition of an interface that performs the required registrations without a real implementation of TransactionManager. I don't believe that this is a reason to hold up the release. I hope we will be supporting pre-Java EE 5 servers for several years to come. Craig On Oct 18, 2006, at 3:39 PM, Marc Prud'hommeaux wrote: Kevin- On Oct 18, 2006, at 2:43 PM, Kevin Sutter wrote: -0 Although it looks like you already have the three +1 votes to publish the 0.9.5 release, I'm hesitant with this publish since the current OpenJPA implementation is using internal WebSphere methods. I knew about the problem of not using the TransactionSynchronizationRegistry interface (OPENJPA-61), but I didn't realize the implications of using internal WebSphere methods to get around this issue. Specifically, OpenJPA is using the following method: com.ibm.ws.Transaction.TransactionManagerFactory.getTransactionManage r Is this a problem because you would like to see OpenJPA using more modern methods of getting at the TM, or because there are other serious consequences to calling this method? Note that Kodo has been using this method fine for years, and it looks like a number of other popular frameworks (Spring, Castor, and Hibernate, after some quick Googling on the method name) also use this method to get the WAS TM, so it doesn't seem uncommon. I would like to see this get resolved before we publish the 0.9.5 release. The OPENJPA-61 report has two aspects to it. One is to use the new JTA interface for Java EE 5 compliant environments. That's one problem. But, the other, more immediate, problem is to remove the usage of internal WebSphere methods for existing WebSphere environments. We will attempt resolve this immediate problem first. And, then worry about the TransactionSynchronizationRegistry. If it is just a matter of using a more modern method to get the same TM functionality, then we can pretty quickly implement this by adding a new WASManagedRuntime that gets the TM in whatever way we want. However, if it doesn't work with older WAS versions, we should keep the old method around as well, since otherwise people won't have any way of integrating with this version. Ideally, of course, everything would move towards the TransactionSynchronizationRegistry, but as I mentioned in OPENJPA-61, we currently rely internally on having a javax.transaction.Transaction instance for both managed and stand- along transactions. The quickest route to getting this to work would be to make some TransactionSynchronizationRegistryTransactionManagerFacade that returns a TransactionFacade implementation of javax.transaction.Transaction whose begin()/commit() methods are just no-ops or throw exceptions (since I don't think the Broker should ever be calling those methods when the openjpa.TransactionMode is set to managed). However, I haven't yet experimented with a container that supports the TransactionSynchronizationRegistry, so that implementation work would be best done by someone who has experience with one of those containers (hint :) Anyway, in conclusion, I'm happy to see an update to the methods through which the transaction integration is performed provided we don't break backwards compatibility with older versions. I'd also rather not hold up 0.9.5 just for this ... we can always cut a new release pretty quickly once we get the updated integration working and tested, but in the near term, it'd be nice to get something out there beyond the nightly snapshots that people
Re: TransactionManagerFactory and WAS (was: [VOTE] publish openjpa 0.9.5-incubating podling release)
On 10/18/06, Craig L Russell [EMAIL PROTECTED] wrote: I think that there are a few issues being discussed here. 1. Best practice for OpenJPA when running with servers/containers implementing the TransactionSynchronizationRegistry is to use this interface to register for transaction events of interest. There is a well-known JNDI name for the instance that implements the interface when running in a standard Java EE 5 environment. So this should be done in preference to other proprietary mechanisms. But this interface is not guaranteed to be available if running in older servers. In particular, AFAIK, Websphere does not yet ship a product that implements this interface. So, Agree. TransactionSynchronizationRegistry is the preferred method for Java EE 5 environments. 2. For pre-TSR servers, what do we do? Best practice here is to use the proprietary mechanisms of the servers, but do so in a way that does not require the implementation to be dependent on the proprietary interfaces unless running in that environment. Agree, as long as the interfaces being used are public and documented (as your next bullet outlines). 3. I don't see an issue using proprietary (public, documented) interfaces as an alternative to TSR. This is a time-honored way to get what you need given what is available. Agree again. I've just looked at the implementation of *ManagedRuntime in the org.apache.openjpa.ee package and found best practice 2 to be implemented. Nice job, Marc (attribution in code). Yes, the implementation provided is flexible enough for the various application servers. The problem is with WebSphere's access to the TransactionManager. For reasons that do not directly pertain to this discussion, access to the WebSphere TM was not made available via public, documented interfaces until WAS 5.1. Previous to this, the undocumented, internal TransactionManagerFactory interfaces were used by several vendors. Since these internal interfaces can change and have changed in the past, WAS eventually provided the ExtendedJTATransaction interface. Using TSR will require a non-trivial change to the classes in the package. In particular, there will need to be a definition of an interface that performs the required registrations without a real implementation of TransactionManager. I don't believe that this is a reason to hold up the release. I hope we will be supporting pre-Java EE 5 servers for several years to come. Agree. That's why I voted neutral. I didn't want to hold up the release. Just wanted to voice my concern with the use of these internal interfaces. Mike and I are working on an update. Thanks. Craig On Oct 18, 2006, at 3:39 PM, Marc Prud'hommeaux wrote: Kevin- On Oct 18, 2006, at 2:43 PM, Kevin Sutter wrote: -0 Although it looks like you already have the three +1 votes to publish the 0.9.5 release, I'm hesitant with this publish since the current OpenJPA implementation is using internal WebSphere methods. I knew about the problem of not using the TransactionSynchronizationRegistry interface (OPENJPA-61), but I didn't realize the implications of using internal WebSphere methods to get around this issue. Specifically, OpenJPA is using the following method: com.ibm.ws.Transaction.TransactionManagerFactory.getTransactionManage r Is this a problem because you would like to see OpenJPA using more modern methods of getting at the TM, or because there are other serious consequences to calling this method? Note that Kodo has been using this method fine for years, and it looks like a number of other popular frameworks (Spring, Castor, and Hibernate, after some quick Googling on the method name) also use this method to get the WAS TM, so it doesn't seem uncommon. I would like to see this get resolved before we publish the 0.9.5 release. The OPENJPA-61 report has two aspects to it. One is to use the new JTA interface for Java EE 5 compliant environments. That's one problem. But, the other, more immediate, problem is to remove the usage of internal WebSphere methods for existing WebSphere environments. We will attempt resolve this immediate problem first. And, then worry about the TransactionSynchronizationRegistry. If it is just a matter of using a more modern method to get the same TM functionality, then we can pretty quickly implement this by adding a new WASManagedRuntime that gets the TM in whatever way we want. However, if it doesn't work with older WAS versions, we should keep the old method around as well, since otherwise people won't have any way of integrating with this version. Ideally, of course, everything would move towards the TransactionSynchronizationRegistry, but as I mentioned in OPENJPA-61, we currently rely internally on having a javax.transaction.Transaction instance for both managed and stand- along transactions. The quickest route to getting this to work would be to make some
