geronimo-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Thomas Mahler <thm...@web.de>
Subject Re: [persistence] Some thoughts regarding CMP and JDO
Date Sat, 09 Aug 2003 20:51:35 GMT
Hi Dain,

thanks for the quick reply and for explaining your plans.
I'd like to explain some central concepts of OJB to increase the mutual
understanding and to start a discussion how to get together.

Dain Sundstrom wrote:
> Hello Thomas (and the rest of the OJB team),
> 
> Jeremy Boynes and I (and a few others) wrote the CMP 2.0 implementation 
> in JBoss, and we have been working on the persistence code in the 
> initial Geronimo code base.
> 
> There is some code right now (a compiler and sql generator) and a fairly 
> extensive design, but it looks like we have similar designs.  The design 
> is fairly simple from the high level.  We will support several front end 
> layers simultaneously at runtime (CMP, JDO, maybe Hibernate, heck maybe 
> SQL).

Front end layers means APIs that user code uses for persistence
operations, right?
OJB currently provides support for ODMG, JDO and SODA as standardized
user APIs.

I have some doubts about Hibernate in this context.
1. Hibernate does not expose a standard API, but its own proprietary 
API. Why supporting proprietary APIs?
2. AFAIK Hibernate does not provide any SPI to plug in a user defined 
persistence service. So how do you plan to plug in your central 
persistence service into Hibernate?

>   The job of the front end layer is to handle the life-cycle and 
> callbacks required by the related specification, but all real work will 
> be delegated to a centralized persistence service.  This persistence 
> service handles caching, locking, versioning, clustering and so on.  
> When persistence service actually needs to manipulate data it delegates 
> to a store manager service.  

We have a similar design in OJB. The JDO front end layer is currently
being implemented on top of a transaction manager called OTM (Object
Transaction Manager). The OTM layer is reponsible for
management of distributed object level transactions (incl. JTA and JCA), 
lockmanagement (our lockmanagement is working accross clusters already!) 
and cache management (We provide wrappers to JCS, Tangosol Coherence, etc.).

The OTM layer itself is written on top our persistence microkernel 
called PersistenceBroker. It seems to correspond to your StoreManager. 
The PersistenceBroker is a pluggable component. That is the existing 
implementation for RDBMS persistence can be replaced at runtime by other 
implementations targeting LDAP, Filebased, XML or whatsoever.

Our ODMG implementation is currently built directly on top of the 
PersistenceBroker, but we plan to refactor it to be also based on OTM.

The resulting layering is shown in the attached images.

> The target initial store managers include 
> SQL 92, SQL 99, Oracle (which is not really SQL), file based (XML 
> maybe), and we have plans to add LDAP, clustered database layer and some 
> legacy systems. 

OJB PersistenceBroker already provides support for more than 10 of the 
most popular RDBMS platforms. I'm working on a file based solution 
(based on Prevayler), others have been playing around with LDAP.

  The following ASCI picture sums this up (if it comes
> through):
> 
>                 ---------------
> CMP ----------> |             | ------> SQL
> JDP ----------> | persistence | ------> Oracle
> Hibernate ----> | manager     | ------> LDAP
>                 |             | ------> CICS (whatever)
>                 ---------------
> 
> Now the persistence manager has a huge job, so it is broken down into 
> plugins for caching, locking and so on, which effectively  makes the 
> persistence manager just a coordinator of the plugins.

I mentioned it before: this is exactly how we do it in OJB already. We 
are using a pluggable component concept. We are thinking of enabling our 
components for Avalon and/or JMX.

> 
> Anyway, this is getting a little too technical for right now, 
> considering the initial code doesn't even have Entity beans.  From what 
> I have seen, we have a similar vision, and I think we should talk about 
> merging our efforts into a common persistence engine (maybe we can even 
> get Gavin and the Hibernate team to sync up with us).

I don't know how much of your plans are production level code already?
For OJB the answer is quite clear. It's a working and mature codebase.

It provides a robust and scalable architecture that matches the 
requirements you mention to a high degree.

So in my eyes it would save geronimo *a lot of work* to base the 
persistence engine on OJB.
Writing generic persistence engines is a tricky business. We are doing 
it with OJB for about 3 years now...

>  I think it would 
> be really positive for Java to at least have all of us at least talking 
> so our systems can play well together, but if we joined forces....  :D

I agree. I'd like to see this happen.
I believe that OJB already implements 80 - 90 % of the scope you 
describe. It would make a lot of sense to use OJB as the base 
persistence framework and to simply add the yet missing pieces like 
StoreManagers for LDAP or CICS and a CMP code generator.

So before embarking on the adventure of writing yet another generic 
persistence engine, I think we should look at the things that 
db.apache.org already offers today.
We should define the gaps and should come to a solid make or buy decision.

Dain, please get me right. I'm not trying to sell something. I simply 
want to avoid double work and to push integration of Apache projects.

We won't think about implementing our own servlet engine when we simply 
can use Tomcat, would we?

cheers,
Thomas

PS: I'm offline for the next 2 weeks, so I won't be able to reply to you 
in this time. But other OJB developers are available.

> 
> -dain
> 
> 
> On Thursday, August 7, 2003, at 07:02 AM, Mahler Thomas wrote:
> 
>> Hello all,
>>
>> I'm contacting you on behalf of the Apache OJB development team
>> (http://db.apache.org/ojb).
>> OJB is part of the Apache DB subproject and aims at providing first class
>> standards based object relational mapping technology. We are currently
>> finalizing our 1.0 release.
>>
>> Our team is excited to have a complete J2EE implementation at Apache 
>> and we
>> are willing to contribute to your project.
>>
>> OJB is heavily used in Tomcat, JBOSS and other application server
>> environments and supports JTA and JCA.
>> OJB provides special support for implementing BMP solutions easily.
>> It provides ODMG and JDO compliant APIs.
>>
>> That's why we feel that OJB is a natural choice if you are thinking 
>> about a
>> persistence engine to implement CMP (and maybe JDO). We are willing to
>> integrate all necessary changes into our codebase.
>> Who is working on persistence concepts? Whom could we contact to get
>> involved into the respective discussions?
>> If you have any questions don't hesitate to contact me or the ojb 
>> developer
>> mailing list.
>>
>> cheers and all the best for this new project,
>> Thomas Mahler
>>
>> OJB developer
>> mailto:thma@apache.org
> 
> 
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: ojb-dev-unsubscribe@db.apache.org
> For additional commands, e-mail: ojb-dev-help@db.apache.org
> 
> 


Mime
View raw message