[ODMG] Making graphs with M:N relations persistent

[Bart Molenkamp <b....@bizzdesign.nl>]

Hi,

I've read in the release-notes.txt file that persisting graphs with M:N
relations is not possible and is a known issue. It works when I create
objects without the relations, store them using the persistence broker,
then add the relations and lock them, so that the ODMG transaction
stores the objects with the relations.

That work-around is fine for me, except that an ODMG transaction is not
the same thing as a PB (database) transaction. Using the PB's store()
method stores the object directly in the database (which is not what I
want, I want it to be stored when ODMG transaction's commit() is
called).

Two questions:
1. How difficult is it to fix this m:n issue in ODMG? Where should I
start looking? I'll be happy if I can fix it and donate a patch!
2. If it is difficult (which it probably is ;), what would be the best
strategy to intercept in an ODMG transaction's commit() and then use the
persistence broker then as workaround (so that my whole graph of objects
is stored inside a single database transaction)?

Thanks,
Bart.

---------------------------------------------------------------------
To unsubscribe, e-mail: ojb-dev-unsubscribe@db.apache.org
For additional commands, e-mail: ojb-dev-help@db.apache.org

Re: [ODMG] Making graphs with M:N relations persistent

[Armin Waibel <ar...@apache.org>]

Hi Bart,

Bart Molenkamp wrote:
> Hi,
> 
> I've read in the release-notes.txt file that persisting graphs with M:N
> relations is not possible and is a known issue. It works when I create
> objects without the relations, store them using the persistence broker,
> then add the relations and lock them, so that the ODMG transaction
> stores the objects with the relations.
> 
> That work-around is fine for me, except that an ODMG transaction is not
> the same thing as a PB (database) transaction. Using the PB's store()
> method stores the object directly in the database (which is not what I
> want, I want it to be stored when ODMG transaction's commit() is
> called).
> 
> Two questions:
> 1. How difficult is it to fix this m:n issue in ODMG?

It will be tricky, because the odmg-implementation can't use the 
auto-xxx feature of the PB-api to handle m:n relations, thus we have to 
deal with the indirection table write/delete queries on odmg-level.
On insert first store n and m side, then write to indirection table. On 
delete vice versa,....

> Where should I
> start looking? I'll be happy if I can fix it and donate a patch!

Start looking in ObjectEnvelopeTable/ObjectEnvelope. I will start 
refactoring/fixing of known issues in odmg-api ASAP (hope to check in a 
first version in next two or three weeks). Any hints and help will be 
welcome.


> 2. If it is difficult (which it probably is ;), what would be the best
> strategy to intercept in an ODMG transaction's commit() and then use the
> persistence broker then as workaround (so that my whole graph of objects
> is stored inside a single database transaction)?
> 

This shouldn't be a problem, try

// cast to OJB extensions
TransactionExt tx = (TransactionExt)...get Transaction
tx.beginTransaction();
... do your odmg work

PersistenceBroker pb = tx.getBroker();
// do insert m:n relations using PB-api
// !!! don't close the used PB!!!

...
tx.commit()

good luck!

regards,
Armin

> Thanks,
> Bart.
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: ojb-dev-unsubscribe@db.apache.org
> For additional commands, e-mail: ojb-dev-help@db.apache.org
> 
> 
> 

---------------------------------------------------------------------
To unsubscribe, e-mail: ojb-dev-unsubscribe@db.apache.org
For additional commands, e-mail: ojb-dev-help@db.apache.org