Cayenne handles correct ordering of operations automatically, based
on dependencies derived from relationships.The algorithm has a few
limitations though. It can't handle cycles (when Entity A depends on
Entity B, but also Entity B depends on Entity A). This probably also
includes entities that have relationships to the same entity (I
assume this is the case the original post was referring to).
There are few solutions:
1. (a workaround, rather than a solution) Do commit in two steps.
2. Define FK constraints in question as DEFERRABLE and INITIALLY
DEFERRED (supported by Postgres 8.*)
3. Set a custom org.objectstyle.cayenne.map.EntitySorter on the
DataNode.
Andrus
On May 20, 2006, at 3:00 AM, Marcin Skladaniec wrote:
Hm. Strange. I do really complex commits, sometimes 7 or more
related records (I mean 7 levels of relationship, not seven
entities), related by many-to-many many-to-one relationships and
never get those problems. And it doesn't matter if the records are
new or old. Could you describe how you are creating objects and how
do you commit them ?
Regards
Marcin
On 20/05/2006, at 4:31 PM, Tomi NA wrote:
On 5/20/06, Jeff de Vries <[EMAIL PROTECTED]> wrote:
I don't know if it is related or not, but I've also had problems
in the
past when I try to create a new parent and several child objects
related
to that parent all at once and then try to commit. The problem
looks
like Cayenne is INSERTing the child records into the database first,
before the parent record, and the database complains that the
children
have an invalid foreign key (and, yes, I have the ON UPDATE and ON
DELETE rules for the foreign key set to DO NOTHING and I still
get the
error from PostgreSQL). To get around it I just committed the
parent
first, then committed all the children.
I had the same problem, IIRC: I was very surprised that cayenne
couldn't handle such a commit, although truth be told, I can't
imagine
everything that's going on under the hood of the operation that would
make implementing this feature difficult.
I would certainly love to see this fixed (if at all possible) as I
wasn't to happy to have to commit in the middle of what had to be an
atomic transaction. It'd also make the framework a lot more flexible,
e.g. enabling the user to have long inter-commit sessions with
complex
data updates, if the user so desires.
t.n.a.
