3:34 p.m.
hi!
I'm progressing on the implementation of the Neo4j integration for
Hibernate OGM and I think that I've bumped into a missing piece of
info when mapping associations.
Let me explain: in Neo4j (in any grap database really) it would be
very natural to map an association of 2 entities as a relationship
among 2 nodes (where each node represents an entity). This was the
path I was taking but then I've realized that I'm probably missing a
piece of info that would allow me to connect 2 nodes together.
Looking at the OGM docs and debugging I've realized that (at least
today) I've got only following pieces of info available in the
AssociationKey:
- table name for the association owning entity
- a set of column names / column values corresponding to a primary key
of the target entity.
Moreover, 'columnNames' in AssociationKey got a predefined names in
form of [o:property name]_[t:pk_field1], [o:property
name]_[t:pk_field2] etc. (where 'o' stands for the owning entity and
't' for the target one).
What I would like to (ideally) do is to map an association as a Neo4j
relationship between 2 nodes (each node corresponding to an entity)
where the relationship would be named as [o:property name].
Unfortunately I'm facing 2 issues:
1) I don't have access to EntityKey's of 2 ends of the association (or
at least I don't know how to access this info). I've noticed that
Emmanuel works on providing a valid EntityKey inside a AssociationKey
(https://github.com/emmanuelbernard/hibernate-ogm/tree/OGM-177) - I
guess it would solve half of my problems :-)
2) Naming strategy of column names in AssociationKey ([o:property
name]_[t:pk_field1]) makes it difficult to get a hand on a [property
name] - I would like to avoid splitting on '_' as it is very fragile.
I'm still quite new to Hibernate OGM so I'm not sure if I'm
approaching the problem from the right angle. I would really like to
keep mapping as natural to Neo4j / graph DB way of thinking so I would
be grateful for any feedback.
Cheers,
Pawel
8:31 a.m.
That's an interesting problem that we started to scratch in the document oriented
approach.
In the key/value pair, the underlying system has no notion of association. The table +
column(s) mapping works great.
In a Document approach, there is no (useful) notion of association either but one can
gather a set of elements under a property name. This is useful for collection of embedded
and collection of associated entities that one want to embed. But the property name under
which we would set the collection is not property rendered in the table + column(s)
approach. That's why associationKey has getCollectionRole(); I wish I had a better
approach but today that's what we've got.
In a graph approach, true associations are expressible in the system. So we both need some
kind of association property name (which can be the collection role) and also a way to
identify the entity being associated.
I have a few questions for you Pawel.
- In Neo4J, is there a notion of (optional) schema where we would not need to repeat that
we are linking from A to B in property foo?
- Do you expect the same association name when linking A to B and B to A? In other words
are associations (always) bidirectional and what's the (perf) cost of making them
bidirectional?
We will need at some point to introduce the notion of @AssociationName to separate the
object model from the association representation in document and graph datastores.
I am not thrilled at the idea of giving you the owning and (set of) target EntityKey for
each association but I don't see any better way. I think the target EntityKey is not
too hard to construct but as mentioned on OGM-177 getting the owning one is going to be
harder.
Pawel, I'd recommend you use collectionRole for your association name. Let me know if
that works even in the bidirectional case. You can also explore the idea of filling up the
EK of the target entity in AssociationKey. Let me know how hard that is. I will likely
need to dive on this one but if you can explore a bit that will help.
Emmanuel
On 28 mai 2012, at 22:34, Pawel Kozlowski wrote:
hi!
I'm progressing on the implementation of the Neo4j integration for
Hibernate OGM and I think that I've bumped into a missing piece of
info when mapping associations.
Let me explain: in Neo4j (in any grap database really) it would be
very natural to map an association of 2 entities as a relationship
among 2 nodes (where each node represents an entity). This was the
path I was taking but then I've realized that I'm probably missing a
piece of info that would allow me to connect 2 nodes together.
Looking at the OGM docs and debugging I've realized that (at least
today) I've got only following pieces of info available in the
AssociationKey:
- table name for the association owning entity
- a set of column names / column values corresponding to a primary key
of the target entity.
Moreover, 'columnNames' in AssociationKey got a predefined names in
form of [o:property name]_[t:pk_field1], [o:property
name]_[t:pk_field2] etc. (where 'o' stands for the owning entity and
't' for the target one).
What I would like to (ideally) do is to map an association as a Neo4j
relationship between 2 nodes (each node corresponding to an entity)
where the relationship would be named as [o:property name].
Unfortunately I'm facing 2 issues:
1) I don't have access to EntityKey's of 2 ends of the association (or
at least I don't know how to access this info). I've noticed that
Emmanuel works on providing a valid EntityKey inside a AssociationKey
(https://github.com/emmanuelbernard/hibernate-ogm/tree/OGM-177) - I
guess it would solve half of my problems :-)
2) Naming strategy of column names in AssociationKey ([o:property
name]_[t:pk_field1]) makes it difficult to get a hand on a [property
name] - I would like to avoid splitting on '_' as it is very fragile.
I'm still quite new to Hibernate OGM so I'm not sure if I'm
approaching the problem from the right angle. I would really like to
keep mapping as natural to Neo4j / graph DB way of thinking so I would
be grateful for any feedback.
Cheers,
Pawel
_______________________________________________
hibernate-dev mailing list
hibernate-dev(a)lists.jboss.org
https://lists.jboss.org/mailman/listinfo/hibernate-dev
12:36 p.m.
Hi Emmanuel, All,
Thnx for taking time looking into this, my comments below.
On Tue, May 29, 2012 at 3:31 PM, Emmanuel Bernard
<emmanuel(a)hibernate.org> wrote:
In a Document approach, there is no (useful) notion of association
either but one can gather a set of elements under a property name. This is useful for
collection of embedded and collection of associated entities that one want to embed. But
the property name under which we would set the collection is not property rendered in the
table + column(s) approach. That's why associationKey has getCollectionRole(); I wish
I had a better approach but today that's what we've got.
While checking under debugger I can see that the 'collectionRole'
field is not filled in AssociationKey (it is null). I can observe this
with the master branch of https://github.com/hibernate/hibernate-ogm.
Is there any other repo / branch where the work on the collectionRole
is more advanced? Or am I missing sth here?
In a graph approach, true associations are expressible in the system.
So we both need some kind of association property name (which can be the collection role)
and also a way to identify the entity being associated.
Exactly, this is my reasoning as well.
- In Neo4J, is there a notion of (optional) schema where we would not
need to repeat that we are linking from A to B in property foo?
Hmm, not sure if I'm reading you correctly here, but I'm not aware of
any such optional schema... Not sure if by A,B here you are referring
to node instances or node "classes". Because - as far as I'm aware -
in Neo4j you don't have any concept of a Node type. Node is just a
Node, it differs from other nodes only by:
- its internal id (really internal, assigned by Neo4j and not overradible)
- its properties
- its relations (associations)
- Do you expect the same association name when linking A to B and B
to A? In other words are associations (always) bidirectional and what's the (perf)
cost of making them bidirectional?
If you've got a relationships among 2 nodes such a relationship has a
direction but it is actually possible to trawerse in both directions
(there shouldn't be any performance hit regarding the direction of
traversal). In short: Neo4j stores the info regarding the relationship
direction but those associations do behave as bidirectional when it
comes to traversing.
so yes, answering your question, I would expect the same association
name when linking A to B and B to A.
We will need at some point to introduce the notion of
@AssociationName to separate the object model from the association representation in
document and graph datastores.
Hmm, ok, I'm not sure I can see imediatelly a need for the
@AssociationName. Let's assume I've got a User entity that has 2
OneToOne associations to an Address entity (say, stored in homeAddress
and workAddress fields). Given such object model I would map a user to
a Node, 2 addresses to a nodes and would have the following
relationships:
UserNode(id=user1)--homeAddress-->AddressNode(id=adress1)
UserNode(id=user1)--workAddress-->AddressNode(id=adress2)
Not sure if I'm making myself clear here, if not I can prepare a ghist
+ screen shoots from Neo4j editor to illustrate it better.
I am not thrilled at the idea of giving you the owning and (set of)
target EntityKey for each association but I don't see any better way. I think the
target EntityKey is not too hard to construct but as mentioned on OGM-177 getting the
owning one is going to be harder.
Yep, I kind of expected that you guys won't be thrilled by this as
this somehow 'breaks' the current 'table model'. At the same time I
believe that the mapping should be as natural for a given store as
possible.
Pawel, I'd recommend you use collectionRole for your association
name. Let me know if that works even in the bidirectional case.
This should work in the bidirectional case, just need to understand
why it is null in the current implementation. Do you have any idea
about what might be going on here?
You can also explore the idea of filling up the EK of the target
entity in AssociationKey. Let me know how hard that is. I will likely need to dive on this
one but if you can explore a bit that will help.
OK, sure, will look into this to do preliminary investigations.
Once again, thnx for you input so far!
Cheers,
Pawel
3:16 a.m.
> In a Document approach, there is no (useful) notion of
association either but one can gather a set of elements under a property name. This is
useful for collection of embedded and collection of associated entities that one want to
embed. But the property name under which we would set the collection is not property
rendered in the table + column(s) approach. That's why associationKey has
getCollectionRole(); I wish I had a better approach but today that's what we've
got.
While checking under debugger I can see that the 'collectionRole'
field is not filled in AssociationKey (it is null). I can observe this
with the master branch of https://github.com/hibernate/hibernate-ogm.
Is there any other repo / branch where the work on the collectionRole
is more advanced? Or am I missing sth here?
Yes that's OGM-177 :(
You can start by using @ElementCollection instead of @OneToMany. You will get the
information you are looking for in this situation.
> - In Neo4J, is there a notion of (optional) schema where we would not need to repeat
that we are linking from A to B in property foo?
Hmm, not sure if I'm reading you correctly here, but I'm not aware of
any such optional schema... Not sure if by A,B here you are referring
to node instances or node "classes". Because - as far as I'm aware -
in Neo4j you don't have any concept of a Node type. Node is just a
Node, it differs from other nodes only by:
- its internal id (really internal, assigned by Neo4j and not overradible)
- its properties
- its relations (associations)
That answers my question.
> We will need at some point to introduce the notion of @AssociationName to separate
the object model from the association representation in document and graph datastores.
Hmm, ok, I'm not sure I can see imediatelly a need for the
@AssociationName. Let's assume I've got a User entity that has 2
OneToOne associations to an Address entity (say, stored in homeAddress
and workAddress fields). Given such object model I would map a user to
a Node, 2 addresses to a nodes and would have the following
relationships:
UserNode(id=user1)--homeAddress-->AddressNode(id=adress1)
UserNode(id=user1)--workAddress-->AddressNode(id=adress2)
Not sure if I'm making myself clear here, if not I can prepare a ghist
+ screen shoots from Neo4j editor to illustrate it better.
Understood, @AssociationName would be to decorrelate the property name homeAddress from
the association name in case the object model name changes but you don't want to
change all your data in your database.
> Pawel, I'd recommend you use collectionRole for your
association name. Let me know if that works even in the bidirectional case.
This should work in the bidirectional case, just need to understand
why it is null in the current implementation. Do you have any idea
about what might be going on here?
OGM-177 basically :)
4590
days inactive
4592
days old
3 comments
2 participants
participants (2)
-
Emmanuel Bernard -
Pawel Kozlowski