On 13/06/12 12:53am, Schönfisch, Jörg wrote:
Thanks for your detailed answer!
I think what I'm trying to achieve is a bit more difficult than this. I'm 
already using polymorphic joins and it works really well. However, in the 
scenario I try to model here I don't want to use single table inheritance for 
performance reasons.

What performance problem do you see? We have tables with several hundred 
thousand records managed in this way without the slightest issue. And Cayenne 
maps queries to proper SQL with joins which the db optimises nicely.


I'm implementing some kind of history for my entities. If a relation between 
two entities is changed, the old relation is written to a history table and 
then gets replaced by the new one. The historized relation still references the 
two entities in the normal tables. If now one of these entities is deleted, it 
is also put in a separate history table. The historized relation from before 
now references one normal entity and one historized entity. So what is 
referenced by the relation changed without this information being explicitly 
available anywhere.
The IDs for all these entities and relations are permanent, so the normal and 
the corresponding historized entities always have the same ID and only differ 
by their revision (and the changed values, of course). The normal and the 
history tables should be split to improve query performance if the information 
from the history is not needed.

That just made my head hurt. I can see what you are doing... but it is 
complicated. Are you sure you need to save this to a new table? Will it really 
be a performance problem or are you just assuming it will be? More records in a 
table need not slow things down if the indexes work well for you.

It sounds like you have two parallel schemas. One for current data and another 
for all the history.


It might be possible to model this as vertical inheritance with NormalEntity 
and HistorizedEntity being subclasses to a common superclass. But this would 
complicate my model and introduce additional tables which is what I am trying 
to avoid.

But above you said that "it is also put in a separate history table". I'm 
unclear whether you have separate database tables or not. At any rate, whether you have 
separate database tables, it sounds like a Java subclass approach will give you some 
benefits.

Ari




So I want to select only the IDs from the relationship table and then do two 
additional queries for the normal and the historized entities identified by 
these. But I have not found a possible to model this.


Cheers,
Joerg


-----Ursprüngliche Nachricht-----
Von: Aristedes Maniatis [mailto:[email protected]]
Gesendet: Dienstag, 12. Juni 2012 15:53
An: [email protected]
Betreff: Re: Mapping a relationship with type List<Long>

On 12/06/12 7:01pm, Schönfisch, Jörg wrote:
Hi,

I have two tables, one relates to an object entity, the other is a join table. A FK 
relationship exists between those two. However the "other" IDs in the join 
table may be present in several different tables so I cannot define a FK relationship for 
them. Thus I directly want to read a list of IDs and not convert them into object 
entities. The mapping of IDs to objects will later take place in the business code.
Any hints on how to do this or if it's possible at all?

What you are describing is a polymorphic join. In Rails it looks like this:

     
http://guides.rubyonrails.org/association_basics.html#polymorphic-associations

There is no built-in construction in Cayenne, but it isn't hard to map. Let's pretend you 
want to have the ability to add "attachments" to various other entities. 
Contacts, Artists, Paintings. These attachments might be a picture or other information.

So you create:

[Attachment]
id
other columns

[AttachmentRelation]
foreignRecordId (int)
foreignTable (string or enum)

[ContactAttachmentRelation]
[PaintingAttachmentRelation]
[ArtistAttachmentRelation]

These last three classes you create in Cayenne model and you make them a 
subclass of AttachmentRelation. There is no database table that goes with them, 
so effectively you are creating them as single table inheritance. Three 
subclasses, and only table table in the database.

Put a method like this in the AttachmentRelation:

     public abstract void setAttachedRelation(Attachable attachable);


And then code like this goes in the ContactAttachmentRelation subclass:

    public void setAttachedRelation(Attachable attachable) {
      setAttachedContact((Contact) attachable);
    }

The model will have a qualifier like this:

    <obj-entity name="ContactAttachmentRelation" superEntityName="AttachmentRelation" 
className="x.y.z.ContactAttachmentRelation" clientClassName="x.y.z.ContactAttachmentRelation">
      <qualifier><![CDATA[entityIdentifier like "Contact"]]></qualifier>
    </obj-entity>



There is a bit more to do, but it is pretty simple. The main trick is to use 
the Cayenne inheritance and let it handle all your SQL generation. It works 
really nicely and we've had something like this in production for many years.



Ari




--
-------------------------->
Aristedes Maniatis
GPG fingerprint CBFB 84B4 738D 4E87 5E5C  5EFA EF6A 7D2E 3E49 102A


Reply via email to