cassandra-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Pedro Gomes <pedrogo...@lsd.di.uminho.pt>
Subject An ORM like plugin for Cassandra under Datanucleus
Date Fri, 21 May 2010 19:22:12 GMT
Hi all 

In the following weeks I have developed a plugin to the java persistence platform Datanucleus,
similar to the one presented by Google for App Engine and a Hbase already present in the platform
. Datanucleus: http://www.datanucleus.org/project/download.html

For now it allows the persistence of simples fields and one-one, one-many, and many-many relations,
supporting collections and maps, arrays are next through the Java Data Objects (JDO) API.
 
The platform allows a query language JDOQL, but I didn't add the support for it. 
You can access the project under :

	http://github.com/PedroGomes/datanucleus-cassandra 

Any feedback or doubts are welcome.

I still have, though, some doubts on the implementation. To store many-many, and one to many
relations in Cassandra, i.e., one collection/map of pointers to other objects I see two options:

-The column, in the Column family class, correspondent to the relation contains a collection
or map that codes the relation, i.e. it stores the ids to the objects,. This can be dangerous
if the collection/map grows large. 
-You have a column family that codes relations, one simple Column family per relation, or
a Super Column family that codes all relations . This approach however,  makes each fetch
from the collection takes 3 read steps, but each object id is stored separately. 

Other option that I'm thinking is a hybrid solution where the user chooses the pretended option.

If someone can give some advice on this, I would be thankful. 

Grateful
Pedro Gomes
 






Mime
View raw message