jena-on-cassandra

An implementation of the Jena storage layer on the Cassandra storage engine.

Create a keyspace on the Cassandra server for each DataSet or collection of graphs.

Create an instance of CassandraConnection.

Use the keyspace name and the CassandraConnection to create a DatasetGraphCassandra or GraphCassandra instance.

Once constructed they should work as any normal DatasetGraph or Graph.

Cassandra Table Design

There are 4 tables created in the Cassandra keyspace. The tables have 4 primary segments identified:

• G = Graph Name
• S = Subject
• P = Predicate
• O = Object

The tables are identified by their column order and are SPOG, PGOS, OSGP, and GSPO.

In addition each table has 3 additional columns with indexes.

• obj_dtype - text - The rdf datatype of the object.
• obj_int - int - the integer value of the object if it is an integer or can be converted to an integer.
• obj_value - text - The string representation of the object value

Query Design

When the Graph.find(), DatasetGraph.find() or DatasetGraph.findNG() is called the query engine determines which table can best answer the query given the g, s, p or o values.

Depending on the value of the o the indexes are added to the primary segments.

Cassandra queries have some particular requirements:

1. the primary key must be specified. If it is not specified then token( col ) > Long.MIN_VALUE will return all the values

2. the rest of the key columns do not have to be specified except that if a key segment has a value all earlier segments must also have values.

To handle the case where a previous key segment is missing we will stop at the first missing segment. If there are any further specified segments we will use a result filter to properly filter them.

We always remove the object value if any non primary data (obj_dtype, obj_int, obj_value) are available as we will use the other columns for the primary query, and then the indexes to locate the proper values.