Agamemnon is a thin library built on top of pycassa.
It allows you to use the Cassandra database (http://cassandra.apache.org) as a graph database. Using cassandra provides an extremely high level of reliability and scalability that is not available in other graph databases. Cassandra provides integrated support for both data partitioning as well as replication via configuration.

Much of the api was inspired by the excellent neo4j.py project (http://components.neo4j.org/neo4j.py/snapshot/), however the support in this package has diverged from that project.

Agamemnon also has integrated RDF support through RDFLib (http://www.rdflib.net/)

The following is an example of how to use Agamemnon in your own code

from agamemnon.factory import load_from_settings

First, we can decide which kind of data store we are creating. In this case we're creating an InMemory data store

config = {'backend': 'agamemnon.memory.InMemoryDataStore'}
graph_db = load_from_settings(config)

In honor of The Simpsons Movie, we'll create a node called spiderpig

spiderpig = graph_db.create_node('test_type', 'spiderpig', {'sound':'oink'})

Now we will retrieve the spiderpig from the graph and check that the attributes were correct.

>>> spiderpig = graph_db.get_node('test_type', 'spiderpig')
>>> spiderpig['sound']
'oink'

Now we will create a friend for the spiderpig (who also happens to be his alter ego). Again, let's check to confirm that the node and it's attributes were created correctly.

>>> harry_plopper = graph_db.create_node('test_type', 'Harry Plopper', {'sound':'plop'})
>>> harry_plopper = graph_db.get_node('test_type','Harry Plopper')
>>> harry_plopper['sound']
'plop'

Nodes can have different types as well. Here we create a node of type simpson, with name Homer. This node has different attributes than the previous nodes.

>>> homer = graph_db.create_node('simpson', 'Homer', {'sound':'Doh', 'job':'Safety Inspector'})
>>> homer = graph_db.get_node('simpson', 'Homer')
>>> homer['sound']
'Doh'
>>> homer['job']
'Safety Inspector'

Nodes by themselves are not very useful. Let's create a relationship between spiderpig and Harry Plopper.

>>> rel = spiderpig.friend(harry_plopper, key='spiderpig_harry_plopper_alliance', alter_ego=True, best=False)

This creates a relationship of type friend. The key has been specified in this case, although it is not necessary. If no key is supplied a uuid will be generated for the relationship.

Every node type has a "reference node". This is a metanode for the type and functions as an index for all nodes of a given type.

>>> reference_node = graph_db.get_reference_node('test_type')

Getting the instances from the test_type reference node should return the Harry Plopper node and the spiderpig node.

>>> sorted([rel.target_node.key for rel in reference_node.instance.outgoing])
['Harry Plopper', 'spiderpig']

The spiderpig should only have one friend at this point, and it should be Harry Plopper

>>> friends = [rel for rel in spiderpig.friend]

>>> len(friends)
1

>>> friends[0].target_node.key
'Harry Plopper'

Now let's confirm that Harry Plopper is friends with spider pig as well:

>>> 'spiderpig' in harry_plopper.friend
True

And, once more, make sure that spider pig is Harry Plopper's only friend:

>>> friends = [rel for rel in harry_plopper.friend]

>>> len(friends)
1

>>> friends[0].source_node.key
'spiderpig'

They should not be best friends. Let's confirm this:

>>> friends[0]['best']
False

Homer is spiderpig's best friend:

>>> rel = homer.friend(spiderpig, best=True, alter_ego=False, type='love', strength=100)

Here we added additional attributes to the relationship.

Now spiderpig should have 2 friends.

>>> friends = [rel for rel in spiderpig.friend]
>>> len(friends)
2

You can get a list of all of the relationships of a particular type between a node and other nodes with a particular key

>>> homer_spiderpig_love = spiderpig.friend.relationships_with('Homer')
>>> len(homer_spiderpig_love)
1

>>> homer_spiderpig_love = spiderpig.friend.relationships_with('Homer')
>>> print homer_spiderpig_love[0]['strength']
100

This project is an extension of the globusonline.org project and is being used to power the upcoming version of globusonline.org. I'd like to thank Ian Foster and Steve Tuecke for leading that project, and all of the members of the cloud services team for participating in this effort, especially: Vijay Anand, Kyle Chard, Martin Feller and Mike Russell for helping with design and testing. I'd also like to thank Bryce Allen for his help with some of the python learning curve.