Python networkx and persistence (perhaps in neo4j)

Question

I have an application that creates many thousands of graphs in memory per second. I wish to find a way to persist these for subsequent querying. They aren't particularly large (perhaps max ~1k nodes).

I need to be able to store the entire graph object including node attributes and edge attributes. I then need to be able to search for graphs within specific time windows based on a time attribute in a node.

Is there a simple way to coerce this data into neo4j ? I've yet to find any examples of this. Though I have found several python libs including an embedded neo4j and a rest client.

Is the common approach to manually traverse the graph and store it in that manner?

Are there any better persistence alternatives?

Answer 1

Networkx has several serialization method.

In your case, I would choose graphml serialization :

http://networkx.github.io/documentation/latest/reference/readwrite.graphml.html

It's quite simple to use :

import networkx as nx
nx.write_graphml('/path/to/file')

To load it in Neo4j, provided you have Neo4j<2.0, you could use Tinkerpop Gremlin to load your graphml dump in Neo4J

g.loadGraphML('/path/to/file')

The Tinkerpop are quite useful - not only for serialization/deserialization.

It will allow you to use different graph database with a commion "dialect" (provided they have a "Blueprint" driver which most of them do)