Skip to content

Neo4jBatch Implementation

okram edited this page Jan 24, 2012 · 7 revisions
<dependency>
   <groupId>com.tinkerpop.blueprints</groupId>
   <artifactId>blueprints-neo4jbatch-graph</artifactId>
   <version>??</version>
</dependency>
Graph graph = new Neo4jBatchGraph("/tmp/neo4j");
// do your batch insertion
graph.shutdown();
graph = new Neo4jGraph("/tmp/neo4j");

Neo4jBatchGraph provides support for the bulk insertion of data into a Neo4j graph. This Blueprints implementation is single-threaded and non-transactional. It is meant for initial dumps of raw data into a Neo4j graph and is much faster than doing the equivalent process using Neo4jGraph (see Neo4j Implementation).

Note: This is not a pure Blueprints implementation. Consider the following issues when using Neo4jBatchGraph:

  1. Delete methods (except for Element.removeProperty()) are not supported.
  2. There are no default Index.VERTICES and Index.EDGES indices created at startup.
  3. The Neo4j “reference node” (vertex 0) is not automagically deleted upon graph creation.

Unlike Neo4jGraph, and like TinkerGraph, Neo4jBatchGraph can accept a long id on vertex creation:

long id = 1L;
graph.addVertex(id)

When utilizing Neo4jBatchGraph it is important to make use of the flush method on Neo4jBatchIndex. This method must be called before querying the index for data to ensure that it returns results consistently. This method is not a standard Index API method and thus, be sure to typecast the index to Neo4jBatchIndex. Note that calling flush (and using indices in general) has a noticeable impact on the performance when writing to the graph. Finally, see Neo4jBatchGraph.flushIndices() for flushing all indices at once.