Queries on moving from Neo4J Embedded+SDN to Neo4J Server

Question

I have few queries w.r.t moving from Neo4J Embedded to Neo4J Server and would appreciate if you can provide clarifications on those.

Background

We have been using Neo4J (1.8.x) in embedded mode for one of our applications (event management platform), SDN (2.x) compatible with Neo4J 1.8.x and thereby leveraging the infrastructure provided by SDN (NodeEntity, RelationshipEntity, etc).

CUD operations are routed through the Neo4JTemplate & Reads are 100% done through parameterized Cypher queries.

Almost every domain object has its own index (index name specified via @Indexed) and we also use a combination of exact, full-text and spatial indexes.

Our traversals are not complex (max of 3-4 levels deep). We have a total of 60K+ nodes, 120K+ relationships, & 400K+ properties

We would now like to move from the embedded to the server mode and thus go with REST api's supported by the server mode. Apparently, switching to the server mode was easy i.e. switch to SpringRestGraphDatabase and things started working fine (except for the issue where we cannot post large text beyond 20KB (https://github.com/neo4j/neo4j/issues/1232)).

The problem that we have at hand is the performance. I have gone through several forums/topics talking about performance issues over REST and there were several points mentioned about making the calls course-grained or using rest-batch api or writing a server extension.

Queries

• I was exploring the Neo4J-Java-REST bindings which supports batching (i.e. BatchCallback). Going this route, I can now group together all my operations as a batch but then I would have to deal with Node objects and not the SDN mapped domain objects.

  Is this correct?

  If yes, is there any mechanism where I can deal with my domain objects instead of core entities (Node, etc) within the rest-batch-api using SDN

  Most of our domain object creations involve the below steps:

  • Create entity (say master) with the specified parameters

  • Build relationships between the entity and the owner (happens to be a user)

  • Some use-cases internally create additional entities and automatically build the relationship between the master and these

  • As almost every entity have their own index, the created entities are added to the respective index (via @Indexed)

  • All these steps are considered as an atomic operation

  With the batching api (java-REST), are they considered as an atomic operation? My impression was that they were sent as a batch but not necessarily executed as an atomic operation.

  Is my understanding correct? If yes, how do we treat them as atomic? Does this warrant us to develop a server extension?

  Also, are there any guidelines/best-practices for moving from SDN/Embedded to SDN/REST?

• As we were using 1.8 previously, we had used the manual indexes and now in 2.x we have the labels which we would like to adopt. Are there any guidelines/tools available to move the manual-indexed data over to labels?

• Is there any web-page which talks about the roadmap of neo4j releases?

Thanks & Regards,

Chandra

Answer 1

Regarding the use of SDN/REST, I posted a question here

To make things short, SDN/REST is not production ready.

Answer 2

This was the exact reason we moved from the server version to the embedded version. The performance differences are truly astonishing. I haven't since attempted using the latest versions of neo4j but I really hope this has been addressed as it is such a good and promising graph database. Sorry I can't provide a solution for you that you want to hear but if you do get one then I am very keen to hear about it. Cheers