Neo4j embedded mode use of GDS

Question

I am attempting to use GDS 1.8.2 as part of a system running an embedded Neo4j 4.4.3 server. The embedded server has been an operational component for several years, and several versions of Neo4j, so that component on its own has been time-tested and functions great. This is the first attempt to add support for graph algorithms into that component.

My first test is simply to submit a CQL query:

CALL gds.graph.create("someNamedGraph", ["SomeNodeLabel"], ["SomeRelationshipType"])

In the course of getting this to work, I found I had to register the org.neo4j.gds.catalog.GraphCreateProc class in the GlobalProcedures registry of the graph database. This seems to have been successful because, while I was initially encountering a CQL exception saying the gds.graph.create procedure is unknown, now it appears to execute without exception. However, I am now seeing that the transaction doesn't produce the named graph (verified by checking the graph database using out-of-the-box Neo4j Community Edition server mode). It only runs for perhaps 0.1 seconds (vs several seconds when done through the Neo4j Community Edition server mode where it works just fine).

What I now see is that the Query Execution Type (as indicated in the Result object coming back from the execution) is marked as READ_ONLY. There are no exceptions, notifications etc. I have verified that a subsequent write transaction in the same test code, which creates a simple node (as a test), succeeds in writing a node and the Result object provides all the verifying information for that transaction.

Can anyone suggest why the gds.graph.create procedure would seem to execute with no exceptions yet somehow is getting marked as a READ_ONLY transaction? Is this even the reason why the named graph isn't getting created?

Thank you for suggestions or tips! I'm happy to provide more details if anyone has exploratory questions that might help unearth the root cause for this.

score 0 · Answer 1 · answered Feb 12 '22 at 15:22

Providing an answer to my own question as this was resolved with assist from Mats Rydberg. The issue was that the call alone does not execute the operation. The result has to be iterated.

So a more fitting way to do this in embedded mode would be:

CALL gds.graph.create("someNamedGraph", ["someNodeLabel"], ["someRelationshipType"]) YIELD graphName

And, on the server-side:

try(Transaction tx = graphDb.beginTx())
{
    Result resultSet = tx.execute(cql);
    while(resultSet.hasNext())
        resultSet.next();
}

which indeed produces the named graph.

The root problem, to me, is that my original query (with no YIELD clause) works correctly in the built-in browser app for neo4j community edition server-mode, suggesting there are things happening behind the scenes, such that its obfuscated as to how that ends up working. Regardless, the solution to the problem is now understood and hopefully, in the future, there will be more documentation about making GDS work in embedded mode specifically.

Neo4j embedded mode use of GDS

1 Answers1