gds.graph.create.cypher ArrayIndexOutOfBoundsException error

[tomasonjo]

I stumbled upon this more by accident than anything. Basically, if we project 0 nodes and 1 relationship, we get the array index out of bounds exception. We could try and catch this error and return a more descriptive error such as "You failed to project any node" or something.

Create graph:

CREATE (a:Node)-[:REL]->(b:Node)

Project graph

CALL gds.graph.create.cypher(
  'not_exist',
  'MATCH (n:NotExist) RETURN id(n) AS id',
  'MATCH (n)-->(m) RETURN id(n) AS source, id(m) AS target'
)

We get the following error:

Failed to invoke procedure gds.graph.create.cypher: Caused by: java.lang.ArrayIndexOutOfBoundsException: 0

[Mats-SX]

I agree that this should fail in a better way. I think for any graph creation, we should check the node count and if it is < 1 we should fail with a helpful message.

We should also check that this happens for native projections. We already fail for non-existent labels, but I'm not sure what happens for an existing, but empty label.

[Mats-SX]

I just checked: if we try and natively project a label that exists but has a nodecount = 0, we just accept that. I think we should raise an error for this case consistent with your case detailed above.

[FlorentinD]

A recently added config-parameter called validateRelationships, which is set to true by default for gds.graph.create.cypher now correctly throws an error:
Failed to invoke procedure gds.graph.create.cypher: Caused by: java.lang.IllegalStateException: Failed to load relationship with unknown source-node id 0.

So the behavior should be fixed in the next release

[Mats-SX]

Yes, using validateRelationships should help -- but we will also make the system refuse any graph without nodes consistently, regardless of the validateRelationships feature. Coming soon.