Several integration tests from QueryServiceTest are timing out

[gaurav]

I discovered this running the integration tests locally. They also timeout on cam-kp-api-dev.renci.org but run successfully on cam-kp-api. This includes:

• {"message":{"query_graph":{"nodes":{"n0":{"ids":["NCBIGene:3105","NCBIGene:3106","NCBIGene:64167","NCBIGene:60412","NCBIGene:26191"],"categories":["biolink:GeneOrGeneProduct"]},"n1":{"categories":["biolink:GeneOrGeneProduct"]}},"edges":{"e0":{"predicates":["biolink:affects"],"subject":"n0","object":"n1"}}}}}
• {"message":{"query_graph":{"nodes":{"n0":{"categories":["biolink:GeneOrGeneProduct"]},"n1":{"categories":["biolink:BiologicalProcess"]}},"edges":{"e0":{"predicates":["biolink:has_participant"],"subject":"n1","object":"n0"}}}}}

I've tried restarting cam-kp-sparql but that didn't solve the problem.

[gaurav]

I've tried increasing the memory on cam-kp-sparql from 32Gi to 64Gi, but that didn't fix it.

It might be related to this error I've been seeing in the CAM-KP SPARQL log -- maybe I've misconfigured something? The Helm chart I'm using is at https://github.com/helxplatform/translator-devops/pull/526

[gaurav]

Okay, I've found an example SPARQL query that works fine on the old triplestore, but that times out on the new triplestore. @balhoff Do you have any idea why the two Blazegraph databases might react so differently to this?

[balhoff]

@gaurav not exactly related to the performance issue, but I don't think that ?n0_class and ?n1_class should be projected out of the sub-select (and also shouldn't be in the GROUP BY).