How can I change graphd service type to LoadBalancer? currently ClusterIp hardcoded

[porscheme]

Hi @wey-gu, how can we change graphd service type to LoadBalancer? currently ClusterIp hardcoded!

[wey-gu]

Dear @porscheme,

Nebula GraphClient/GraphD endpoint are load-balanced from the client-side, if you look into those clients, we specify all graphD addresses to the connetion_pool and it will round-robin from them.

To me, the load balancer/api gateway pattern cannot be applied to the nebula graph.

@MegaByte875 @veezhang @kqzh kindly add/correct if I understood it wrongly.

Thanks!

[porscheme]

nebula-operator:v1.1.0
Nebula version: v3.1.0
Helm Charts

• Client has to be with K8S cluster to reach graphd, since its using ClusterIP
• Also, below simple query taking 20 minutes, Any tips t-shoot slow response?

MATCH (s:Student)-[r]-(a:CourseTcode)-[rr]-(b)
WHERE a.CourseTcode.id == 522687
RETURN s, r, a, rr, b limit 3

[MegaByte875]

@porscheme https://github.com/vesoft-inc/nebula-operator/tree/master/config/samples we also provide graph-nodeport service and nginx-ingress, you can have a try.

[wey-gu]

@porscheme

Client has to be with K8S cluster to reach graphd, since its using ClusterIP

Now I understood that you would like to expose its endpoint rather than questioning on load balancing part, we should create individual services per each graphD instance (as I mentioned above, client-side load balancing) as @MegaByte875 shared in the samples.

@MegaByte875, is it possible to provide those nodeport or ingress services per graphD by the operator itself? I believe this would be super helpful for users as the application layer in most cases is not in the same namespace or even cluster of NebulaClsuter. If it makes sense and doable, I would create the feature issue then :)

[porscheme]

Thaanks @wey-gu & @MegaByte875.

• Our Spark & Nebula are in two different K8S cluster

@porscheme https://github.com/vesoft-inc/nebula-operator/tree/master/config/samples we also provide graph-nodeport service and nginx-ingress, you can have a try.

Can you make this part of the helm chart?

[MegaByte875]

@wey-gu Yes, this can be a feature @porscheme OK, I will make the service configurable by helm charts

[porscheme]

@wey-gu Yes, this can be a feature @porscheme OK, I will make the service configurable by helm charts
This feature is very important for us to use Nebula-spark-connector, check with @wey-gu for more details.
This is blocking us.
Can we do this over this weekend, I can be your tester :)

[MegaByte875]

#147 @porscheme

[MegaByte875]

I will close the issue, please re-open it if you have other questions.