Repository navigation
Run:ai: test LoadBalancer and Kubernetes Gateway API exposure #1426
Description
Follow-up from review on #1425 (#1425 (review)).
#1425 added optional Run:ai support to the Kubernetes path, but only NodePort exposure was exercised. LoadBalancer is documented as untested, and Kubernetes Gateway API is not covered.
Scope
- Add test coverage for Run:ai exposed through a LoadBalancer service.
- Add test coverage for Run:ai using the Kubernetes Gateway API, per the Run:ai self-hosted v3.0 networking guide: https://docs.nvidia.com/run-ai/self-hosted/v3.0/documentation/installation-and-configuration/config/networking/kubernetes-gateway-api
- Update the Run:ai docs to drop the "LoadBalancer is untested" note once covered, and document Gateway API setup.
Done when
- The Run:ai validator passes in platform mode with LoadBalancer and with Gateway API exposure.
- Docs describe both options.
Activity
dholt commentedon Oct 20, 2026
Both exposure paths now pass the Run:ai validator in platform mode on a live single-node cluster (LoadBalancer through MetalLB with the HAProxy ingress, and Gateway API through KGateway), with clean uninstall after each. Change and test details are in #1432, which will close this issue when merged.