set port to connect to for frontend configmap

[DaedalusG]

closes PLAT-757

Description

The executor Kubernetes deployment configurations contain incorrect or incomplete frontendUrl values that cause the executor to silently hang on startup, never connecting to the Sourcegraph instance.

The Sourcegraph frontend Kubernetes service exposes port 30080 in both Helm and Kustomize deployments. The Kustomize executor ConfigMap patches ship with http://sourcegraph-frontend (no port), and the Helm dind chart's values.yaml provides no guidance on the correct port. Both result in a silent connection failure with no error log.

When a URL omits the port, the HTTP client defaults to port 80 per the HTTP spec. The frontend Kubernetes service does not listen on port 80, so the connection attempt hangs indefinitely — no TCP RST (because the pod IP exists), no application-level error (because the request never reaches the frontend), and no timeout (because Go's default HTTP client has no dial timeout configured). The executor simply blocks forever on its initial connection attempt with no indication of what went wrong.

How I Identified It

During end-to-end testing of the executor dind Helm chart in a local kind cluster (Apple Silicon / OrbStack with Rosetta), the executor container started successfully but never registered with the Sourcegraph instance. The executor logs showed:

{"Body":"Connecting to Sourcegraph instance","Attributes":{"url":"http://sourcegraph-frontend:3080"}}

No follow-up log ever appeared — no "Connected", no error, no timeout. The container sat indefinitely on this line. I confirmed the frontend pod was healthy (1/1 Running, HTTP 200 via port-forward).

Testing connectivity from within the executor pod (via the dind sidecar) revealed the issue:

$ wget -q -O- --timeout=5 http://sourcegraph-frontend:3080/sign-in
# → download timed out

$ wget -q -O- --timeout=5 http://sourcegraph-frontend:30080/sign-in
# → 200 OK

The frontend service definitions confirmed it:

# Helm: charts/sourcegraph/templates/frontend/sourcegraph-frontend.Service.yaml
# Kustomize: base/sourcegraph/frontend/sourcegraph-frontend.Service.yaml
spec:
  ports:
    - name: http
      port: 30080        # ← service port (what in-cluster clients must use)
      targetPort: http    # ← maps to container port 3080

After changing frontendUrl to http://sourcegraph-frontend:30080, the executor connected immediately.

Note: The incorrect port (3080) in my test came from my own override file, not from a docs example — the docs don't show any port at all, which is part of the problem.

Checklist

• Follow the manual testing process
• Update changelog
• Update Kubernetes update doc

Test plan

[DaedalusG]

This recommendation follows from Sander's old comment for this setting https://github.com/sourcegraph/deploy-sourcegraph-helm/blame/b66aacea07ad62cf5d309e9044326bcefecdc09d/charts/sourcegraph-executor/k8s/values.yaml#L73-L75