Helm chart: per-pod media LoadBalancer, signaling port default and disabled preStop hook #115

Found while updating the AKS docs (#114). None of these have been verified against a cluster.

1. external.yaml renders every media Service with the same name

templates/external.yaml loops scale.maxReplicaCount times. Each iteration emits a Service named {{ $fullName }}-external, and there is no --- separator between iterations (unlike service.yaml). With more than one replica this either fails to render or leaves only one media port/pod mapping. Meanwhile each bot advertises InstancePublicPort + podNumber (AzureSettings.cs), so pods other than the surviving one would get no media.

Suggested fix: name the Service {{ $fullName }}-external-{{ $i }} and add --- between iterations. Check with helm template ... -s templates/external.yaml --set scale.maxReplicaCount=3.

2. The default internal.port doesn't match the Services' target port

The bot serves plain HTTP on CallSignalingPort + 1 (AppHost.cs), and the deployment sets CallSignalingPort from internal.port. The per-pod and routing Services hard-code targetPort: 81. This only works with internal.port=80, which the CI sets but values.yaml doesn't (default 9441 → HTTP on 9442).

Suggested fix: set the values.yaml default to 80, or derive targetPort from internal.port + 1.

3. The preStop hook is commented out

The halt_termination.ps1 preStop hook in templates/deployment.yaml is commented out. Without it, terminationGracePeriod doesn't wait for ongoing calls, so scaling down or rolling updates can drop calls that are in progress.

Suggested fix: re-enable the hook, or document why it is disabled.

🤖 Generated with Claude Code