Shut down inference containers when stream is down

[dthaler]

Here’s a simplified example that scales down the inference-system Deployment in the point-robinson namespace:

using System;
using System.Threading.Tasks;
using k8s;
using k8s.Models;

class Program
{
    static async Task Main(string[] args)
    {
        // Load kubeconfig (default: ~/.kube/config)
        var config = KubernetesClientConfiguration.BuildConfigFromConfigFile();
        IKubernetes client = new Kubernetes(config);

        string namespaceName = "point-robinson";
        string deploymentName = "inference-system";

        // Get the Deployment
        var deployment = await client.ReadNamespacedDeploymentAsync(deploymentName, namespaceName);

        // Set replicas to 0
        deployment.Spec.Replicas = 0;

        // Update the Deployment
        await client.ReplaceNamespacedDeploymentAsync(deployment, deploymentName, namespaceName);

        Console.WriteLine($"Scaled down {deploymentName} in {namespaceName} to 0 replicas.");
    }
}

• This only affects the Deployment in that namespace — other namespaces remain untouched.
• Once replicas = 0, the ReplicaSet won’t recreate pods, so CPU usage drops to zero for that workload.
• You can later set deployment.Spec.Replicas = 1 (or more) to bring it back.