Introduction

The NGINX Ingress Controller is supposed to watch Ingress resources and reload its configuration when paths change. If new routes are not being used, the problem is often not "how do I update the controller binary" but "why did the controller not reconcile the latest Ingress definition?" The fix usually starts with verifying that the right controller instance is watching the right Ingress objects.

Confirm the Ingress Object Is Actually Valid

Start by checking the manifest that Kubernetes accepted.

kubectl get ingress my-app -o yaml
kubectl describe ingress my-app

Look for:

• the expected host and path rules
• the correct ingressClassName
• the right backend service name and port
• warning events about rejected configuration

If the resource itself is wrong, restarting the controller will not help. Fix the Ingress object first.

Verify the Controller Watches This Ingress Class

A common cause of stale routes is an ingress class mismatch. Your Ingress may be using one class name while the controller deployment is watching another.

Example Ingress:

1apiVersion: networking.k8s.io/v1
2kind: Ingress
3metadata:
4  name: my-app
5spec:
6  ingressClassName: nginx
7  rules:
8    - host: app.example.com
9      http:
10        paths:
11          - path: /api
12            pathType: Prefix
13            backend:
14              service:
15                name: api-service
16                port:
17                  number: 80

If the controller was installed with a different class, it may ignore this object entirely. Always confirm the deployment and the Ingress agree on class ownership.

Check the Controller Logs and Config Reloads

When the controller detects a change, it should log reconciliation or reload activity.

kubectl logs -n ingress-nginx deploy/ingress-nginx-controller

You are looking for evidence that:

• the new Ingress was seen
• the generated NGINX config was reloaded
• no annotation or path validation error blocked the update

If the logs stay quiet after the Ingress changes, the controller may not be watching that namespace or class, or the event stream may be unhealthy.

Roll Out an Updated Controller When Needed

If the issue really is controller version drift, update the controller with the same mechanism used to install it. With Helm, that is usually:

helm repo update
helm upgrade ingress-nginx ingress-nginx/ingress-nginx \
  --namespace ingress-nginx

Afterward, verify the rollout:

kubectl rollout status deployment/ingress-nginx-controller -n ingress-nginx

If you only need the controller to re-read current resources after config drift or a bad state, a restart may be enough:

kubectl rollout restart deployment/ingress-nginx-controller -n ingress-nginx

Use restart for reconciliation problems, and use upgrade for actual version changes.

Validate the Effective Routing End to End

After any change, test the full chain:

• Ingress object
• controller logs
• generated routes
• backend service endpoints

If the path still fails, the problem may be downstream. For example, the controller might route /api correctly while the backend service has no healthy endpoints or the service port is wrong.

Common Pitfalls

• Restarting or upgrading the controller before checking whether the Ingress object itself is valid.
• Forgetting that ingressClassName must match the controller that should reconcile the resource.
• Blaming route staleness on NGINX when the backend service or endpoints are misconfigured.
• Upgrading the controller when a simple rollout restart would have been enough.
• Ignoring controller logs, which usually reveal whether the path update was accepted or rejected.

Summary

• New Ingress paths should normally be picked up automatically by the NGINX Ingress Controller.
• First verify the Ingress object, class, and backend details.
• Use controller logs to confirm whether reconciliation and config reload actually happened.
• Use helm upgrade for version updates and kubectl rollout restart for stuck reconciliation state.
• Validate routing end to end so you do not mistake a backend problem for a controller update problem.