Draft: not yet run against a live cluster. Treat output as representative and check it on yours.
An Ingress object does nothing on its own. It’s configuration for a controller you have to install, and most edge failures come from the gap between what you wrote and what the controller actually loaded.
Layer
Ingress model
Gateway API model
Which implementation
IngressClass
GatewayClass
The entry point (listeners, ports, TLS)
Implicit in the controller
Gateway, owned by the platform team
Routing rules (hosts, paths, backends)
Ingress
HTTPRoute (and other route types), owned by app teams
Where status is reported
Mostly nowhere
status.conditions on every object
The controller is a reverse proxy running as pods. Traffic goes: load balancer, controller pods, then directly to pod IPs from the Service’s EndpointSlices.
Gateway API is the newer, role-separated model, and the place new routing features land. The Ingress API itself still works and is still everywhere. But the community ingress-nginx controller, the one most tutorials assume, was retired by the Kubernetes project in March 2026: no further releases, bug fixes, or security updates. Don’t choose it for new work, and if you run it today, plan the migration. (F5’s “NGINX Ingress Controller” is a different, vendor-maintained project with a confusingly similar name.)
The controller watches its objects. It only acts on Ingresses whose ingressClassName (or Gateways whose gatewayClassName) matches it. Anything else is silently ignored.
It builds proxy configuration from hosts, paths, TLS Secrets, and backend Services, then reloads or hot-updates the proxy.
It resolves backends to pod IPs. Most controllers read EndpointSlices and send traffic straight to pods, skipping kube-proxy. So a Service with no Ready endpoints means 503 at the edge.
TLS is terminated at the controller using the Secret named in the Ingress tls section or the Gateway listener’s certificateRefs. If the Secret is missing or invalid, most controllers fall back to a default self-signed certificate rather than failing loudly.
The steps below use a Gateway API implementation. Install one for kind by following its own documentation (Envoy Gateway, Istio, Cilium, NGINX Gateway Fabric, and others implement it), and install the Gateway API CRDs at the version that implementation requires.
apiVersion : gateway.networking.k8s.io/v1
gatewayClassName : <your-implementation-class>
apiVersion : gateway.networking.k8s.io/v1
parentRefs : [{ name : edge }]
- backendRefs : [{ name : web , port : 80 }]
kubectl get gateway edge -o jsonpath='{range .status.conditions[*]}{.type}={.status} {.reason}{"\n"}{end}'
kubectl get httproute web -o jsonpath='{range .status.parents[*].conditions[*]}{.type}={.status} {.reason}{"\n"}{end}'
Healthy output shows Accepted=True and Programmed=True on the Gateway, and Accepted=True and ResolvedRefs=True on the route.
Failure: Rule ignored 404 from the default backendWhat you see. The controller answers, but with its default 404 page.
Root cause. The request didn’t match any rule the controller loaded: the Host header doesn’t match the rule’s host, the path type is Exact where you needed Prefix, or the ingressClassName points at a different controller so your Ingress was never loaded at all.
Diagnose. kubectl describe ingress <name> and look for events from the controller. No events often means no controller claimed it. Test with an explicit host: curl -H 'Host: web.local' http://<lb-address>/.
Failure: Backend unavailable 502 Bad Gateway / 503 Service UnavailableWhat you see. The route matches, but the edge returns 502 or 503.
Root cause. 503 usually means the backend Service has no Ready endpoints. 502 usually means the controller reached a pod but the connection failed: wrong Service port, the app speaks HTTPS while the controller sends HTTP, or the app closed the connection.
Fix. Work backwards: Service endpoints, then the pod’s port, then protocol. See Ingress returns 502 or 503 .
Failure: Route not attached HTTPRoute Accepted=False or ResolvedRefs=FalseWhat you see. Traffic doesn’t flow, and the HTTPRoute’s status shows a False condition with a reason such as NotAllowedByListeners, NoMatchingParent, BackendNotFound, or RefNotPermitted.
Root cause. Gateway API is explicit about trust. A listener only accepts routes from namespaces its allowedRoutes permits, and a route can only reference a backend in another namespace if that namespace has a ReferenceGrant allowing it.
Fix. Read the reason. It names the exact rule that blocked the attachment.
Use Gateway API for new work. Its status conditions alone save hours of debugging, and the platform/app ownership split matches how real teams operate.
Run at least two controller replicas with a PodDisruptionBudget , spread across nodes. The edge proxy is a shared single point of failure.
Monitor certificate expiry at the edge , not just in the Secret. A misconfigured reference serves the controller’s fallback certificate without any error.
In regulated production PCI DSS v4.0, SAMA CSFStrong TLS only. PCI DSS Requirement 4 expects strong cryptography for cardholder data over open networks: TLS 1.2 or higher, no weak ciphers. Set the minimum version and cipher list in the controller’s configuration, and keep a scan result as evidence.
Decide where TLS ends. If it terminates at the edge, traffic from the controller to pods is plaintext inside the cluster. For in-scope services, either re-encrypt to the backend or use a mesh with mTLS, and document which one you chose.
On OpenShift , Routes and the ingress operator play this role. The same questions apply: TLS termination type (edge, reencrypt, passthrough) and minimum TLS version in the IngressController’s TLS security profile.
You created an Ingress, and kubectl describe shows no events at all. What's the likely cause? No controller claimed it. Usually the ingressClassName doesn’t match an installed controller, or no controller is installed.
503 versus 502 at the edge: which usually means no Ready endpoints?
A 502 usually means a pod was reached but the connection or protocol failed.
An HTTPRoute references a Service in another namespace and shows RefNotPermitted. What's missing? A ReferenceGrant in the Service’s namespace that allows HTTPRoutes from the route’s namespace to reference it.
Maps to: CKA: Services and Networking. Check the current exam curriculum, since domain names change between revisions.