Q: Ingress shows `Address: <pending>`.
No LoadBalancer IP assigned yet. On bare-metal, need MetalLB. On cloud, wait 1-2 min for cloud LB provisioning.
Want to master this scenario in a live sandbox? KodeKloud's CKA & CKAD Hands-On Certification Track covers this exact problem with hands-on terminal drills.
🛠️ Production Runbook & Step-by-Step Resolution
Direct Answer: When an <strong>Ingress shows Address pending</strong> (or an empty ADDRESS column), no controller has written a load balancer address to the Ingress status. That means the controller is missing, is ignoring the Ingress because of the class, or its own Service of type LoadBalancer never received an external IP. Interviewers ask this because an Ingress object is only configuration; without a running controller and a reachable entry point it does nothing, and candidates should trace the chain from Ingress to controller to Service to cloud.
What Populates the Ingress Address
The ADDRESS field is written by the controller, not by Kubernetes core.
- Controller required: creating an Ingress without an ingress controller (NGINX, AWS Load Balancer Controller, Traefik, GCE and others) leaves ADDRESS empty indefinitely.
- Class selection: a controller only reconciles Ingresses whose
spec.ingressClassNamematches its IngressClass (or that match its legacy annotation rules). This is why creating an Ingress alone, with no IngressClass in the cluster, never produces an address, no matter how correct the rules are. - Address source: controllers such as ingress-nginx copy the external IP or hostname from their own LoadBalancer Service. If that Service is pending, the Ingress stays pending.
- Cloud-managed controllers: the AWS Load Balancer Controller creates an ALB and writes its DNS name, so IAM, subnet tags and quotas matter.
- Quick triage rule: if the Ingress has events but no address, look at the controller; if it has no events at all, look at the class, because no controller is even considering it.
Hands-on: Ingress Address Pending Triage
Walk down the chain and stop at the first thing that is wrong.
kubectl get ingress my-app -n prod
kubectl describe ingress my-app -n prod | tail -15
kubectl get ingressclass
kubectl get pods -A | grep -Ei 'ingress|traefik|alb|haproxy'
kubectl get svc -n ingress-nginx
# EXTERNAL-IP <pending> on the controller Service is the common root cause
kubectl describe svc ingress-nginx-controller -n ingress-nginx | tail -15
kubectl logs -n ingress-nginx deploy/ingress-nginx-controller --tail=50
- Ingress events:
kubectl describe ingressoften contains controller errors in Events at the bottom. - Controller present: list IngressClasses and the controller pods; none means you need to install one. For a cloud-managed controller, also read the controller's own logs because address assignment depends on a successful call to the cloud API.
- Controller Service: look at EXTERNAL-IP of its LoadBalancer Service;
<pending>there is the real issue.
Root Causes and Fixes
Match the signature to the cause.
- No controller installed: no IngressClass and no controller pods. Install one with Helm or manifests, then recreate or wait for reconciliation.
- Class mismatch: the Ingress uses an
ingressClassNamethat does not exist, or none is set and there is no default IngressClass (annotationingressclass.kubernetes.io/is-default-class: "true"). Fix the class name. - Controller Service pending, bare metal or on-prem: no cloud provider exists to fulfil type LoadBalancer. Install MetalLB or kube-vip, or expose the controller with NodePort or hostNetwork and put an external load balancer in front.
- Cloud provisioning failure: events on the Service or Ingress show quota exhaustion, missing subnet tags (for AWS,
kubernetes.io/role/elbon public subnets), or missing IAM permissions for the controller. checkkubectl logs -n kube-system deploy/aws-load-balancer-controllerfor subnet discovery or IAM errors, and make sure the Ingress annotations select the intended scheme (internet-facing or internal). - Local clusters: minikube needs
minikube tunnelor the ingress addon; kind needs extra port mappings and a ready controller.
Verification
Prove the address exists and routes.
# Confirm routing without touching DNS
ADDR=$(kubectl get ingress my-app -n prod -o jsonpath='{.status.loadBalancer.ingress[0].hostname}{.status.loadBalancer.ingress[0].ip}')
curl -sv -H 'Host: app.example.com' "http://$ADDR/healthz"
# Expect 200, or a 404 from the controller if host or path rules do not match
- Address populated:
kubectl get ingress my-app -n prod -o jsonpath='{.status.loadBalancer.ingress}'returns an IP or hostname. Remember that an ADDRESS shown for a pending-then-fixed Ingress can take a short while to appear after the controller Service receives its IP, so re-run the command a few times before concluding the fix failed. - Route test before DNS:
curl -sv -H 'Host: app.example.com' http://<address>/checks routing without waiting for DNS changes. - Backends: a 503 or 404 after the address appears points to the backend Service or the host and path rules, which is a different problem.
- DNS follow-up: once the address exists, point the host name at it with a CNAME or A record, ideally automated by ExternalDNS.
Production Practice and Alternatives
Keep the entry point declarative and monitored.
- Declare the class explicitly: always set
ingressClassNameso the intended controller reconciles the Ingress. Declaring the class also makes the manifest portable across clusters where a different default exists. - Alert on pending: monitor controller Service external IP assignment and controller error logs in infrastructure tests.
- Check controller lifecycle: project maintenance status changes over time, so verify the support status of your controller (including ingress-nginx) and evaluate Gateway API for new designs.
- An Ingress is only configuration, and its address is written by a controller after its entry point exists.
- I check IngressClass, controller pods and logs, then the controller Service EXTERNAL-IP and cloud events.
- I set ingressClassName explicitly, use MetalLB on bare metal, and test routing with a Host header before DNS.
Senior hiring managers and bar raisers evaluate how you prioritize during outages. Avoid these common anti-patterns:
- Editing the Ingress YAML repeatedly The Ingress is rarely wrong; the missing piece is the controller, class or load balancer.
- Leaving ingressClassName unset with no default class No controller claims the Ingress, so it stays unreconciled and pending.
- Expecting type LoadBalancer to work on bare metal Without a cloud provider or MetalLB nothing assigns the external IP.
- Ignoring controller logs and cloud events Quota, IAM and subnet tag errors are only reported there.