Kubernetes Simulation

#IaaS #Onboarding #Concepts

1. HPA (Horizontal Pod Autoscaler) Issue

Colleague:

“Why doesn’t our service automatically scale up Pods even when CPU is maxed out?”

Chris:

“HPA relies on the Metrics Server to pull CPU/memory usage. If the Metrics Server isn’t installed or lacks RBAC permissions, the values come back as 0 and HPA won’t trigger. Also, if there are no Resource Requests set on the Deployment, HPA can’t calculate a baseline threshold. So you need to verify the Metrics Server is working properly and check that Requests are configured.”

2. Ingress Configuration Conflict

Interviewer:

“If you’re using multiple Ingress resources under a single Ingress Controller and a conflict occurs, how would you resolve it?”

Chris:

  • First, check with kubectl describe ingress to see which rules are being applied with priority.
  • Ingress Controllers (like NGINX) follow path-based/host-based routing rules, and priority conflicts can arise.
  • The solutions are clearly separating rules by Host, setting priority via Annotations, and if necessary, isolating controllers using IngressClass.

3. RBAC Issue

Colleague:

“Deployments are failing in the CI/CD pipeline, and the error log says Forbidden: user does not have access.”

Chris:

“That’s a ServiceAccount permissions issue. It happens when the SA running the pipeline doesn’t have permission to modify Deployments. Check the permissions with kubectl get clusterrolebinding, and if needed, create a Role/ClusterRole and bind it to the SA using kubectl create rolebinding.”

4. PersistentVolume Conflict

Interviewer:

“A PVC is stuck in Pending state and won’t bind. How would you go about identifying the cause?”

Chris:

  • First, verify that the StorageClass requested by the PVC matches the StorageClass of the PV.
  • If the requested capacity is larger than the PV’s available capacity, they won’t match.
  • A mismatch in AccessMode (RWO/ROX/RWX) can also cause issues.
  • In an actual cloud environment, check the CSI driver logs to verify that provisioning completed successfully.

5. Node Resource Shortage

Colleague:

“A Pod is stuck in Pending and won’t schedule. The event shows Insufficient CPU.”

Chris:

“That means the cluster needs an Autoscaler or additional nodes. First, check the scheduling failure reason with kubectl describe pod, then review the node resource status with kubectl top nodes. If the requested resources are large, consider adjusting the Pod spec, and also check whether Taints/Tolerations are restricting placement to specific nodes.”

6. Service Discovery Failure

Interviewer:

“A Pod tried to call another Service by DNS name but the connection failed. How would you troubleshoot this?”

Chris:

  1. Check whether the CoreDNS Pod is running properly (kubectl get pods -n kube-system).
  2. Verify that the nameserver inside /etc/resolv.conf is pointing to the CoreDNS ClusterIP.
  3. Check whether a NetworkPolicy is blocking the communication.
  4. Finally, note that DNS records differ depending on whether the Service is Headless or a regular ClusterIP — make sure they match accordingly.