Troubleshooting Common Cluster Issues
What You'll Learn
You'll learn systematic approaches to diagnosing and resolving the three most common managed Kubernetes cluster problems: node failures, networking issues, and resource constraints.
Core Troubleshooting Framework
Think of cluster troubleshooting like medical diagnosis—you observe symptoms, gather data, form hypotheses, and apply targeted fixes. In managed Kubernetes, most issues fall into predictable categories.
Node Failures
When nodes become unavailable or unhealthy, check:
- Node status: Use
kubectl get nodesto identify NotReady nodes - Cluster Autoscaler events: Review logs for scale-up failures or quota issues
- Node conditions: Inspect disk pressure, memory pressure, or PID exhaustion with
kubectl describe node - Provider console: Verify VM health, security groups (EKS), firewall rules (GKE), or NSG rules (AKS)
Quick action: Drain and replace problematic nodes with fresh ones from your node pool.
Networking Issues
Pod-to-pod or service connectivity problems often stem from:
- CNI plugin issues: Check VPC CNI (EKS), Azure CNI, or GKE network plugin logs
- Network policies: Verify policies aren't blocking legitimate traffic
- DNS failures: Test with
kubectl run busybox --image=busybox --rm -it -- nslookup kubernetes.default - Load balancer misconfigurations: Verify security groups allow health check traffic
Resource Constraints
When pods remain Pending or get OOMKilled:
- Check cluster capacity: Use
kubectl top nodesand review available resources - Inspect pod events:
kubectl describe podreveals scheduling failures - Review resource requests: Ensure pods have appropriate CPU/memory requests
- Cluster Autoscaler: Confirm it's enabled and has permission to add nodes
Key Takeaway: Systematic troubleshooting of node health, network connectivity, and resource availability—combined with provider-specific tools like CloudWatch, Cloud Monitoring, or Azure Monitor—resolves most cluster issues quickly. Always start with kubectl describe and provider console logs.