Troubleshooting Common Issues

Azure Compute Solutions

Florin Angelescu

Azure Cloud Architect

Why troubleshooting matters

 

 

 

• Pods may fail to start

 

• Services might not route traffic

 

• Nodes could run out of resources

Why troubleshooting matters

 

• Structured approach:
  • Observe
  • Identify
  • Test
  • Resolve
• Ensures you don't waste time chasing symptoms instead of root causes.

Pod failures

 

 

 

 

• Pods failing to start.
• Causes:
  • Incorrect container images
  • Missing secrets
  • Insufficient resources

Pod failures

• kubectl describe pod - events
• kubectl logs - output

• Check image pull policies
• Check registry credentials

• Investigate application logs
• Investigate resource requests

• Readiness and liveness probes
• Detect unhealthy pods and restart them

Networking problems

 

 

 

• Networking issues:

  • Prevent services from reaching pods
  • Or external clients

• Verify:

  • Services are correctly defined
  • Selectors match pod labels

Networking problems

• kubectl get svc
• kubectl get endpoints

• Ingress controllers - additional config
• TLS certificates or path rules

• Testing connectivity
• kubectl exec, curl

• Packet capture tools
• Azure Network Watcher

Scaling challenges

 

 

 

 

• Auto-scaler settings are mis-configured.
• Nodes lack capacity.

Scaling challenges

 

• Check Horizontal Pod Autoscaler metrics:
  • kubectl get hpa
• Ensure the Cluster Autoscaler is enabled.
• Review resource requests and limits:
  • Overly restrictive values can prevent pods from scheduling
• Inspect node pool quotas and adjust thresholds.
• Simulate load during testing.
• Monitor scaling events in Azure Monitor.

Resource constraints

 

 

• Nodes can run out of:
  • CPU
  • Memory
  • Disk space
• Causing pods to be evicted.
• Monitor resource usage:
  • Azure Monitor
  • kubectl top

Resource constraints

• Over-commit resources leads to instability
• Define realistic requests and limits

• Taints and tolerations -> pod placement
• Ensure critical workloads have priority

• Audits of resource allocation and quotas
• Prevent bottlenecks

• Multiple node pools with different VM sizes
• Balance workloads efficiently

Recap

 

 

 

• Troubleshooting in AKS involves diagnosing:
  • Pod failures
  • Networking issues
  • Scaling challenges
  • Resource constraints

Recap

 

 

 

• Combining Kubernetes tools with Azure integrations:

  • You can resolve problems quickly and maintain reliability

• Troubleshooting play-book for your team:

  • Consistent responses and faster resolution times

Let's practice!

Azure Compute Solutions