Shamsher Khan

Thanks for the clarification—that’s really helpful. Makes sense that the core behavior is similar since the underlying components are the same.

Great article—really appreciate the clear breakdown of AKS capabilities, especially around scalability.
I’m curious, how does AKS autoscaling (cluster autoscaler + HPA) typically behave under sudden traffic spikes compared to self-managed Kubernetes clusters?

I appreciate that! Stay tuned for more deep dives like this.

Great insights, Vivek!

The tool uses kubectl get pods to detect unhealthy pods, including probe failures. No OpenAI needed for that.

OpenAI is used for diagnosis. When a pod fails, the tool sends probe configs, logs, and events to GPT-4 to determine why it's failing.

For probe failures: kubectl shows the probe failed, but GPT-4 analyzes whether it's slow startup (increase initialDelaySeconds), wrong endpoint, timeout settings, or an actual app problem.

Detection = kubectl

Diagnosis = GPT-4

You can skip the LLM and read kubectl output yourself - that's traditional troubleshooting. The LLM reads the same data and suggests fixes.

Thanks for reading — curious how others use AI in Kubernetes diagnostics