Evaluate Kubernetes Node Scaling with LLM

Pipes Kubernetes node resource utilization metrics into an LLM to analyze cluster provisioning and suggest instance type optimizations.

$ kubectl top nodes | llm -m gpt-4o "Analyze this node utilization. Are we over-provisioned or under-provisioned? Suggest instance type changes."

1. [01] kubectlkubectl top nodes
   Fetches current CPU and memory utilization metrics for all nodes in the Kubernetes cluster.
2. [02] bash|
   Pipes the tabular metrics output directly into the standard input of the LLM CLI.
3. [03] llmllm -m gpt-4o
   Invokes the LLM CLI using the gpt-4o model to process the piped data.
4. [04] llm"Analyze this node utilization..."
   The prompt instructing the LLM on how to interpret the metrics and what specific recommendations to provide.

Based on the provided metrics:
- Node 1 (CPU: 12%, Mem: 45%) and Node 2 (CPU: 15%, Mem: 50%) show low CPU utilization.
- You are currently CPU over-provisioned.
Recommendation: Consider scaling down or switching to memory-optimized instances (e.g., AWS r6g.large instead of m6g.xlarge) to reduce costs while meeting memory demands.

• Footgun: `kubectl top` only shows a point-in-time snapshot, which may not reflect peak load or daily traffic spikes.
• Permission/Cost: Requires an active OpenAI API key which incurs usage costs, and cluster read permissions for node metrics.