SKILL.md
Kubernetes Cost Optimization
Cost analysis and optimization using kubectl-mcp-server's cost tools.
When to Apply
Use this skill when:
- User mentions: "cost", "savings", "optimize", "expensive", "budget"
- Operations: cost analysis, right-sizing, cleanup unused resources
- Keywords: "how much", "reduce", "efficiency", "waste", "overprovisioned"
Priority Rules
| Priority | Rule | Impact | Tools |
|---|---|---|---|
| 1 | Find and delete unused PVCs | CRITICAL | findorphanedpvcs |
| 2 | Right-size overprovisioned pods | HIGH | getresourcerecommendations |
| 3 | Identify idle LoadBalancers | HIGH | get_services |
| 4 | Scale down non-prod off-hours | MEDIUM | scale_deployment |
| 5 | Consolidate small namespaces | LOW | Analysis |
Quick Reference
| Task | Tool | Example |
|---|---|---|
| Namespace cost | getnamespacecost |
getnamespacecost(namespace) |
| Cluster cost | getclustercost |
getclustercost() |
| Unused PVCs | findorphanedpvcs |
findorphanedpvcs(namespace) |
| Right-sizing | getresourcerecommendations |
getresourcerecommendations(namespace) |
Quick Cost Analysis
Get Cost Summary
get_namespace_cost(namespace)
get_cluster_cost()Find Unused Resources
find_unused_resources(namespace)
find_orphaned_pvcs(namespace)Resource Right-Sizing
get_resource_recommendations(namespace)
get_pod_metrics(name, namespace)Cost Optimization Workflow
1. Identify Overprovisioned Resources
get_resource_recommendations(namespace="production")
get_pod_metrics(name, namespace)
get_resource_usage(namespace)2. Find Idle Resources
find_orphaned_pvcs(namespace)
find_unused_resources(namespace)3. Analyze Node Utilization
get_nodes()
get_node_metrics()Right-Sizing Guidelines
| Current State | Recommendation |
|---|---|
| CPU usage < 10% of request | Reduce request by 50% |
| CPU usage > 80% of request | Increase request by 25% |
| Memory < 50% of request | Reduce request |
| Memory near limit | Increase limit, monitor OOM |
Cost by Resource Type
Compute (Pods/Deployments)
get_resource_usage(namespace)
get_pod_metrics(name, namespace)Storage (PVCs)
get_pvc(namespace)
find_orphaned_pvcs(namespace)Network (LoadBalancers)
get_services(namespace)Multi-Cluster Cost Analysis
Compare costs across clusters:
get_cluster_cost(context="production")
get_cluster_cost(context="staging")
get_cluster_cost(context="development")Cost Reduction Actions
Immediate Wins
- Delete unused PVCs:
findorphanedpvcs()then delete - Right-size pods: Apply
getresourcerecommendations() - Scale down dev/staging: Off-hours scaling
Medium-term Optimizations
- Use Spot/Preemptible nodes: For fault-tolerant workloads
- Implement HPA: Auto-scale based on demand
- Use KEDA: Scale to zero for event-driven workloads
Long-term Strategy
- Reserved instances: For stable production workloads
- Multi-tenant clusters: Consolidate small clusters
- Right-size node pools: Match workload requirements
Automated Analysis Script
For comprehensive cost analysis, see [scripts/find-overprovisioned.py](scripts/find-overprovisioned.py).
KEDA for Cost Savings
Scale to zero with KEDA:
keda_scaledobjects_list_tool(namespace)
keda_scaledobject_get_tool(name, namespace)KEDA reduces costs by:
- Scaling pods to 0 when idle
- Event-driven scaling (queue depth, etc.)
- Cron-based scaling for predictable patterns
Related Skills
- [k8s-autoscaling](../k8s-autoscaling/SKILL.md) - HPA, VPA, KEDA
- [k8s-troubleshoot](../k8s-troubleshoot/SKILL.md) - Resource debugging