Skip to main content

FinOps: Cost Optimization

Key Takeaways for AI & Readers
  • Billable Requests: Cloud costs are determined by resource "requests" rather than actual usage; over-requesting leads to wasted spend.
  • Visibility Tools: Use Kubecost or OpenCost to attribute spending to specific teams and identify optimization opportunities.
  • Right-Sizing: Vertical Pod Autoscaler (VPA) and automated recommendations help align resource requests with actual historical performance.
  • Spot Instances: Leverage heavily discounted spare cloud capacity for stateless, interruption-tolerant workloads to significantly reduce infrastructure costs.

Cloud providers charge you for the Nodes you provision. If your developers request too much CPU/RAM, you are paying for "Ghost Resources"—capacity that is reserved but never used.

1. Requests vs. The Bill​

In Kubernetes, your cost is driven by Requests, not actual usage.

spec:
replicas:
template:
spec:
resources:
requests:
cpu: m
memory: Mi
Estimated Monthly Cost
$18.00
Efficiency (Requests vs Usage)42%
💡 Kubecost suggests reducing CPU to 100m.
In Kubernetes, you pay for what you Request, not what you use. Tuning these numbers is the key to FinOps.

If you have a Pod that uses 10MB of RAM but you request 1GB, you are paying for that 1GB for as long as the Pod is scheduled.

2. Kubecost & OpenCost​

To manage this, you need visibility. Kubecost is the leading tool for:

  • Allocation: Seeing exactly which Team or Namespace is spending the most money.
  • Efficiency: Finding "Idle" containers that have massive requests but low usage.
  • Right-sizing: Providing specific recommendations (e.g., "Change CPU request from 500m to 100m to save $40/mo").

3. Automated Savings​

Vertical Pod Autoscaler (VPA)​

Automatically adjusts the requests of your pods based on historical usage. If an app uses more than it requested, VPA increases the request.

Spot Instances (Karpenter / Cluster Autoscaler)​

Use "Spare Capacity" from AWS/GCP at a 70-90% discount.

  • The Catch: The cloud provider can take the node back with a 2-minute notice.
  • Strategy: Use Spot for stateless apps (Web servers) and On-Demand for stateful apps (Databases).

4. Resource Quotas​

As an admin, set ResourceQuotas on Namespaces to prevent a single team from accidentally running up a $10,000 bill overnight.