Grappling with Kubernetes costs effectively requires continuously monitoring of both resource allocation and utilization. Most platform, DevOps, and FinOps teams struggle with:

• Overprovisioned workloads: Default resource requests and resource limits are typically set with a buffer across Kubernetes clusters to safeguard performance, leading to wasted compute resources.
• Manual resource management: Tuning CPU and memory requests and limits manually across large Kubernetes infrastructure estates is incredibly time consuming and error-prone.
• Ensuring Reliability: Rightsizing resources often comes at the expense of performance. Using common tools like the horizontal pod autoscaler make it difficult to cut costs while maintaining application performance.
• Unrestricted Cloud Spending: Ungated growth of the Kubernetes cluster without continuous optimization leads to high compute costs across different cloud providers.

Managing Kubernetes resource utilization effectively comes down to properly setting CPU and memory requests, and limits on those workloads. In Kubernetes configuration, requests are used to set the minimum resources a container is guaranteed to access, while limits constrain the maximum resources a container can consume on a node. Finding the optimal values can be extremely challenging — especially at scale.

Rafa Brito learned about this the hard way as the platform engineering manager of a large bank in 2016 when he opened the first Kubernetes cluster to hundreds of users. He experienced what we now see as the typical resource management journey. 

The second source of Kubernetes overspending according to the CNCF survey is a lack of clarity around responsibilities for both teams and individuals, cited by 45% in the CNCF survey. 

Typically, teams rely on developers who are incentivized to deliver new features and avoid reliability issues as much as possible. That means cost always takes a back seat to performance. There ends up being a lot of finger pointing and hot potato passing when it comes to setting CPU and memory requests and limits.  

Continuously automate Kubernetes deployments for cost reduction, adjusting resources on a schedule to ensure efficiency. Recommendations can be deployed on demand or gated on a threshold, ensuring only changes that you've specified get deployed.

Optimize Live layers on top of tools that already exist in the Kubernetes ecosystem. You can implement horizontal auto-scaling using the HPA or KEDA, and double efficiency improvements by adding Karpenter for node resource scaling. 

StormForge integrates seamlessly with popular cloud cost reporting tools like Cloudbolt, providing granular insights into your Kubernetes costs. Together these tools provide cost monitoring and a detailed view of cloud spending and savings opportunities to quickly take action on.