No video of the event yet, sorry

In a large organization, developers shouldn’t need to manage infrastructure complexity. Most configurations are predefined through standardized templates, keeping teams focused on code and business logic rather than cluster tuning. Our challenge was that Kubernetes nodes appeared fully allocated while much of their capacity remained unused due to oversized resource requests. By introducing dynamic overcommit through our operator, we improved node utilization, reduced infrastructure usage for non critical workloads, and maintained full stability for critical services, all without requiring any changes from development teams.

At peak events large-scale Kubernetes environments must handle extreme load variations. Static resource requests often cause massive hidden capacity nodes to appear full while most CPU or memory remains unused. A single, uniform overcommit policy can’t safely account for differences between critical and non-critical workloads. This approach allows us to reserve more resources for the most critical applications while still improving overall cluster utilization.

We built k8s-overcommit Operator, an open-source Kubernetes operator that automates fine-grained overcommit control. It defines reusable Overcommit Classes through Custom Resource Definitions (CRDs) and enforces safety with validating admission webhooks. This enables consistent, policy-driven overcommit across workloads without manual configuration.

These policies can be applied at the pod level, namespace level, or set as a default for the entire cluster.

Designed to be vendor-agnostic, the operator runs seamlessly on any Kubernetes distribution. It helps organizations to improve node density, and reduce cost while preserving the stability of critical workloads.