Amazon Web Services (AWS) introduced an advanced control plane configuration for its Kubernetes platform.
According to the hyperscaler, the changes to Amazon Elastic Kubernetes Service (EKS) give users direct control over selected settings in the Kubernetes API server, scheduler, and controller manager without requiring them to operate customized control-plane components themselves.
Designed to work cluster-wide as opposed to per workload, the feature could aid organizations operating high-churn workloads such as batch processing, AI compute, and continuous integration and continuous delivery (CI/CD). It also benefits customers migrating Kubernetes environments that rely on heavily-tuned scheduler settings.
AWS said these use cases previously saw some users needing to build and maintain bespoke solutions, adding operational and maintenance overhead.
Under the Amazon hood #
With the EKS changes, users can change four main control-plane parameters covering pod placement, autoscaling responsiveness, Kubernetes event retention, and NodePort allocation. The scheduler can now be switched between a position which favors nodes with more spare resources, and one which preferentially places Kubernetes units, known as pods, on nodes such as physical servers or virtual machines (VMs) that are already more heavily allocated. The latter helps concentrate workloads onto fewer nodes, leaving less partially idle instances and potentially reducing compute costs. Customers can also give different weights to CPU, memory, and accelerator resources.
A controller-manager configuration meanwhile adds control over the Horizontal Pod Autoscaler (HPA) synchronization period. A shorter period causes the HPA to evaluate metrics and scaling decisions more frequently, enabling workloads to respond faster to demand increases.
AWS explained that Kubernetes normally retains events for an hour, but workloads that continuously create and destroy pods can consume storage while slowing API-server list operations. With updates to the Kubernetes API server, customers can configure deployment time periods with supported values between 10 and 60 minutes. Reducing the retention period here potentially clears short-lived diagnostic data sooner, improving overall compute performance. Longer retention meanwhile provides administrators with more historical debugging information.
Amazon is also letting users change the range of ports Kubernetes can use for NodePort services, which let outside traffic reach an application running inside Kubernetes. The configuration capability can help companies match existing firewall and network rules, support more NodePort services, and migrate older applications that depend on specific port numbers without having to modify the applications or add a proxy.
The Advanced Kubernetes control plane configuration is available for Amazon EKS clusters running Kubernetes 1.31 or later across all commercial AWS Regions, AWS GovCloud (U.S.), and AWS China Regions in which EKS is available. AWS is providing the feature itself at no additional charge, although customers using parameters such as the configurable HPA synchronization period must use the separately billed EKS Provisioned Control Plane tier.