How we host homogeneous scalable Kubernetes control planes across multiple cloud providers using Gardener

Gardener is an SAP-pioneered open-source project that was born to solve the common problems in Kubernetes control-plane management. Gardener provides Kubernetes clusters-as-a-service with minimal TCO, and infrastructure pervasiveness across a wide range of cloud providers such as AWS, GCP, Azure and Alibaba cloud.

Gardener is a standard Kubernetes extension and adheres to the same concepts by design. It adds a custom API server, a controller-manager, and a scheduler to create and manage the lifecycle of Kubernetes clusters. It extends the Kubernetes API with custom resources, most prominently the Gardener cluster specification (a Shoot resource in the Gardener world), that can be used to create and manage a Kubernetes cluster in a declarative way.

This inception design (kubeception) enables the control plane to be deployed as a native Kubernetes workload into separate clusters (a.k.a Seeds), reducing the need for dedicated master VMs. This approach not only lowers the total cost of ownership but also leverages mature Kubernetes features for robust and simplified day-2 operations, such as cluster updates and maintenance.

Today, we host over 6000 clusters across our landscapes spanning multiple providers, utilizing over 32k nodes, 450k core CPUs and a massive 2.5 Petabytes of memory. In this talk, we describe how we achieved this incredible feat and how we envision the future of this project.

You can read more about Gardener at https://gardener.cloud and https://github.com/gardener/gardener.