Running Large-Scale Scheduling Simulations with Virtual Kubelet - David Morrison

Don't miss out! Join us at our next Flagship Conference: KubeCon + CloudNativeCon Europe in Paris from March 19-22, 2024. Connect with our current graduated, incubating, and sandbox projects as the community gathers to further the education and advancement of cloud native computing. Learn more at kubecon.io
Running Large-Scale Scheduling Simulations with Virtual Kubelet - David Morrison, Applied Computing Research Labs
In this talk, we show how you can simulate scheduling in a large Kubernetes cluster without breaking the bank. Pod scheduling is one of the trickiest parts of Kubernetes infrastructure to get right. Done correctly, it can have enormous benefits in terms of cost and efficiency; done poorly, it can cause outages or lasting damage to your bottom line. The Kubernetes scheduler comes with a host of parameters that you can tune to control its behaviour, as well as allowing users to inject custom plugins at almost every stage of the scheduling process. In a large-scale cluster, the interactions between all these knobs can lead to unexpected emergent behaviour that is nearly impossible to reason about. To simulate this, we use a virtual kubelet that presents itself as a fully-configurable, low-overhead Kubernetes node object. In this presentation we show how to use this technique to collect data about scheduling decisions that are made for a variety of different kube-scheduler parameters.