Scaling Your Kubernetes Clusters Without Going Broke

The benefits of running your containers in Kubernetes are clear: you get your slice of the resources available and can scale as much as you need, as long as Kubernetes has a slice to give you. 

What happens when your cluster doesn't have enough resources? Typically, clusters are configured with an autoscaler that works with your cloud provider to provision more nodes, but what type of node will be provisioned? Is it going to be an instance with 96 vCPUs when you only need a fraction of one? What is it going to cost?

In this talk, we will go over the various capabilities of  Karpenter to get just-in-time provisioning for the resources you need. We will also show you how flexible provisioning can be, allowing you to take advantage of a wide array of instance types and architecture while being optimized for cost by utilizing features of your cloud provider, such as reserved and spot instances.

This talk will be heavily geared towards sharing our experiences and how we leveraged Karpenter to provision thousands of nodes for burst traffic while dramatically reducing our cloud spend.