Cluster autoscalers automatically adjust the number of nodes in your cluster without operator intervention. Scaling up is important to meet resource demand while scaling down is useful for cost savings when nodes are underutilized. Within the Kubernetes ecosystem, there are various methods for achieving autoscaling.

We’ll do a deep dive into Cerebral, an open source, provider agnostic, preemptive autoscaler with pluggable metrics backends and autoscaling engines. We’ll demo how to use it with a standard backend such as Prometheus, as well as explore use cases for custom backends: scaling based on a growing application queue, for example. This methodology will be contrasted with others such as the Kubernetes Cluster Autoscaler. The audience will also leave with an understanding of different scaling strategies: random, least waste, cost-optimized, and so on.