Production Kubernetes

Chapter 13. Autoscaling

The ability to automatically scale workload capacity is one of the compelling benefits of cloud native systems. If you have applications that encounter significant changes in capacity demands, autoscaling can reduce costs and reduce engineering toil in managing those applications. Autoscaling is the process whereby we increase and decrease the capacity of our workloads without human intervention. This begins with leveraging metrics to provide an indicator for when application capacity should be scaled. It includes tuning settings that respond to those metrics. And it culminates in systems to actually expand and contract the resources available to an application to accommodate the work it must perform.

While autoscaling can provide wonderful benefits, it’s important to recognize when you should not employ autoscaling. Autoscaling introduces complexity into your application management. Besides initial setup, you will very likely need to revisit and tune the configuration of your autoscaling mechanisms. Therefore, if an application’s capacity demands do not change markedly, it may be perfectly acceptable to provision for the highest traffic volumes an app will handle. If your application load alters at predictable times, the manual effort to adjust capacity at those times may be trivial enough that investing in autoscaling may not be justified. As with virtually all technology, leverage them only when the long-term benefit outweighs the setup and maintenance ...