Kubesprawl is one of the big pain points for organisations operating Kubernetes. The emerging pattern of having Kubernetes clusters per team or even per developer, along with separate environments for dev, test, staging and production, can mean that operations teams are dealing with many 10’s or even 100’s of Kubernetes clusters. Aside from the complexities of deployment, lifecycle management and automation, this can lead to significant expense, in both virtual and physical environments. In this talk we’ll introduce how we solved that problem in DC/OS, by bin packing multiple Kubernetes clusters onto the same underlying infrastructure to build high density infrastructure. This approach has a number of interesting problems in itself, including handling complex isolation and networking, and we’ll deep dive into how this works.
Proliferation of Kubernetes clusters is a big problem for a lot of operators, both from a complexity perspective and from a cost perspective. Achieving high compute utilisation and packing multiple Kubernetes clusters onto the same infrastructure can reduce both cost and complexity significantly.