Deploying on Kubernetes can be likened to flying with an alligator - it's manageable as long as the rules are followed, but things get complicated when scaling a highly available stateful system. This article discusses the challenges of scaling a CockroachDB Dedicated cluster using Kubernetes Statefulsets and how they were overcome by writing a custom Kubernetes scheduler plugin. The key points include understanding the constraints of Kubernetes' Statefulset, handling data redistribution and replication in stateful systems like databases, and ensuring that resources are distributed across multiple failure domains for high availability.