As I understand it, the main benefit of Kubernetes is the ability to pack multiple applications in a single VM. This way, one uses fewer paid resources to deplo

As I understand it, the main benefit of Kubernetes is the ability to pack multiple applications in a single VM. This way, one uses fewer paid resources to deploy a service mesh. In return for actually being able to show that you are not wasting any RAM or IO, you give up isolation. The "declarative" part of Kubernetes control buried in YAML seems to fail to live up to that label. Is there any literature that shows that the cost savings from packing more than one service per VM are significant enough to outweigh the cost of the messes created by middling developers who have trouble reasoning about the functioning of single process application at non-Google companies create when they think they should do what Google does?