Content Addressable CRDs: Type Uniqueness Across Kubernetes Clusters

The presentation discusses the challenges of identifying types in a Kubernetes cluster and proposes a solution using package revisions and content addressability of OCI images.

• Identifiers of types in a Kubernetes cluster are mostly meaningless outside of a single physical cluster
• Types are not portable and can't be moved across clusters with a guarantee that the syntax has the same semantics
• Behavior is attached to types in both imperative and declarative programming models
• Package revisions and content addressability of OCI images can be used to expand the definition of types and include their behavior
• The demo shows the installation of a type using package revisions and dependencies

The presenter demonstrates the installation of a type using package revisions and dependencies in a local kind cluster. The type includes dependencies on provider AWS and provider Helm, and composes primitive types into a higher-level representation of an EKS cluster.

As the Kubernetes extensibility story has evolved over time, the system has grown to more closely resemble a programming interface. As a result, more and more projects have standardized on Kubernetes, utilizing Custom Resource Definitions to define the schemas for interacting with the functionality they offer. However, defining higher level types outside the context of an individual cluster is difficult due to the fact that the unique identifier of a type – its Group, Version, and Kind – are not globally universal. In this talk we’ll explore a proposal for content addressable storage of type definitions for Kubernetes, inspired by the OCI distribution specification. The focus will be on the ergonomics of the system, how it compares to existing packaging and installation mechanisms for Custom Resource Definitions, and what future use cases could be enabled by aligning on a shared standard. Live demos will illustrate the full workflow and serve as inspiration for subsequent collaboration.