Lessons Learned Building Kubernetes Controllers • Dave Cheney • YOW! 2018

The speaker explains Kubernetes as an open-source system for automating containerized application deployment, scaling, and management, acknowledging its perceived complexity. He breaks down Kubernetes into four layers: a replicated data store (like etcd), an API server, controllers and operators that manage state, and a container runtime. The focus then shifts to ingress controllers, which manage external traffic to applications within a Kubernetes cluster, providing services like HTTP/TLS management, load balancing, and reverse proxying.

Ingress controllers offer key benefits such as traffic consolidation to reduce the number of public IPs, simplified TLS management, abstract configuration for web applications, and path-based routing. The speaker introduces Contour as an ingress controller that utilizes Envoy proxy as its data plane, highlighting Envoy's dynamic configuration capabilities as a strong match for Kubernetes' declarative API. Contour acts as a translator, converting Kubernetes REST API information into the format Envoy requires, enabling real-time configuration updates without reloads.

The discussion delves into the software engineering aspects of building Contour. Key points include using Go as the programming language, maintaining a healthy code-to-test ratio (around 1:2.5 for Contour), keeping the main function concise by offloading business logic, and structuring packages effectively by naming them for their functionality. The use of an internal directory for non-exportable code is recommended to prevent unintended dependencies. The talk also covers Go routine management, dependency management using dep, and integrating Docker into the development workflow. Finally, the speaker advocates for optimizing the developer inner loop by avoiding frequent Docker builds and pushes, favoring faster iteration cycles and robust, non-flaky functional testing.

Kubernetes Architecture Overview [1:52]

• Kubernetes is described as an open-source system for automating deployment, scaling, and management of containerized applications.
• The perceived complexity of Kubernetes is a matter of perspective, similar to the underlying complexity of simple devices.
• Kubernetes can be understood in four layers:
  • Replicated data store: Defaults to etcd; considered a "live hand grenade" for operators due to its stateful nature.
  • API server: Sits on top of the data store, handles authentication, schema validation, CRUD operations, and crucially, "watch" capabilities.
  • Controllers and operators: Watch the API server and strive to match the current state of the world to the desired state in the database. The API server can be modeled as a message bus.
  • Container runtime: The component that actually runs containerized applications, with Docker being a common default.

This breakdown provides a foundational understanding of Kubernetes' core components and their interactions, essential for developers and operators.

"Kubernetes is an open-source system for automating deployment, scaling, and management of containerized applications."

Related Video Summary

• Secrets to Finally Getting a DevOps Job