A/B/n testing, or split testing, is a testing process by which user traffic is randomly distributed between two or more versions of an application (or application component). Business metrics are evaluated to identify a winning version — the version which results in the greater profit or business value. For example, a shopping application might use revenue and user engagement as business metrics.
We focus on A/B/n testing of backend services deployed in Kubernetes. For example, in the figure below, the frontend might be a Node.js based online store. It relies on a backend recommendation service to make product suggestions to users. We are interested in A/B/n testing multiple versions of the recommendation service. In the figure, we have two versions, v1 (the current or default version) and v2 (the candidate version).