Apache Spark has become the de-facto standard for writing big data processing pipelines. While the business logic of Spark applications is often at least as complex as what we have been dealing with in a pre-big data world, enabling developers to write comprehensive, fast unit test suites has not been a priority in the design of Spark. The main problem is that you cannot test your code without at least running a local SparkContext. These tests are not really unit tests, and they are too slow for pursuing a test-driven development approach.
In this talk, I will introduce thekontextfrei library, which aims to liberate you from the chains of the SparkContext. I will show how it helps restoring the fast feedback loop we are taking for granted. In addition, I will explain how kontextfrei is implemented and discuss some of the design decisions made and look at alternative approaches and current limitations.