CircuitFlow: A Domain Specific Language for Dataflow Programming (with appendices) (2111.12420v1)

Abstract: Dataflow applications, such as machine learning algorithms, can run for days, making it desirable to have assurances that they will work correctly. Current tools are not good enough: too often the interactions between tasks are not type-safe, leading to undesirable run-time errors. This paper presents a new declarative Haskell Embedded DSL (eDSL) for dataflow programming: CircuitFlow. Defined as a Symmetric Monoidal Preorder (SMP) on data that models dependencies in the workflow, it has a strong mathematical basis, refocusing on how data flows through an application, resulting in a more expressive solution that not only catches errors statically, but also achieves competitive run-time performance. In our preliminary evaluation, CircuitFlow outperforms the industry-leading Luigi library of Spotify by scaling better with the number of inputs. The innovative creation of CircuitFlow is also of note, exemplifying how to create a modular eDSL whose semantics necessitates effects, and where storing complex type information for program correctness is paramount.