Cherry Picking in Database Languages

摘要

To avoid expensive round-trips between the application layer and the database layer it is crucial that data-intensive processing and calculations happen close to where the data resides - ideally within the database engine. However, each application has its own domain and provides domain-specific languages (DSL) as a user interface to keep interactions confined within the well-known metaphors of the respective domain. Revealing the innards of the underlying data layer by forcing users to formulate problems in terms of a gen eral database language is often not an option. To bridge that gap, we propose an approach to transform and directly compile a DSL into a general database execution plan using graph transformations. We identify the commonalities and mismatches between different models and show which parts can be cherry-picked for direct translation. Finally, we argue that graph transformations can be used in general to translate a DSL into an executable plan for a database.