Data Provenance for Recursive SQL Queries

The adoption of recursion in SQL-framed either in terms of recursive common table expressions (CTEs) or recursive user-defined functions (UDFs)-marked a jump in the expressivity of the query language. The resulting queries can perform complex computation close to database-resident data but, at the same time, often prove challenging to understand and debug. We build on earlier work on the derivation of where- and why-provenance for complex (yet non-recursive) SQL queries to also embrace recursive SQL CTEs and UDFs. Fine-grained data provenance for recursive SQL is derived through language-level query rewriting and a two-phase evaluation strategy that does not invade the underlying RDBMS.