ETLDiff: A Semi-automatic Framework for Regression Test of ETL Software

Abstract

Modern software development methods such as Extreme Programming (XP) favor the use of frequently repeated tests, so-called regression tests, to catch new errors when software is updated or tuned, by checking that the software still produces the right results for a reference input. Regression testing is also very valuable for Extract–Transform–Load (ETL) software, as ETL software tends to be very complex and error-prone. However, regression testing of ETL software is currently cumbersome and requires large manual efforts. In this paper, we describe a novel, easy–to–use, and efficient semi–automatic test framework for regression test of ETL software. By automatically analyzing the schema, the tool detects how tables are related, and uses this knowledge, along with optional user specifications, to determine exactly what data warehouse (DW) data should be identical across test ETL runs, leaving out change-prone values such as surrogate keys. The framework also provides tools for quickly detecting and displaying differences between the current ETL results and the reference results. In summary, manual work for test setup is reduced to a minimum, while still ensuring an efficient testing procedure.