Declarative rules for inferring fine-grained data provenance from scientific workflow execution traces

Fine-grained dependencies within scientific workflow provenance specify lineage relationships between a workflow result and the input data, intermediate data, and computation steps used in the result's derivation. This information is often needed to determine the quality and validity of scientific data, and as such, plays a key role in both provenance standardization efforts and provenance query frameworks. While most scientific workflow systems can record basic information concerning the execution of a workflow, they typically fall into one of three categories with respect to recording dependencies: (1) they rely on workflow computation steps to declare dependency relationships at runtime; (2) they impose implicit assumptions concerning dependency patterns from which dependencies are automatically inferred; or (3) they do not assert any dependency information at all. We present an alternative approach that decouples dependency inference from workflow systems and underlying execution traces. In particular, we present a high-level declarative language for expressing explicit dependency rules that can be applied (at any time) to workflow trace events to generate fine-grained dependency information. This approach not only makes provenance dependency rules explicit, but allows rules to be specified and refined by different users as needed. We present our dependency rule language and implementation that rewrites dependency rules into relational queries over underlying workflow traces. We also demonstrate the language using common types of dependency patterns found within scientific workflows.