Workflow Provenance: from Modeling to Reporting

Workflow provenance is a crucial part of a workflow system as it enables data lineage analysis, error tracking, workflow monitoring, usage pattern discovery, and so on. Integrating provenance into a workflow system or modifying a workflow system to capture or analyze different provenance information is burdensome, requiring extensive development because provenance mechanisms rely heavily on the modelling, architecture, and design of the workflow system. Various tools and technologies exist for logging events in a software system. Unfortunately, logging tools and technologies are not designed for capturing and analyzing provenance information. Workflow provenance is not only about logging, but also about retrieving workflow related information from logs. In this work, we propose a taxonomy of provenance questions and guided by these questions, we created a workflow programming model 'ProvMod' with a supporting run-time library to provide automated provenance and log analysis for any workflow system. The design and provenance mechanism of ProvMod is based on recommendations from prominent research and is easy to integrate into any workflow system. ProvMod offers Neo4j graph database support to manage semi-structured heterogeneous JSON logs. The log structure is adaptable to any NoSQL technology. For each provenance question in our taxonomy, ProvMod provides the answer with data visualization using Neo4j and the ELK Stack. Besides analyzing performance from various angles, we demonstrate the ease of integration by integrating ProvMod with Apache Taverna and evaluate ProvMod usability by engaging users. Finally, we present two Software Engineering research cases (clone detection and architecture extraction) where our proposed model ProvMod and provenance questions taxonomy can be applied to discover meaningful insights