Collaborative data-driven workflows: think global, act local

Abstract

We introduce and study a model of collaborative data-driven workflows. In a local-as-view style, each peer has a partial view of a global instance that remains purely virtual. Local updates have side effects on other peers' data, defined via the global instance. We also assume that the peers provide (an abstraction of) their specifications, so that each peer can actually see and reason on the specification of the entire system. We study the ability of a peer to carry out runtime reasoning about the global run of the system, and in particular about actions of other peers, based on its own local observations. A main contribution is to show that, under a reasonable restriction (namely, key-visibility), one can construct a finite symbolic representation of the infinite set of global runs consistent with given local observations. Using the symbolic representation, we show that we can evaluate in PSPACE a large class of properties over global runs, expressed in an extension of first-order logic with past linear-time temporal operators, PLTL-FO. We also provide a variant of the algorithm allowing to incrementally monitor a statically defined property, and then develop an extension allowing to monitor an infinite class of properties sharing the same temporal structure, defined dynamically as the run unfolds. Finally, we consider an extension of the language, that permits workflow control with PLTL-FO formulas. We prove that this does not increase the power of the workflow specification language, thereby showing that the language is closed under such introspective reasoning.