Plug-in Based Analysis Framework for LHC Post-Mortem Analysis

Plug-in based software architectures [1] are extensible, enforce modularity and allow several teams to work in parallel. But they have certain technical and organizational challenges, which we discuss in this paper. We gained our experience when developing the Post-Mortem Analysis (PMA) system, which is a mission critical system for the Large Hadron Collider (LHC). We used a plugin-based architecture with a general-purpose analysis engine, for which physicists and equipment experts code plugins containing the analysis algorithms. We have over 45 analysis plugins developed by a dozen of domain experts. This paper focuses on the design challenges we faced in order to mitigate the risks of executing third-party code: assurance that even a badly written plugin doesn't perturb the work of the overall application; plugin execution control which allows to detect plugin misbehaviour and react; robust communication mechanism between plugins, diagnostics facilitation in case of plugin failure; testing of the plugins before integration into the application, etc.