COMPASS Production System Overview

Migration of COMPASS data processing to Grid environment has started in 2015 from a small prototype, deployed on a single virtual machine. Since summer of 2017, the system works in production mode, distributing jobs to two traditional Grid sites: CERN and JINR. Now the infrastructure of COMPASS Grid Production System includes 6 virtual machines, each is reserved for one production service: database, PanDA, Auto Pilot Factory, Monitoring, CRIC information system and, finally, production system (ProdSys) management instance, which provides a user interface for production manager and hosts services of automatic processing. Support of COMPASS virtual organization is provided by CERN IT. CRIC is also deployed at CERN Cloud Service. Other ProdSys services are deployed at JINR Cloud Service. There are two storage elements at CERN: EOS for short-term storage and Castor for long-term storage. During last year, along with providing a 24/7 service, the system was instrumented by many features, which allow automating data processing as much as possible. Recently, Blue Waters HPC has become a member of the computing infrastructure of the experiment. Details of implementation, workflow management, and infrastructure overview are presented in this article.