Improving Security in the ATLAS PanDA System

Abstract

The security challenges faced by users of the grid are considerably different to those faced in previous environments. The adoption of pilot jobs systems by LHC experiments has mitigated many of the problems associated with the inhomogeneities found on the grid and has greatly improved job reliability; however, pilot jobs systems themselves must then address many security issues, including the execution of multiple users' code under a common 'grid' identity. In this paper we describe the improvements and evolution of the security model in the ATLAS PanDA (Production and Distributed Analysis) system. We describe the security in the PanDA server which is in place to ensure that only authorized members of the VO are allowed to submit work into the system and that jobs are properly audited and monitored. We discuss the security in place between the pilot code itself and the PanDA server, ensuring that only properly authenticated workload is delivered to the pilot for execution. When the code to be executed is from a 'normal' ATLAS user, as opposed to the production system or other privileged actor, then the pilot may use an EGEE developed identity switching tool called gLExec. This changes the grid proxy available to the job and also switches the UNIX user identity to protect the privileges of the pilot code proxy. We describe the problems in using this system and how they are overcome. Finally, we discuss security drills which have been run using PanDA and show how these improved our operational security procedures.