Containerized Batch System Monitoring

Abstract

Running a batch system for grid jobs and for local users, we investigate a generic solution to monitor the resource usages of jobs. We extend a standard toolkit for monitoring container performance metrics also for non-containerized applications, so that we can make easy use of a popular industry solution. By concentrating on the basic kernel features also used by containers frameworks, we envisage to use the same tools on non-containerized batch systems as well as container orchestrators. For deployment, we encapsulate the tools as Singularity containers and distribute them via CVMFS.