Scaling Puppet and Foreman for HPC

ABSTRACTThe Ohio Supercomputer Center has deployed a Puppet configuration management and Foreman provisioning environment that scales to almost one thousand servers that are a mix of HPC cluster compute and service nodes as well as storage, web, and infrastructure servers. The motivation for using a common provisioning and configuration management environment for all computers at OSC was to minimize the unique environments with their own management tools, as well as to provide for unified management of network booting, DNS, and audit interfaces for change management. In this paper, we describe several novel methods used to scale Puppet to a large number of servers, as well as the use of the Foreman provisioning environment as a global key-value store. We also present our implementation of multiple environment configurations that allow us to support development and test environments in parallel and integrated with production configuration management.