The best practice for add-on facter modules is to put them in a separate directory from the facter installation itself, and to use RUBYLIB, LOAD_PATH, or FACTERLIB to load them (as mentioned in http://docs.puppetlabs.com/guides/custom_facts.html ).  The current placement does not allow for openshift_facts to be omitted or overridden, and it is critical for OpenShift Online to do so.  

More context on this: the reason this has come up is that openshift_facts duplicates the data collection done by one of our other cron jobs, but it only stores counts and throws away the actual mappings of gear uuid / uid / state that we need for our monitoring system.  Collecting this data multiple times is very expensive on a busy system.  We would prefer it if such a metadata cache were part of the core node codebase, but since it is not, we need to keep our own cache implementation for now.