This bug is motivated by post-mortem findings at http://requiem/doc/postmortem103237

The puppet sources make it hard to recognize the actual roles
and uses of the two read-only database replicas.  Moreover, the
usage as it existed prior to the failure made every database server
a single point of failure.

The issues can be addressed with these changes in the puppet sources:
 A) Designate one of the slaves as the "primary slave".  Designate the
    other as the "backup slave."
 B) Use that terminology consistently throughout the puppet sources.
 C) Assign both the RPC and CI archiver services to the primary slave.
 D) Make clear in comments what services are assigned to the primary
    slave.  Also make clear that the primary slave is a mission-critical
    single point of failure.

Step C) has already been done in order to address bug 869754.  The other
three steps are needed to ratify the status quo, and make sure there's
no confusion if we need to change things (or just debug a failure).