via the -c and -E/-F options:

* On normal nodes, we used to use evproxy to capture experiment events
  on ops and forward them to the node local pubsubd (where all local
  agents connect to). Now we restart pubsubd with the -c and a -F option
  that says to subscribe to all events '(EXPT == "pid/eid")' at the
  parent pubsubd on ops (-c event-server).

* On shared hosts, we used to start kill pubsubd and start clusterd. Now
  we restart pubsubd with just "-c event-server" which causes it to
  subscribe to all events at the parent pubsubd on ops. This is how
  clusterd did it as well.

Another change is that instead of just kill pubsubd and starting with
new options, we now write a pubsubd.conf file into /usr/local/etc. The
rc.d startup files look for this file and source it as a typical .conf
file and pass the options to pubsubd.

I added deletion of these file to the prepare script.