The TeraGrid operates two high availability central aggregating MDS4 services.
Some of the content in these services needs to persist thru an aggregating MDS4
service restart, even if the downstream that originally provided the
information is down and not able to report newer information.

We've worked around this limitation by implemented the functionality external
to MDS by configuring upstream registration to not aggregate information,
coding central cron jobs that aggregate downstream data to files in a
directory, and configuring the aggregating MDS the contents of files on disk.

It would be helpful if MDS4 had a way to persist data (hard state) thru service
restarts.