'gmetad' should throttle the frequency at which it prints some of its
error messages. In particular: "/usr/sbin/gmetad: data_thread()
got no answer from any [clustername] datasource". Two plausible
implementations are: (1) print this when a problem first occurs --
suppress printing until the operation is successful, and then print
again the next time a problem occurs; or (2) use exponential backoff on
the frequency at which the problem is reported, resetting the backoff
when the operation is successful.