"Ed L." <pgadmin@bluepolka.net> writes:
> I've been seeing the same problem intermittently over the past 2 years
> among ~30 production clusters.
Hm, I suddenly have a theory. Do you guys use LISTEN/NOTIFY in the
databases that are exhibiting the problem? Is it possible that the
backends that aren't cooperating have not done anything since they
last received a NOTIFY? (That would imply that the connected frontend
didn't issue any SQL commands in response to the notification.)
regards, tom lane