Martijn van Oosterhout <kleptog@svana.org> writes:
> The only issue I can think of is that constraints might be violated
> (duplicate keys in unique index) because one of the records might have
> been invisible when the second was created...
More generally, application-driven updates of derived data might be
wrong because they omitted consideration of data that had become
invisible.
This might be a good time to press your buddy to move to 8.1 ;-)
PG 8.1 contains logic that should positively prevent a wraparound, by
shutting down the server if wraparound gets too close.
regards, tom lane