"Roberts, Adam" <Adam.Roberts@echostar.com> writes:
> So, my main question is, is it reasonable to say that a trans id
> wraparound failure could create a situation in which you could
> use/manipulate user data tables if you refer to the data tables directly
> but if you tried to use a util (such as pgdump) or an internal psql
> query like \d or \df it would appear that you have no data tables?
Yeah, it is possible, because the system's internal catalog fetches use
SnapshotNow rules, which only look to see if a row's inserting/deleting
transaction(s) committed or not; they don't apply any comparison to the
current transaction ID. So what you've got is a situation where the
tables' pg_class rows have wrapped around and become invisible to SQL
queries, but the system's internal operations are still happy.
regards, tom lane