Jan Wieck <JanWieck@Yahoo.com> writes:
> The fact that the backend cannot open a segment of a TOAST table raises
> the question "how did it get removed?".
I am guessing that the segment never existed, and that the real problem
is corruption of an entry in the TOAST table's index. A bad record
number in an index tuple's heap pointer could produce this symptom.
It might be worth dumping the TOAST index with pg_filedump or some such
tool and looking for silly block numbers. Examining the pattern of
corruption on the bad page, once found, might give us some hint how it
happened.
regards, tom lane