"James Farrugia" <james.farrugia@gmail.com> writes:
> I wonder whether any of you can help me out with this problem.
What PG version is this?
> To get vacuum the TOAST object we created a temporary table foo (col1
> char(1)) and assigned its reltoastrelid (up till now set to 0) to
> pg_toast_35027430's OID and then vacuumed foo. The plan worked and
> immediately age(datfrozenxid) in pg_database reflected a much younger XID.
> We then decided to get rid of pg_toast_35027430 by dropping foo. Foo
> disappeared but pg_toast_35027430 persisted.
Well, yeah, because you didn't create a pg_depend link.
We have seen a couple prior reports of toast tables not going away when
their parent was dropped, but nobody's been able to create a
reproducible case yet. The most likely idea is probably that pg_depend
got corrupted somehow, causing the toast table not to get found by DROP.
Can you find any rows in pg_depend having objid equal to the OID of
pg_toast_35027430? Try reindexing pg_depend and then see if you find
any.
regards, tom lane