Re: pg_class.relnamespace NOT IN pg_namespace.oid

Ireneusz Pluta <ipluta(at)wp(dot)pl> writes:
> W dniu 2012-02-27 21:59, Tom Lane pisze:
>> I'd do the latter I think. Keep in mind that there are probably also
>> entries in pg_depend linking the tables to the schemas. If your goal is
>> only to get to a clean dumpable state and then dump and reload the
>> database, you probably don't need to worry about fixing pg_depend.
>> However, if you intend to keep on using the database without a reload,
>> it'd be prudent to make sure pg_depend is straightened out as well.

> I found rows of missing namespace oid in pg_depend.refobjid. I understand that I update them to oid
> of a newly created empty schema?

Right.

> But, as in the menatime I played with the case separately on a test database, I found also pg_type
> entries need fixing. I was not aware of pg_depend, but found that after only tweaking both pg_class,
> and pg_type the database got dumpable.

Hm. We've seen occasional reports of this sort of behavior (that is,
DROP of a schema failing to cascade to all the contained objects) but
never been able to reproduce it. If you do see it happen again, and
can work out a scenario that causes it (even only intermittently)
we'd love to have a test case.

One possible theory for cascaded drops to fail like that is that the
indexes on pg_depend are corrupt, so you might want to consider
REINDEXing that catalog, just in case.

regards, tom lane