On 2012-09-28 19:01, Tom Lane wrote:
> Looks like you had some data corruption from the spike. If you're
> lucky, it's just this one block of pg_attribute and you can reconstruct
> things. If not, there may be more problems ...
Looks like there are more problems unfortunally. See below.
>> Any point of running a vacuum full on the database or is that a bad idea?
>
> I wouldn't try that. It might be worth trying a REINDEX on
> pg_attribute. That will not fix the table damage but it will at least
> make sure the indexes are consistent with what's now in the table.
> After that, I'd suggest manually removing the broken view's pg_class
> row and then seeing if you can pg_dump with sane-looking results.
> If you can, a dump and restore of at least this database would be
> prudent.
I tried the reindex and then the manual delete of the view from
pg_class. That took care of the rows in pg_attribute but when I tried
to run pg_dump I now get this:
pg_dump: failed sanity check, parent table OID 16547 of pg_rewrite entry
OID 16550 not found
/Mikael