On Sat, Sep 26, 2009 at 1:23 PM, Gerhard Wiesinger <lists@wiesinger.com> wrote:
> On Sat, 26 Sep 2009, Tom Lane wrote:
>
>> Gerhard Wiesinger <lists@wiesinger.com> writes:
>>>
>>> Research on the pg_bufferache showed that a lot of pages don't have any
>>> assigned relfilenodesin pg_class, even when they are dirty (in this case
>>> inserts are done).
>>
>> There should never be any buffered pages for nonexistent tables. The
>> method you are using will only show relnames for pages belonging to
>> tables of the current database ... maybe there are a lot of dirty pages
>> for other databases in the cluster? Or maybe the tables are newly
>> created (not committed) and so you can't see their pg_class rows yet?
>>
>
> OK, this is a visibility problem of the databases. Is it possible to GRANT
> visibility for e.g. one user to all databases for pg_class?
It's not a question of granting it, it's just not there in the other
databases. you could join on a dblink'd result set I guess.