* SL Baur <steve@beopen.com> [000524 02:59] wrote:
> Alfred Perlstein <bright@wintelcom.net> writes in pgsql-hackers@postgresql.org:
>
> > while doing a pg_dump of a table after postgresql made a mess of itself:
>
> > dumpRules(): SELECT failed for table website. Explanation from
> > backend: 'ERROR: cache lookup of attribute 1 in relation 9892634
> > failed '.
>
> I just got a message like that earlier this afternoon. My problem was
> that I had created a view and later dropped and recreated one of the
> tables the view referenced. Dropping and recreating the view fixed
> things.
I'm not using views afaik.
There seems to be a serious corruption problem when a transaction
is aborted, I'll see if I can have a reproduceable bug report
tomorrow.
Afaik it has to do with a transaction aborting after inserting or
updating into a table. Something seems to go seriously wrong.
We're also getting some problems when we don't "SET ENABLE_SEQSCAN=OFF;"
for certain queries, Postgresql takes a really unoptimal path and
will loop forever.
Btw, is there any way to specify an abort timeout for a query if it's
taking too long to just fail?
--
-Alfred Perlstein - [bright@wintelcom.net|alfred@freebsd.org]
"I have the heart of a child; I keep it in a jar on my desk."
