Tom Lane wrote:
> Alvaro Herrera <alvherre@commandprompt.com> writes:
> > Well, if a transaction modifies a table in some way, even without
> > changing the data, should generate an unfreeze event, because it will
> > need to lock the table; for example AlterTable locks the affected
> > relation with AccessExclusiveLock. It's important for the
> > non-transactional change to the pg_class tuple be the very first in the
> > transaction, because otherwise the change could be lost; but other than
> > this, I don't think there's any problem.
>
> You can't guarantee that. Consider for instance manual updates to
> pg_class:
>
> BEGIN;
> UPDATE pg_class SET reltriggers = 0 WHERE relname = ...
> ... alter table contents ...
> COMMIT or ROLLBACK;
>
> I believe there are actually patterns like this in some pg_dump output.
> Will you hack every UPDATE operation to test whether it's changing
> pg_class and if so force an "unfreeze" operation before changing any
> row? No thanks :-(
Oh, true, I hadn't thought of direct updates to pg_class.
> >> I'm wondering if we need a second pg_class-derived catalog that carries
> >> just the nontransactional columns.
>
> > I hope we don't need to do this because ISTM it will be a very big change.
>
> (Yawn...) We've made far bigger changes than that. The important
> thing is to get it right.
Yeah, I know -- I've been involved in some of them. I hereby volunteer
to do it for 8.2 because I'd really like to see this patch in.
--
Alvaro Herrera http://www.CommandPrompt.com/
PostgreSQL Replication, Consulting, Custom Development, 24x7 support