Re: GetOldestXmin going backwards is dangerous after all

On 2013-02-04 11:52:05 -0500, Tom Lane wrote:
> Andres Freund <andres(at)2ndquadrant(dot)com> writes:
> > I absolutely hate to suggest it, but what about storing the last
> > vacuum's xmin horizon in the main table's pg_class.options in the back
> > branches?
>
> Not workable. This would require a non-in-place update of the table's
> pg_class row (at least in cases where the option wasn't stored already).
> Which would require VACUUM to have an XID, which would make a whole lot
> of assumptions fall over.

Don't get me wrong, I hate that solution, but that specific issue seems
relatively easily solvable by pre-creating the option intially in
vacuum_rel() in a separate transaction before entering
lazy_vacuum_rel().

I unfortunately don't yet see a robust way without storing the last used
horizon :(.

Greetings,

Andres Freund

--
Andres Freund http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services