Re: GetOldestXmin going backwards is dangerous after all

On 2013-02-04 17:21:50 +0000, Simon Riggs wrote:
> On 4 February 2013 17:02, Andres Freund <andres@2ndquadrant.com> wrote:
> 
> > I unfortunately don't yet see a robust way without storing the last used
> > horizon :(.
> 
> We can't go backwards, but we can go forwards.
> 
> We can move the next xid forwards by an amount equal to the increase
> in vacuum_defer_cleanup_age.

Don't think that helps, the problem is not new writes but already
removed rows in the toast table.

Besides, advancing the next xid by vacuum_defer_cleanup_age every
restart could get expensive. Unless we find a space to store the old
value which we haven't really got in the back branches (control file?)...

Greetings,

Andres Freund

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