Re: Multixid hindsight design

On 2015-06-05 10:45:09 +0100, Simon Riggs wrote:
> On 1 June 2015 at 20:53, Thomas Munro <thomas(dot)munro(at)enterprisedb(dot)com> wrote:
>
> > On Tue, May 12, 2015 at 9:20 AM, Heikki Linnakangas <hlinnaka(at)iki(dot)fi>
> > wrote:
> > > The beauty of this would be that the TED entries can be zapped at
> > restart,
> > > just like pg_subtrans, and pg_multixact before 9.3. It doesn't need to be
> > > WAL-logged, and we are free to change its on-disk layout even in a minor
> > > release.
> >
> > What about prepared transactions? They can lock rows FOR SHARE that
> > survive server restarts.
> >
>
> Interesting comment. I'm not aware that we do.
>
> If we do support row locking that survives server restart, how did it work
> before 9.3?

Multixacts were persistent before 9.3 as well. A good number of the bugs
existed then as well, but their effect was much more limited. The
difference is that now multixacts don't just have to survive till the
last locker isn't running anymore (which was determined by a horizon),
but that they have to live till they're vacuumed away, since xmax might
be stored in the multixact.