Hi,
On 2020-03-06 16:33:18 -0300, Alvaro Herrera wrote:
> On 2020-Mar-06, Michael Paquier wrote:
> > On Thu, Mar 05, 2020 at 09:40:54AM -0500, Robert Haas wrote:
> > > Seems reasonable, but it would be better if people proposed these
> > > kinds of changes closer to the beginning of the release cycle rather
> > > than in the crush at the end.
> >
> > +1, to both points.
>
> Why? Are you saying that there's some actual risk of breaking
> something? We're not even near beta or feature freeze yet.
>
> I'm not seeing the reason for the "please propose this sooner in the
> cycle" argument. It has already been proposed sooner -- seven years
> sooner. We're not waiting for users to complain anymore; clearly nobody
> cared.
Yea. There are changes that are so invasive that it's useful to go very
early, but in this case I'm not seeing it?
+1 for removing non-fast promotions.
FWIW, I find "fallback promotion" a confusing description.
Btw, I'd really like to make the crash recovery environment more like
the replication environment. I.e. have checkpointer, bgwriter running,
and have an 'end-of-recovery' record instead of a checkpoint at the end.
Greetings,
Andres Freund