Em seg., 30 de mar. de 2020 às 18:14, Andres Freund <and...@anarazel.de>
escreveu:

> Hi,
>
> On 2020-03-30 14:10:29 -0700, Andres Freund wrote:
> > On 2020-03-30 17:08:01 -0300, Ranier Vilela wrote:
> > > Em seg., 30 de mar. de 2020 às 16:05, Andres Freund <
> and...@anarazel.de>
> > > escreveu:
> > >
> > > > Hi,
> > > >
> > > > On 2020-03-30 15:07:40 -0300, Ranier Vilela wrote:
> > > > > I'm not sure that the patch is 100% correct.
> > > >
> > > > This is *NOT* correct.
> > > >
> > > Anyway, the original source, still wrong.
> > > What is the use of testing PageIsNew (page) twice in a row, if nothing
> has
> > > changed.
> >
> > Yea, that can be reduced. It's pretty harmless though.
> >
> > We used to require a cleanup lock (which requires dropping the lock,
> > acquiring a cleanup lock - which allows for others to make the page be
> > not empty) before acting on the empty page in vacuum. That's why
> > PageIsNew() had to be checked again.

Well, this is what the patch does, promove reduced and continue to check
PageIsNew after unlock.

regards,
Ranier Vilela

Reply via email to