Re: New vacuum option to do only freezing - Mailing list pgsql-hackers

From Masahiko Sawada
Subject Re: New vacuum option to do only freezing
Date
Msg-id CAD21AoDu0wHfBmgJSQdH4Yn1vJingqgdEQ4bUioJf8v=RQ9K=g@mail.gmail.com
Whole thread Raw
In response to Re: New vacuum option to do only freezing  (Kyotaro HORIGUCHI <horiguchi.kyotaro@lab.ntt.co.jp>)
Responses Re: New vacuum option to do only freezing
List pgsql-hackers
On Thu, Apr 4, 2019 at 9:18 AM Kyotaro HORIGUCHI
<horiguchi.kyotaro@lab.ntt.co.jp> wrote:
>
> Hello.
>
> At Wed, 3 Apr 2019 11:55:00 -0400, Robert Haas <robertmhaas@gmail.com> wrote in
<CA+Tgmoas581jpJ0TPaA38OhjXHgbLy8z1fuuHH7CaNkrboZJeA@mail.gmail.com>
> > On Wed, Apr 3, 2019 at 1:32 AM Masahiko Sawada <sawada.mshk@gmail.com> wrote:
> > > Attached the updated version patches including the
> > > DISABLE_PAGE_SKIPPING part (0003).
> >
> > I am confused about nleft_dead_tuples.  It looks like it gets
> > incremented whenever we set tupgone = true, regardless of whether we
> > are doing index cleanup.  But if we ARE doing index cleanup then the
> > dead tuple will not be left.  And if we are not doing index vacuum
> > then we still don't need this for anything, because tups_vacuumed is
> > counting the same thing.  I may be confused.  But if I'm not, then I
> > think this should just be ripped out, and we should only keep
> > nleft_dead_itemids.
>
> tups_vacuumed is including heap_page_prune()ed tuples, which
> aren't counted as "tupgone".

Yes. tup_vacuumed counts not only HOT pruned tuples but also tuples
that became dead after heap_page_prune(). When index clenaup is
disabled, the former leaves only itemid whereas the latter leaves
itemid and heap tuple as we don't remove. nleft_dead_tuples counts
only the latter to report precisely. I think nleft_dead_tuples should
be incremented only when index cleanup is disabled, and the that part
comment should be polished.

>
> > As far as VacOptTernaryValue, I think it would be safer to change this
> > so that VACOPT_TERNARY_DEFAULT = 0.  That way palloc0 will fill in the
> > value that people are likely to want by default, which makes it less
> > likely that people will accidentally write future code that doesn't
> > clean up indexes.
>
> It's convincing. My compalint was enabled=0 and disabled=1 is
> confusing so I'm fine with default=0, disabled=1, enabled=2.

Okay, fixed.

Attached the updated version patch.

Regards,

--
Masahiko Sawada
NIPPON TELEGRAPH AND TELEPHONE CORPORATION
NTT Open Source Software Center

Attachment

pgsql-hackers by date:

Previous
From: Andres Freund
Date:
Subject: Re: allow online change primary_conninfo
Next
From: Amit Langote
Date:
Subject: Re: Inadequate executor locking of indexes