Feedback-ID: ia2694551:Fastmail
Date: Wed, 19 Jun 2024 19:21:25 +0200
From: Alvaro Herrera <alvherre@alvh.no-ip.org>
To: "Anton A. Melnikov" <a.melnikov@postgrespro.ru>
Cc: PostgreSQL Hackers <pgsql-hackers@lists.postgresql.org>
Subject: Re: Don't process multi xmax in FreezeMultiXactId() if it is already
 marked as invalid.
Message-ID: <202406191721.pp2tjhvp7dsp@alvherre.pgsql>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <21ece835-b671-4656-9f35-008b9f355034@postgrespro.ru>
Archived-At: <https://www.postgresql.org/message-id/202406191721.pp2tjhvp7dsp%40alvherre.pgsql>
Precedence: bulk

On 2024-Jun-14, Anton A. Melnikov wrote:

> Hello!
> 
> The src/backend/access/heap/README.tuplock says about HEAP_XMAX_INVALID bit
> that "Any tuple with this bit set does not have a valid value stored in XMAX."
> 
> Found that FreezeMultiXactId() tries to process such an invalid multi xmax
> and may looks for an update xid in the pg_multixact for it.
> 
> Maybe not do this work in FreezeMultiXactId() and exit immediately if the
> bit HEAP_XMAX_INVALID was already set?
> 
> For instance, like that:
> 
> master
> @@ -6215,6 +6215,15 @@ FreezeMultiXactId(MultiXactId multi, uint16 t_infomask,
>         /* We should only be called in Multis */
>         Assert(t_infomask & HEAP_XMAX_IS_MULTI);
> +       /* Xmax is already marked as invalid */
> +       if (MultiXactIdIsValid(multi) &&
> +               (t_infomask & HEAP_XMAX_INVALID))

Hmm, but why are we calling FreezeMultiXactId at all if the
HEAP_XMAX_INVALID bit is set?  We shouldn't do that.  I think the fix
should appear in heap_prepare_freeze_tuple() to skip work completely if
HEAP_XMAX_INVALID is set.  Then in FreezeMultiXactId you could simply
Assert() that the given tuple does not have HEAP_XMAX_INVALID set.

-- 
Álvaro Herrera        Breisgau, Deutschland  —  https://www.EnterpriseDB.com/
"It takes less than 2 seconds to get to 78% complete; that's a good sign.
A few seconds later it's at 90%, but it seems to have stuck there.  Did
somebody make percentages logarithmic while I wasn't looking?"
                http://smylers.hates-software.com/2005/09/08/1995c749.html