Feedback-ID: i0fe9450f:Fastmail
Date: Sun, 24 Dec 2023 11:30:21 +0900
From: Michael Paquier <michael@paquier.xyz>
To: Robert Haas <robertmhaas@gmail.com>
Cc: "pgsql-hackers@postgresql.org" <pgsql-hackers@postgresql.org>
Subject: Re: pgsql: Prevent tuples to be marked as dead in subtransactions on
 standb
Message-ID: <ZYeXvRzLAUMo0oVK@paquier.xyz>
References: <E1rD5HX-009rzG-AO@gemulon.postgresql.org>
 <CA+TgmobQNViB1E+b4njpXxcYH-ZG7f_c_yOvAautU26neY8q1w@mail.gmail.com>
MIME-Version: 1.0
Content-Type: multipart/signed; micalg=pgp-sha512;
	protocol="application/pgp-signature"; boundary="OEjk6uY5XvMrahpU"
Content-Disposition: inline
In-Reply-To: 
 <CA+TgmobQNViB1E+b4njpXxcYH-ZG7f_c_yOvAautU26neY8q1w@mail.gmail.com>
Archived-At: 
 <https://www.postgresql.org/message-id/ZYeXvRzLAUMo0oVK%40paquier.xyz>
Precedence: bulk


--OEjk6uY5XvMrahpU
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline

On Tue, Dec 19, 2023 at 03:15:42PM -0500, Robert Haas wrote:
> I don't think this is a good commit message. It's totally unclear what
> it means, and when I opened up the diff to try to see what was
> changed, it looked nothing like what I expected.

(Not my intention to ignore you here, my head has been underwater for
some time.)

> I think a better message would have been something like
> "startedInRecovery flag must be propagated to subtransactions".

You are right, sorry about that.  Something like "Propagate properly
startedInRecovery in subtransactions started during recovery" would
have been better than what I used.

> And I
> think there should have been some analysis in the commit message or
> the comments within the commit itself of whether it was intended that
> this be propagated to subtransactions or not. It's hard to understand
> why the flag would have been placed in the TransactionState if it
> applied globally to the transaction and all subtransactions, but maybe
> that's what happened.

Alvaro has mentioned something like that on the original thread where
we could use comments when a transaction state is pushed to a
subtransaction to track better the fields used and/or not used.
Documenting more all that at the top of TransactionStateData is
something we should do.

> Instead of discussing that issue, your commit message focuses in the
> user-visible consequences, but in a sort of baffling way. The
> statement that "Dead tuples are ignored and are not marked as dead
> during recovery," for example, is clearly false on its face. If
> recovery didn't mark dead tuples as dead, it would be completely
> broken.

Rather than "dead" tuples, I implied "killed" tuples in this sentence.
Killed tuples are ignored during recovery.
--
Michael

--OEjk6uY5XvMrahpU
Content-Type: application/pgp-signature; name="signature.asc"

-----BEGIN PGP SIGNATURE-----
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=0HnE
-----END PGP SIGNATURE-----

--OEjk6uY5XvMrahpU--