Date: Mon, 8 Jun 2026 14:21:59 +0000
From: Bertrand Drouvot <bertranddrouvot.pg@gmail.com>
To: Xuneng Zhou <xunengzhou@gmail.com>
Cc: "Hayato Kuroda (Fujitsu)" <kuroda.hayato@fujitsu.com>,
	Alexander Lakhin <exclusion@gmail.com>,
	pgsql-hackers <pgsql-hackers@postgresql.org>
Subject: Re: t/035_standby_logical_decoding.pl might fail on attempt to read
 wrong timeline
Message-ID: <aibQB2f9h3nYeg/2@bdtpg>
References: <7daef094-abf3-4672-bc23-3df4763b16a3@gmail.com>
 <aiP/X1FThlZRCFiN@bdtpg>
 <OS9PR01MB1214908BA67A7811BD6281208F51C2@OS9PR01MB12149.jpnprd01.prod.outlook.com>
 <CABPTF7WmLKOJcSPod7zT2hynUFQcUs0VoQyR-p=XFsafvmGm7g@mail.gmail.com>
 <CABPTF7U_B9pdC563fFONQLX_FXCZtxZgM+jN9snzMVg9b9MRfg@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: 
 <CABPTF7U_B9pdC563fFONQLX_FXCZtxZgM+jN9snzMVg9b9MRfg@mail.gmail.com>
Archived-At: <https://www.postgresql.org/message-id/aibQB2f9h3nYeg/2%40bdtpg>
Precedence: bulk

Hi,

On Mon, Jun 08, 2026 at 09:00:17PM +0800, Xuneng Zhou wrote:
> I tweaked the reproducer based on the theory outlined above. The main
> changes from the original reproducer are:
> 
> 1) blocks at logical-walsender-after-slot-acquire in walsender.c,
> before the decoding context is created and before the reader starts
> from restart_lsn, matching the delay set by Alexander

> 2) Forces the first read to occur during promotion. It inserts rows
> 1..4, waits for replay, starts promotion with pg_promote(false), holds
> startup at startup-logical-decoding-status-change-end-of-recovery,
> then wakes the walsender.

Yeah, so this existing startup-logical-decoding-status-change-end-of-recovery
injection point already exists in the code tree and is also called after 
CleanupAfterArchiveRecovery() and before RECOVERY_STATE_DONE change in
StartupXLOG().

So this is the same window as the new injection point that was added in the new
tests in v1-0002 shared up-thread [1].

That said, I think I prefer the v1-0002 shared up-thread [1] approach for the
tests as:

- the injection point name clearly describes the tested condition
- no new injection point is added in walsender.c (it pauses startup mid-promotion
and lets the walsender connect)
- the test relies on one injection point (not two)

[1]: https://postgr.es/m/aiaBtENl7tTf2MM8%40bdtpg

Regards,

-- 
Bertrand Drouvot
PostgreSQL Contributors Team
RDS Open Source Databases
Amazon Web Services: https://aws.amazon.com