From: Alvaro Herrera <alvherre@alvh.no-ip.org>
To: Amit Kapila <amit.kapila16@gmail.com>
To: Zhijie Hou (Fujitsu) <houzj.fnst@fujitsu.com>
Cc: Pg Hackers <pgsql-hackers@lists.postgresql.org>
Cc: Gabriele Bartolini <gabriele.bartolini@enterprisedb.com>
Subject: Re: crash with synchronized_standby_slots
Date: Tue, 3 Dec 2024 18:04:49 +0100
Message-ID: <202412031704.thc75krypvts@alvherre.pgsql> (raw)
In-Reply-To: <OS0PR01MB571695C993F2B1738C6358FA942A2@OS0PR01MB5716.jpnprd01.prod.outlook.com> <CAA4eK1JmXd8VsKkH2dpnw5un-GUHd3PjaGwZZW73TDVKrHvN=Q@mail.gmail.com>
On 2024-Nov-29, Amit Kapila wrote:
> I tried it on my Windows machine and noticed that ReplicationSlotCtl
> is NULL for syslogger, so the problem doesn't occur. The reason is
> that we don't attach to shared memory in syslogger, so ideally
> ReplicationSlotCtl should be NULL. Because we inherit everything
> through the fork for Linux systems and then later for processes that
> don't want to attach to shared memory, we call PGSharedMemoryDetach()
> from postmaster_child_launch(). The PGSharedMemoryDetach won't
> reinitialize the memory pointed to by ReplicationSlotCtl, so, it would
> be an invalid memory.
Heh, interesting. I'm not sure if we should try to do something about
invalid pointers being left around after shmem initialization. Also, is
this the first GUC check_hook that needs to take an LWLock?
Anyway, I have pushed this.
BTW it occurs to me that there might well be some sort of thundering
herd problem if every process needs to run the check_hook when a SIGHUP
is broadcast, and they'll all be waiting on that particular lwlock and
run the same validation locally again and again. I bet if you have a
few thousand backends (hi Jakub! [1]) it's problematic. Maybe we need a
different way to validate the GUC, but I don't know what that might be;
but doing the validation once and storing the result in shmem might be
better.
On 2024-Nov-29, Zhijie Hou (Fujitsu) wrote:
> I can also reproduce this bug and confirmed that the bug is fixed
> after applying the patch. In addition to the regression tests, I also
> manually tested the behavior of the postmaster, walsender, and user
> backend after reloading the configuration, and they all work as
> expected.
Many thanks for testing!
[1] https://postgr.es/m/CAKZiRmwrBjCbCJ433wV5zjvwt_OuY7BsVX12MBKiBu+eNZDm6g@mail.gmail.com
--
Álvaro Herrera PostgreSQL Developer — https://www.EnterpriseDB.com/
"I apologize for the confusion in my previous responses.
There appears to be an error." (ChatGPT)
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Reply to all the recipients from your mail client; the link fills in
the headers below: reply via email
Some clients, webmail especially, drop In-Reply-To and References from
a link, and the reply then starts a new thread: use the mbox below.
To: pgsql-hackers@postgresql.org
Cc: alvherre@alvh.no-ip.org, amit.kapila16@gmail.com, houzj.fnst@fujitsu.com, gabriele.bartolini@enterprisedb.com
Subject: Re: crash with synchronized_standby_slots
In-Reply-To: <202412031704.thc75krypvts@alvherre.pgsql>
References: <202412031704.thc75krypvts@alvherre.pgsql>
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox