Date: Wed, 22 Mar 2023 17:18:28 -0500
From: Justin Pryzby <pryzby@telsasoft.com>
To: Michael Paquier <michael@paquier.xyz>
Cc: Kyotaro Horiguchi <horikyota.ntt@gmail.com>, sfrost@snowman.net,
	alvherre@alvh.no-ip.org, andres@anarazel.de,
	nathandbossart@gmail.com, jchampion@timescale.com,
	john.naylor@enterprisedb.com, masao.fujii@oss.nttdata.com,
	noriyoshi.shinoda@hpe.com, pgsql-hackers@postgresql.org,
	rjuju123@gmail.com, sawada.mshk@gmail.com, tgl@sss.pgh.pa.us,
	thomas.munro@gmail.com
Subject: Re: Improve logging when using Huge Pages
Message-ID: <ZBt+tCS/Akj8BvHi@telsasoft.com>
References: <ZBfsbTEUnRcEyNao@paquier.xyz>
 <ZBplvQx7n0MGVr6O@telsasoft.com>
 <ZBqwHY7yZQATAp9N@paquier.xyz>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <ZBqwHY7yZQATAp9N@paquier.xyz>
Archived-At: 
 <https://www.postgresql.org/message-id/ZBt%2BtCS/Akj8BvHi%40telsasoft.com>
Precedence: bulk

On Wed, Mar 22, 2023 at 04:37:01PM +0900, Michael Paquier wrote:
> I have noted something..  For the WIN32 case, we have that:
> 
> +++ b/src/backend/port/win32_shmem.c
> @@ -327,6 +327,8 @@ retry:
>             Sleep(1000);
>             continue;
>         }
> +
> +       huge_pages_active = ((flProtect & SEC_LARGE_PAGES) != 0);
>         break;
> 
> Are you sure that this is correct?  This is set in
> PGSharedMemoryCreate(), part of CreateSharedMemoryAndSemaphores() in
> the startup sequence that creates the shmem segment.  However, for a
> normal backend created by EXEC_BACKEND, SubPostmasterMain() reattaches
> to an existing shared memory segment, so we don't go through the
> creation path that would set huge_pages_active for the process just
> started, (note that InitPostmasterChild() switches IsUnderPostmaster,
> bypassing the shmem segment creation).

Wow, good point.  I think to make it work we'd need put
huge_pages_active into BackendParameters and handle it in
save_backend_variables().  If so, that'd be strong argument for using a
GUC, which already has all the necessary infrastructure for exposing the
server's state.

-- 
Justin