Date: Mon, 14 Jul 2025 16:39:18 -0500
From: Nathan Bossart <nathandbossart@gmail.com>
To: Sami Imseih <samimseih@gmail.com>
Cc: Bertrand Drouvot <bertranddrouvot.pg@gmail.com>,
	pgsql-hackers <pgsql-hackers@postgresql.org>
Subject: Re: Improve LWLock tranche name visibility across backends
Message-ID: <aHV5BsKoSeOkIsNL@nathan>
References: 
 <CAA5RZ0vvED3naph8My8Szv6DL4AxOVK3eTPS0qXsaKi=bVdW2A@mail.gmail.com>
 <aG9/OKQRgPUCxmsX@ip-10-97-1-34.eu-west-3.compute.internal>
 <CAA5RZ0ui5+=Q3ouxbrjSKa1xJvOjDD2tNYfkG-UtJJNNc-4E=g@mail.gmail.com>
 <aHDYHKdEyZiV/Aob@ip-10-97-1-34.eu-west-3.compute.internal>
 <CAA5RZ0vK_EuHwGj8nWkb1UwYjDbqvj+ZyiYPYn3CRsU8uwKUsQ@mail.gmail.com>
 <aHGJBt37dg37f1bE@nathan>
 <CAA5RZ0uic7Us255mFPJSrMFH7UY-RR8YZcjBzNDzfE4gFREDGQ@mail.gmail.com>
 <aHVeqtsltQyBBvAM@nathan>
 <CAA5RZ0vb_CVR7SuCpHrLbV4yDQJ8o9WUudz51YRjUr4PLiDMYg@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: 
 <CAA5RZ0vb_CVR7SuCpHrLbV4yDQJ8o9WUudz51YRjUr4PLiDMYg@mail.gmail.com>
Archived-At: <https://www.postgresql.org/message-id/aHV5BsKoSeOkIsNL%40nathan>
Precedence: bulk

On Mon, Jul 14, 2025 at 03:45:02PM -0500, Sami Imseih wrote:
>> Ah, I missed the problem with postmaster.  Could we have the first backend
>> that needs to access the table be responsible for creating it and
>> populating it with the built-in/requested-at-startup entries?
> 
> We can certainly maintain a flag in the shared state that is set once
> the first backend loads all the tranches in shared memory. That did not
> cross my mind, but it feels wrong to offload such responsibility to a
> normal backend.

Well, we already need each backend to either initialize or attach to the
dshash table, and the initialization would only ever happen once on a
running server.  Adding a new initialization step to bootstrap the built-in
and registered-at-startup tranche names doesn't seem like that much of a
leap to me.

Another random thought: I worry that the dshash approach might be quite a
bit slower, and IIUC we just need to map an integer to a string.  Maybe we
should just use a DSA for LWLockTrancheNames.  IOW we'd leave it as a char
** but put it in shared memory.  

-- 
nathan