Date: Thu, 23 Jan 2025 10:38:37 -0800
From: Noah Misch <noah@leadboat.com>
To: Tom Lane <tgl@sss.pgh.pa.us>
Cc: pgsql-hackers@lists.postgresql.org
Subject: Re: "postmaster became multithreaded" is reachable
Message-ID: <20250123183837.79.nmisch@google.com>
References: <206317.1737656533@sss.pgh.pa.us>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <206317.1737656533@sss.pgh.pa.us>
User-Agent: Mutt/2.2.12 (2023-09-09)
Archived-At: 
 <https://www.postgresql.org/message-id/20250123183837.79.nmisch%40google.com>
Precedence: bulk

On Thu, Jan 23, 2025 at 01:22:13PM -0500, Tom Lane wrote:
> ExitPostmaster() says:
> 
>     /*
>      * There is no known cause for a postmaster to become multithreaded after
>      * startup.  Recheck to account for the possibility of unknown causes.
>      * This message uses LOG level, because an unclean shutdown at this point
>      * would usually not look much different from a clean shutdown.
>      */
>     if (pthread_is_threaded_np() != 0)
>         ereport(LOG,
>                 (errcode(ERRCODE_INTERNAL_ERROR),
>                  errmsg_internal("postmaster became multithreaded"),
>                  errdetail("Please report this to <%s>.", PACKAGE_BUGREPORT)));
> 
> We now have a report[1] of a way to trigger this error message.
> It seems to boil down to "cause an error before PostmasterMain
> has reached its test of pthread_is_threaded_np()", for example
> 
> $ # build with --enable-nls, using MacPorts' gettext package
> $ unset LANG
> $ postgres -D /foo
> postgres: could not access directory "/foo": No such file or directory
> Run initdb or pg_basebackup to initialize a PostgreSQL data directory.
> 2025-01-23 18:16:03.843 GMT [41524] LOG:  postmaster became multithreaded
> 2025-01-23 18:16:03.843 GMT [41524] DETAIL:  Please report this to <pgsql-bugs@lists.postgresql.org>.

That makes sense.  Thanks for confirming it.

> I suggest therefore that we remove the "Please report a bug"
> bit and instead use the much more helpful hint appearing in
> PostmasterMain:
> 
>                  errhint("Set the LC_ALL environment variable to a valid locale.")));

That suffices.  It took a lot of years for the "Please report" message to
yield a report, so I'm not much concerned about losing reports from this
change.

> An alternative idea (that maybe should also be done in PostmasterMain)
> is to report the HINT only #ifdef __darwin__ and on other platforms
> use the "Please report a bug" text.

I gather pthread_is_threaded_np() is still macOS-only, in which case the
!darwin branch would be future-proofing only.

Another alternative would be to track whether we reached the startup-time
check, emitting a message at exit only if we did reach the startup-time check.
If a postmaster passes the startup-time check and fails the exit-time check,
changing the environment won't help.  Even so, I'm content with your first
proposal.

> [1] https://www.postgresql.org/message-id/18783-d1873b95a59b9103%40postgresql.org