Date: Tue, 7 Apr 2026 06:06:17 +0000
From: Bertrand Drouvot <bertranddrouvot.pg@gmail.com>
To: Andres Freund <andres@anarazel.de>
Cc: pgsql-hackers@lists.postgresql.org
Subject: Re: meson: Adjust test timeout for Valgrind builds
Message-ID: <adSe2cj22RY/NHoQ@ip-10-97-1-34.eu-west-3.compute.internal>
References: <ac/UW28t39GolvPB@ip-10-97-1-34.eu-west-3.compute.internal>
 <qqd4723zxp6rfi3c7apyi6vzffijrvaajyhoppd7q2ns3kz45u@duxmofii2f4n>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <qqd4723zxp6rfi3c7apyi6vzffijrvaajyhoppd7q2ns3kz45u@duxmofii2f4n>
Archived-At: 
 <https://www.postgresql.org/message-id/adSe2cj22RY/NHoQ%40ip-10-97-1-34.eu-west-3.compute.internal>
Precedence: bulk

Hi,

On Mon, Apr 06, 2026 at 09:25:46AM -0400, Andres Freund wrote:
> Hi,
> 
> On 2026-04-03 14:53:15 +0000, Bertrand Drouvot wrote:
> > When building with -DUSE_VALGRIND, tests run significantly slower due
> > to Valgrind's instrumentation overhead, causing the default 1000s test's timeout
> > to be exceeded. Example when running the regress test suite:
> > 
> > "
> > $ meson test -C build -q --print-errorlogs --setup running --suite regress-running
> > regress-running - postgresql:regress-running/regress time out (After 1000 seconds)
> > 
> > Summary of Failures:
> > 
> > 1/1 regress-running - postgresql:regress-running/regress TIMEOUT        1000.01s
> > 
> > Ok:                0
> > Fail:              0
> > Timeout:           1
> > "
> > 
> > PFA a patch that detects Valgrind builds using a compiler check, which correctly
> > handles USE_VALGRIND being passed via -Dc_args, CPPFLAGS or CFLAGS and increases
> > the test timeout to 10000s in that case.
> > 
> > I don't have a strong opinion on the new value. In practice, the regress
> > suite runs in about 30 seconds without Valgrind and in about 46 minutes with
> > Valgrind on my setup. Note that the timeout is per test, not for the entire
> > suite so that 10000s looks large enough (I tested to run the entire suite with
> > the patch and it did not produce any timeout).
> > 
> > Another option could be to disable the timeout on a Valgrind build (set timeout
> > to 0) but then a test could block forever.
> > 
> > Note that there are no changes needed for autoconf as it does not set a timeout
> > for the tests.
> 
> The usual way to deal with that is to pass --timeout-multiplier=100 or
> something like that to meson test.

Yeah, that's another option but I see it as an extra step if the patched version
still time out. Currently we know for sure that some tests will time out so I 
thought that was more use friendly to try to prevent that in the first place and
use the multiplier if it's still not enough for any reasons.

Regards,

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