Re: pgbouncer | Update PgBouncer to 1.26.0 (!4)

On Fri, Sep 25, 2026 at 2:05 AM Christoph Berg (@myon)
<gitlab(at)salsa(dot)debian(dot)org> wrote:
>
> Sorry that I didn't properly check the status on jengus.postgresql.org - I thought the package was ready so I tried to release it, but there are some test problems. (I tried running it locally first, that didn't work, but I ignored that. That should have been a warning.)
>

Moving the conversation from salsa.debian.org since I _think_ the issue
is on jengus.postgresql.org and not with the Debian package.

Looking at the logs for latest build of pgbouncer-binaries (#139), I
think the failures are resource contention. Looking at the logs for
forky amd64, there were 5 test failures:

1. test_client_idle_timeout
2. test_transaction_timeout
3. test_idle_transaction_timeout
4. test_query_timeout_when_no_active_query
5. test_server_connect_timeout_establish

All of the tests failed during the teardown of the fixture. In all five
cases the failure was a timeout waiting for the command 'pg_ctl -m fast
stop'. I've looked at the logs for all of the failed amd64 builds [1]
and I see a similar pattern. Additionally, I have not been able to
reproduce a test failure locally.

It looks like pytest is launched with '-n auto' which [2] causes pytest
to

> spawn a number of works processes equal to the number of available
> CPUs.

I try disabling the parallel execution of tests by setting CONCURRENCY=1
in override_dh_auto_test. I've verified that this works locally but
since I don't have a local reproduction, I'm not 100% certain this will
fix the test failures. I also don't have a good sense for what the
impact on build times would be. I also looked at modifying the Jenkins
job to limit the number of conccurrent builds but the only option I
found was to configure the matrix build to be serial.

-- Bradford

[1]: https://jengus.postgresql.org/job/pgbouncer-binaries/139/
[2]: https://pytest-xdist.readthedocs.io/en/stable/