MIME-Version: 1.0
Date: Thu, 06 Nov 2025 20:34:06 +0000
From: Richard Huxton <richard@huxton.org.uk>
To: pgsql-pkg-debian@lists.postgresql.org
Subject: Re: Patroni 4.1.0 causes issues because of systemd notify
In-Reply-To: <emf4c21a82-984c-4bc4-91e8-79bdf35bc018@79dd1d3b.com>
References: <emf4c21a82-984c-4bc4-91e8-79bdf35bc018@79dd1d3b.com>
Message-ID: <ea1c371415349429e65a28cdfedb516c@huxton.org.uk>
Content-Type: text/plain; charset=UTF-8;
 format=flowed
Content-Transfer-Encoding: quoted-printable
Archived-At: 
 <https://www.postgresql.org/message-id/ea1c371415349429e65a28cdfedb516c%40huxton.org.uk>
Precedence: bulk

On 2025-10-22 17:16, Hauke Bruno Wollentin wrote:
> Hi all,
>=20
> while spinning up new Patroni cluster with version 4.1.0 (which has=20
> been added to the Debian repo 2 days ago), my clusters won=E2=80=99t bo=
otstrap=20
> properly anymore.
>=20
> At first, logs looks clear but anyways, all systemd operations like=20
> start/stop/restart will run into a timeout, causing the service to stop=
=20
> which then will break the cluster at all.
>=20
> It seems like the new systemd notify feature is causing this, since=20
> using Patroni 4.0.7 with `Type=3Dsimple` in the systemd unit file works=
=20
> pretty fine.

I think you may well be right - it seems that patroni is sending the=20
notify from a forked child process and that isn't what systemd is=20
expecting

> Oct 22 17:43:09 hetzner-lab01 systemd[1]: patroni.service: Got=20
> notification message from PID 13083, but reception only permitted for=20
> main PID 13059

> Any recommendations how to fix/workaround this?

https://www.freedesktop.org/software/systemd/man/latest/systemd.service.h=
tml#NotifyAccess=3D

I don't have a test patroni installation handy right now, but I think if=20
you set

     NotifyAccess=3Dall

in the service config (which you should be able to do by creating a file=20
with just that line called `/etc/systemd/system/patroni/override.conf`)

It defaults to "main" IIRC which is the most restricted, "exec" might=20
work, but "all" (the least restricted) seems the most likely to work=20
first time.

Hopefully that will do the trick. If so, maybe try "exec" then and let=20
the list know which works and the package can be updated.

--=20
   Richard Huxton