[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: python3-dnspython and eventlet incompatibility

On 11/5/21 2:02 PM, Arnaud Morin wrote:
> Hey team,
> I am working on installing openstack neutron from openstack victoria
> extrepo.
> In my neutron config, I am setting transport-url with a fqdn, e.g.:
> transport_url = rabbit://user:pass@rabbit.arnaudmorin.fr:5672
> But, this is not working, due to a bug between eventlet (which seems to
> be used by neutron) and dnspython (see [1] and [2])
> I have two possible workarounds:
> * using an IP instead of FQDN
> * downgrading dnspython to <2.0.0 (1.16.0)
> I also tested the latest eventlet, but it seems too new for neutron
> victoria.
> I havn't tested xena, but if it's relying on the same python
> dependencies, it will also fail.
> What is the recommendation from the community about this?
> Should I report this as a bug? (If so, I dont remember where, so I'd
> appreciate a hint :))
> Cheers, Arnaud.
> [1] https://github.com/eventlet/eventlet/issues/629
> [2] https://github.com/eventlet/eventlet/issues/619

Hi Arnaud,

It is well known that there is an issue with Eventlet as in Bullseye,
and dnspython 2.0. However, we have a patch for it. I opened a release
team bug in the hope they accept the package, but so far, they haven't
replied to me. Please read this bug report:


I am hereby CC-ing the bug, so that the release team sees it really
affects Debian users.

In advance of the release team acceptance, I have just dropped (a few
minutes ago) the patched package as it will be uploaded when the stable
release team accepts it:


Pleaese try it, and let me know if it solves your troubles.

If it doesn't, there's still a workaround: write in /etc/hosts all of
the entries you need, like the entry for rabbit.arnaudmorin.fr as per
your transport_url directive. This way, Neutron will not use the
Eventlet greendns function (or at least, it will not use dnspython to

Please really let me know (and the Debian bug) if the patched Eventlet
fixes your trouble first though, as I didn't have the issue (because my
setup always write all the cluster nodes in /etc/hosts), and I need to
know if that fixes it.


Thomas Goirand (zigo)

Reply to: