As far as I know, testing were done mostly with 0.8.1 during the cycle. So probably it is less harm, especially considering that Rodolfo wrote that however neutron projects work fine with 0.8.x and 0.9.x versions as well, they are mostly using 0.8.1 anyway and he also suggests capping to 0.8.1 if i understand correctly.

@Matthew : either we cap pyroute2 to 0.8.1 or leave it 0.9.4, is that possible to bump to 0.9.5 (or whatever version fixes the bug) later on after the release on stable/2025.2?

Thanks,
Előd

________________________________________
From: Matthew Thode
Sent: Wednesday, September 10, 2025 17:51
To: Balazs Gibizer
Cc: openstack-discuss
Subject: Re: [requirements]Flamingo pyroute2 regression


On 25-09-10 13:57:06, Balazs Gibizer wrote:

> On Wed, Sep 10, 2025 at 1:53 PM Balazs Gibizer <gibi@redhat.com> wrote:

> >

> > (sending again as I probably sent it originally from an unregistered address)

> > Hi requirements folks,

> >

> > We identified a bug[1] in pyroute2 0.9.1 that causes intermittent

> > failures during interface attach / detach operations. Originally we

> > got the indication that the pyroute2 0.9.4 fixed the issue and our

> > tests showed green results so we bumped to that version. But since

> > then we observed that the problem still happens in 0.9.4 as well.

> >

> > Can we roll back our pyroute2 dependency to < 0.9 in Flamingo so close

> > to the release?

> >

> > If not, can we later roll back the requirement on stable/2025.2? Or

> > the only way is to roll forward on stable/2025.2 (and on master) when

> > the fix appears in pyroute2?

>

> And I forgot to add that I pushed the rollback patch

> https://review.opendev.org/c/openstack/requirements/+/960366

>

> gibi

>

> >

> > Thanks,

> > gibi

> >

> > [1]https://bugs.launchpad.net/os-vif/+bug/2109396

> >

>



Rolling back at this time is somthing I don't think would be a good

idea, pyroute2 is used in a lot of places and I'm not sure what (if any)

changes those projects would have done that could break when rolling

back to 0.8.1.  Is there any workaround the the attach/detach that could

be made instead (retry)?



--

Matthew Thode