VIF unplug operation failing on Rocky Linux 9.3 (Yoga)

[Pierre Riteau]

Description
===========

VIF unplug operations are failing on Rocky Linux 9.3 with stable/yoga. This is caused by incompatibility between pyroute2 0.6.6 (yoga u-c) and the Rocky Linux 9.3 kernel.

Steps to reproduce
==================

* Launch a virtual machine
* Try to resize the virtual machine

Expected result
===============

The resize operation succeeds (assuming enough resources are available).

Actual result
=============

The resize operation will fail and a similar traceback will be logged in nova-compute:

Traceback (most recent call last):
  File "/var/lib/kolla/venv/lib/python3.9/site-packages/os_vif/__init__.py", line 110, in unplug
    plugin.unplug(vif, instance_info)
  File "/var/lib/kolla/venv/lib/python3.9/site-packages/vif_plug_ovs/ovs.py", line 445, in unplug
    self._unplug_bridge(vif, instance_info)
  File "/var/lib/kolla/venv/lib/python3.9/site-packages/vif_plug_ovs/ovs.py", line 376, in _unplug_bridge
    linux_net.delete_bridge(vif.bridge_name, v1_name)
  File "/var/lib/kolla/venv/lib/python3.9/site-packages/oslo_privsep/priv_context.py", line 271, in _wrap
    return self.channel.remote_call(name, args, kwargs,
  File "/var/lib/kolla/venv/lib/python3.9/site-packages/oslo_privsep/daemon.py", line 215, in remote_call
    raise exc_type(*result[2])
pr2modules.netlink.exceptions.NetlinkError: (95, 'Operation not supported')

Note that this was resolved in Ubuntu by backporting a fix from pyroute2 0.6.10. For more details see https://bugs.launchpad.net/ubuntu/+source/pyroute2/+bug/1995469

We successfully reproduced the fix by backporting the following commit: https://github.com/svinota/pyroute2/commit/1eb08312de30a083bcfddfaa9c1d5e124b6368df

Environment
===========

This should be reproducible with Nova stable/yoga at the time of writing, when running on Rocky Linux 9.3:
5.14.0-362.8.1.el9_3.x86_64 #1 SMP PREEMPT_DYNAMIC Wed Nov 8 17:36:32 UTC 2023

Hypervisor is KVM:
qemu-kvm-7.2.0-14.el9_2.5.x86_64
libvirt-daemon-9.0.0-10.3.el9_2.x86_64

Storage is Ceph (Quincy).

Networking is Neutron with OpenVSwitch.

[Gaël THEROND (gtherond)]

I do confirm we have the exact same issue on a yoga based cluster using CentOS Stream 9 as a host.

[Pierre Riteau (priteau)]

Gaël, this was fixed in Kolla, see https://review.opendev.org/c/openstack/kolla/+/904781