Hi Ricardo, We (some operators) are trying to meet (semi-) regularly to exchange problems/experiences like this, next time on March 11 1500 UTC, see: https://lists.openstack.org/archives/list/openstack-discuss@lists.openstack....
With the removal of the linuxbridge mechanism driver, has anyone successfully completed an in-place migrationfrom LinuxBridge to OVN?
At CERN, we plan to migrate most users in-place while running LinuxBridge and OVN at the same time (via cold/live migration). That being said, we only have provider network for the users and will go hypervisor by hypervisor. From what I saw in neutron, it looks like you could have LinuxBridge and OVN as providers at the same time. Running both LXB and OVN on the same hypervisor I haven't tried as well as overlay networks, and this is where the tricky part might start depending on the overlay used and potential downtime you want to have for your users. Best, Daniel ________________________________ From: Sean Mooney <smooney@redhat.com> Sent: Monday, March 3, 2025 15:21 To: Ricardo Cano <ledsole@gmail.com>; openstack-discuss@lists.openstack.org <openstack-discuss@lists.openstack.org> Subject: Re: [neutron] LXB Removal and OVN migration On 03/03/2025 13:50, Ricardo Cano wrote:
Hello openstack-discuss,
With the removal of the linuxbridge mechanism driver, has anyone successfully completed an in-place migrationfrom LinuxBridge to OVN? I did come across this https://www.jimmdenton.com/migrating-lxb-to-ovn/ and that looks good as a one-shoit while biting the bullet and taking a full outage. However, I was wondering if it was possible to run lxb and ovn side-by-side and cut them over, say one network at a time.
This particular deployment was stood up using openstack-ansible back in the rocky days (now running antelope) and it appears that openstack ansible's default was lxb up until Zed.
that is likely because the official default backend of neutron was Linux bridge in the install guide since it creation until it was marked experimental by neutron our devstack ci jobs defaulted to ovs for most of openstack lifetime but ovs only became the offical default backend when linux bridge was marked experimental https://docs.openstack.org/neutron/latest/install/controller-install-ubuntu.... ovn is still not the default in the docs or in most installers even if that is what we use in the devstack jobs by default. its an add on alretrnive procedure for "advanced users only" https://docs.openstack.org/neutron/latest/install/ovn/manual_install.html so openstack-ansibel was jsut followign the documented defautl backend. kolla, devstack and tripleo diverged from that.
Any suggestions or recommendations on the best path forward?
Also, will the dynamic routing agent work with ovn
i believe ovn does DVR differently without a requirement form the agent. i.e. if the vm has a floating ip i belive ovn can route that traffic locally to the internet form the compute without needing to use a central gateway but ill admit i could be missremebering that i have not looked at it in detail in a number of years. https://docs.openstack.org/neutron/latest/ovn/gaps.html """ * Floating IP Port Forwarding in provider networks and with distributed routing Currently, when provider network types like |vlan| or |flat| are plugged to a router as internal networks while the |enable_distributed_floating_ip| configuration option is enabled, Floating IP port forwardings which are using such router will not work properly. Due to an incompatible setting of the router to make traffic in the vlan/flat networks to be distributed but port forwardings are always centralized in ML2/OVN backend. This is being reported in [11] <https://docs.openstack.org/neutron/latest/ovn/gaps.html#id23>. [11] https://bugs.launchpad.net/neutron/+bug/2028846 <https://bugs.launchpad.net/neutron/+bug/2028846> """