Both during adoption and later during updates/upgrades of the EDPM nodes we should make sure that openvswitch package update won't disrupt the dataplane traffic.

In TripleO it was done with hack https://github.com/openstack-archive/tripleo-heat-templates/blob/stable/wallaby/deployment/undercloud/undercloud-upgrade.yaml#L126-L145 and on overclouds https://github.com/openstack-archive/tripleo-heat-templates/blob/stable/wallaby/deployment/tripleo-packages/tripleo-packages-baremetal-puppet.yaml#L468-L515 The hack was effectively skipping some parts of rpm installation for the package.

Right now in the OSP 18, after https://github.com/openstack-k8s-operators/edpm-ansible/pull/539 was merged we don't run update packages for update/upgrades but that may change in future.

As per OVS folks, the right way to update vswitchd is to use flow-restore-wait option in db to affect how vswitchd restarts; then store / restore flows using ovs-ofctl; then disable the option. These steps can be orchestrated in ansible.

Note: a similar problem also applies to podified OVS. There, RPM hack definitely doesn't apply, so it would be wise to come up with some common solution to both scenarios.

Note: this bug is only about EDPM node side. Podified OVS handling of OVS image updates should be tracked in a separate bug (to be reported.)