We are not using kolla, but we have the same issue. Actually, restarting ovs could result in ping loss for instances. Our process here is to empty the compute in case we need to perform this (live migrate) On 23.06.25 - 15:16, İzzettin Erdem wrote:
Hello,
We have a kolla-ansible yoga environment with 9 controller nodes and more than 1000 compute nodes. When I upgrade neutron and openvswitch to a newer version (doesn't matter which, in my case Antelope) access to all virtual machines is lost, but if there are existing ping flows, new ones are not created.
This continues until openvswitch deploy is completed, i.e. neutron_openvswitch restart. In some cases, the access outage can be prolonged. When I send an upgrade to neutron alone, I don't experience any outage. How can I prevent this, is there a way to send a hot-upgrade to neutron&openvswitch without any outage? Or is it normal to have an outage?