Hi in Openstack side we have an issue during VM migration[1] when we see packet drops during VM migration when northd/ovn-controller are loaded and slow to process the events i.e actual flow setup takes time on the destination/additional chassis while the VM get's migrated(started on the destination).
 
We are looking into optimizing it on the CMS side(as we were doing it wrongly till now) * First option we have is to check for additional_chassis set on the Port_Binding, this happens once northd does it work but with this we still be having race condition when ovn-controller on destination chassis is loaded or slow.

• For the ovn-controller case
  • One option is to have an agent running on the chassis where VM runs and that monitors the OVSDB for the updates to external-ids:ovn-installed for the Interace and trigger the migration when that get's detected, we have an OVN agent currently but that's not mandatory one in the deployment so looking for the agent less case

• • The request here is basically for the agentless deployment, so when flows are setup on the destination chassis i.e external_ids:ovn-installed is set for the interface, to get this reflected on the SB DB side for the Port_Binding(something like additional_up=True or some other flag per additional chassis)? so we can just monitor for that SB DB update like other events and trigger the actual migration?
     

[1] https://issues.redhat.com/browse/OSPRH-21767?focusedId=28666552&page=com.atlassian.jira.plugin.system.issuetabpanels%3Acomment-tabpanel#comment-28666552

Slack where discussed originally https://redhat-internal.slack.com/archives/C01G7T6SYSD/p1765884661269999