This bug is related to:

• live-migrated instance in MIGRATING status for a long time: https://bugzilla.redhat.com/show_bug.cgi?id=2185897
• OVN Controller should not claim subport if its parent port is bound to a different chassis: https://bugzilla.redhat.com/show_bug.cgi?id=2220938
  • https://patchwork.ozlabs.org/project/ovn/patch/20230927183003.3498939-1-mheib@redhat.com/

When a VM with a Trunk port is migrated, the parent port is correctly bound to the new chassis. However, the subports are flapping during the process; e.g.: [1] the port is migrated from compute-0 to compute-1. The ovn-controller of each node (compute-0, compute-1) is always trying to claim the port (c34a1f96-d745-4e8c-8687-70d4b18d9f81) once the other node has claimed it.

Because of that, Neutron is receiving multple port binding chassis updates [2] related to the same port. If by accident the source host makes the last claim of the subport, this port will be assigned to an incorrect host (chassis) and it won't be defined as UP. That will make the live-migration to fail.

The goal of this bug is, if possible, to implement a mechanism similar to [3]: the LSP is aware of the multiple chassis and only the destination chassis claims it.

[1]https://paste.opendev.org/show/bIWeAAXaXb68ut0R0YoI/

[2]https://paste.opendev.org/show/bqCx4QxAu82vW9ohJUuq/

[3]https://bugzilla.redhat.com/show_bug.cgi?id=2012179