I am affected by this also (running Stein ovs-agent with native at large scale).
When restarting the agents, they are receiving the RPC messages that finally call add_fdb_flow function.
BUT, the first RPC loop (iter 0) may finish before all add_fdb_flow are successfully executed.
The flows that are problematic are the "flood" flows (table 22), which are not sent using "call" but "cast" on neutron RPC side [1]
[1] https://github.com/openstack/neutron/blob/5a748b7c29cb846d0de3c5e8ea3868873348c0d1/neutron/plugins/ml2/drivers/l2pop/rpc.py#L48
I am affected by this also (running Stein ovs-agent with native at large scale).
When restarting the agents, they are receiving the RPC messages that finally call add_fdb_flow function.
BUT, the first RPC loop (iter 0) may finish before all add_fdb_flow are successfully executed.
The flows that are problematic are the "flood" flows (table 22), which are not sent using "call" but "cast" on neutron RPC side [1]
[1] https:/ /github. com/openstack/ neutron/ blob/5a748b7c29 cb846d0de3c5e8e a3868873348c0d1 /neutron/ plugins/ ml2/drivers/ l2pop/rpc. py#L48