Comment 5 for bug 1934666

Thank you Weronika for the testing. Runing the compute service with dvr_snat mode for L3 agent has some known issues, like the ingress flooding on br-int, packet drop on nodes, gateway MAC miss learning and so on. They are all not easily to fix, especially for DVR+HA routers.
But for this issue here, IMO, maybe spawn radvd process in all node can solve the problem. Remove the "master" check can be a simple way, (aka a revert for [2]). Please have a try for that.

[1] https://github.com/openstack/neutron/blob/master/neutron/agent/l3/ha_router.py
[2] https://github.com/openstack/neutron/commit/2f9b0ce940099bcc82d2940b99bdc387db22d6fc