After patch https://opendev.org/openstack/tripleo-heat-templates/commit/2dc7066b050ecf22dc9e5909061272ffe765ebfc there are "kill scripts" for all processes managed by e.g. L3 agent, even for radvd which isn't run in sidecar container but in neutron-l3-agent container directly. That is causing various issues. First one was in Neutron, and is covered by BZ1823795. But with fix for BZ1823795 we hit another issue. As radvd-kill script is now allowed to be executed by neutron-l3-agent, it finds neutron-l3-agent container and kills it. So every time when radvd is killed by neutron-l3-agent because e.g. IPv6 subnet is deleted from router, agent is also restarted (killed by radvd-kill script and started back by systemd).
I linked current WIP upstream patch, though focus currently is on related bug #1827276
Whole system was reverted for 16.1, so this bug is no longer relevant, closing