neutron l3 agent restart with l3 ha enabled

[Fedor Tarasenko]

I have k-a with OVS & L3 HA enabled.

When I restart neutron_l3_agent container on one node, it removes all keepalived processes from this node, but does not remove IP addresses from namespaces (that's normal behaviour for failed keepalived process). When neutron_l3_agent restarts, it runs neutron-netns-cleanup which removes all namespaces, but this process takes some time (3-5-10 minutes depending on installation). This behavior was created by solving this bug https://bugs.launchpad.net/kolla/+bug/1703078
Оver a certain period of time, I have two same IP-MAC pairs on active and failed l3 nodes, which confuses our network a lot.

The question is - was this behaviour ever discussed and any solution provided?
1. For example, our team want to check if moving keepalived processes to host can help.
2. Another option is to drain l3 agent before restart. But for now we have no checks for this process in k-a.