Comment 4 for bug 1888256

1) Yes. We want to have a "simple" or as we call "Flat" network, which can be used by all projects, and gives public v4/v6 addresses. (next to our "Smart" networks, where dual qrouter, floating IP, VPNaaS etc available, but no IPv6 now)
Since the project administrators cannot manage ports, they got new IP with new instance.

2) Yes.

3) We use the router only for metadata in our "Flat" named networks. (cannot remember why, but we use this method from kilo. ).
As you can see, the "Flat1-subnet-v4" has a static route: (destination='169.254.169.254/32', gateway='193.224.218.251' ) where the 193.224.218.251 is the floating IP of the qrouter.

Maybe I misunderstand you, but I should try with None/None?

I thought, the address-mode=slaac will make the neutron allocate the proper address (which calculated from the MAC) and it's a must for the iptables rules in the back of the qrouters.

However, I used the documentation (https://docs.openstack.org/neutron/rocky/admin/config-ipv6.html) and it's clearly says with RA mode=none, and address mode=slaac: "Guest instance obtains IPv6 address from non-OpenStack router using SLAAC."

For some reason, the radvd process spawned with the qrouter with this configuration, and I cannot really understand, why.

The another odd thing is that our RegionOne, which upgraded from Kilo to Rocky in a past few years works with the old networks in this way. (our Flat1 and Flat2 which created in Kilo) Our new "Flat3" named network in RegionOne, and the two new "Flat" in RegionTwo behave this way only.