router ext gw assigning IPv6 when IPv6 + NAT is unsupported

[Weronika Sikora]

Hi,

this is both a question and a bug report.

Currently, when setting an external gateway on a router by using an external network with both IPv4 and IPv6 subnets, two IPs are assigned - IPv4 and IPv6. However, according to the docs [1] , Openstack Networking does not yet support NAT with IPv6.

This makes the IPv6 address on the external gateway both confusing and without any use. I propose to stop assigning IPv6 for router's external gateway to avoid such confusion. If it's not supported yet, then there's no reason to assign it.

While Master might yet include such support in the future, stable branches such as Stein or Train do not. The nature of those being stable is that they don't include big features, only bug fixes, which means that they will never support IPv6 with NAT. Therefore, even if you decide to leave this assignment in the Master branch, I think that stable branches should have this removed regardless.

Some simple variation of "[v4] if is_gw_port else [v4, v6_stateful]" should suffice - [2] .

[1] https://docs.openstack.org/neutron/latest/admin/config-ipv6.html#nat-floating-ips
[2] https://opendev.org/openstack/neutron/src/commit/9e2a0ac058305dd16d6dedb883d305cd162f5fe4/neutron/db/ipam_pluggable_backend.py#L279

[Brian Haley (brian-haley)]

There is (at least) one out-of-tree neutron driver that does support NAT6-6, think it's Midonet. Think removing the IPv6 address could break that.

[Weronika Sikora (shushuda)]

Ah, good point, I wasn't aware of that.

[Dr. Jens Harbott (j-harbott)]

In addition, IPv6 can be routed without NAT, this is the standard scenario deployed in combination with neutron-dynamic-routing. So the statement of "unsupported" is wrong in general. It should be specified in more detail what exactly the bug is meant to be about.

[Weronika Sikora (shushuda)]

The assumption was about the NAT usage in particular. It was more of a question of "why is IPv6 assigned for ext gw if docs say IPv6 + NAT is unsupported, shouldn't this be removed?". After your explanations, I think this issue is no longer valid and could be closed.

I've changed the title to a better fitting one.