[12:49] <caribou> minimal: FYI the bug for the issue is https://github.com/canonical/cloud-init/issues/4518
[12:49] -ubottu:#cloud-init- Issue 4518 in canonical/cloud-init "ENI and Network-Manager renderers create wrong IPv6 routes" [Open]
[14:00] <caribou> minimal: oops, didn't notice that you weren't there : FYI the bug for the issue is https://github.com/canonical/cloud-init/issues/4518
[14:00] -ubottu:#cloud-init- Issue 4518 in canonical/cloud-init "ENI and Network-Manager renderers create wrong IPv6 routes" [Open]
[14:33] <minimal> yeah saw your msg in the logs, thanks
[15:51] <caribou> My impression is that _add_route() does not take into account the fact that IPv4 and IPv6 routes are mixed together as does _add_nameserver() right below it 
[15:51] <caribou> https://github.com/canonical/cloud-init/blob/d0f00bd54649e527d69ad597cbcad6efa8548e58/cloudinit/net/network_manager.py#L139
[16:06] <minimal> caribou: _add_address appears to act in the same way as those other 2 functions yet it seems to result in v4 and v6 addresses being kept separate
[16:07] <caribou> I'm just testing adding the bits from _add_nameserver() into _add_route() and it looks like it is DTRT
[16:07] <minimal> actually _add_nameserver doesn't take "family" as an argument whereas _add_route and _add_address do
[16:11] <minimal> so as long as "family" is correctly set then _add_route should call _add_numbered with the correct family and _add_numbered should then put the route in the relevant IPv4/IPv6 section
[16:15] <caribou> The problem is the family = "ipv6" if subnet_is_ipv6(subnet) else "ipv4" at the start of the for loop set family as ipv4 for all the routes even if one is v4 and the one pointed to by 'gateway' is 6
[16:16] <caribou> v6
[16:48] <minimal> caribou: that's assuming all the routes in a "subnet" section are of the same type (v4 / v6)
[16:49] <minimal> so the question is why then does a subnet have mixed v4 / v6 data?
[18:46] <fabiomirmar> @noahm, is there any work in progress or plans to update debian's ec2-net-utils wiht what you have upstream?
[18:47] <fabiomirmar> @noahm, we noticed that the postinst file was removed a while ago by Commit 908d2ef9 and we wanted to have it in Ubuntu, so instead of maintaining a bunch of SAUCE stuff in Ubuntu, it would be nice if we can just sync from debian