Instances won't obtain IPv6 address if they have additional IPv4 interface

[Nir Magnezi]

Description of problem:
=======================
I booted an instance with both IPv4 and IPv6 interfaces, yet that instance did no obtain any IPv6 address.
In order make sure nothing is wrong with my IPv6 configuration (which is RADVD SLAAC), I booted additional instance with IPv6 interface only, which obtained an IPv6 address with no issues.

Version-Release number of selected component (if applicable):
=============================================================
openstack-neutron-2014.2-0.7.b3
How reproducible:
=================
Always

Steps to Reproduce:
===================
0. Perior to the test, configure the following:
   a. Neutron router
   b. IPv4 Network & Subnet
   c. IPv6 Network & Subnet (SLAAC in my specific case)
      --> Created with: --ipv6-address-mode slaac --ipv6_ra_mode slaac
   d. Add router interfaces with those networks.

1. spwan an instance with both IPv4 & IPv6 interfaces.

2. spwan an instance with IPv6 interface only.

Actual results:
===============
1. The instance spawed in step 1 obtained IPv4 address and IPv6 link local address only
2. The instance spawed in step 2 obtained IPv6 addrees proparly.

Expected results:
=================
Instances should obtain all IP addresses in both scenarios I mentioned above.

Additional info:
================
Using tcpdump from within the instances I noticed that ICMPv6 Router Advertisments did not reach the NIC.

[Ihar Hrachyshka (ihar-hrachyshka)]

Quoting myself from the Red Hat counterpart bug: https://bugzilla.redhat.com/show_bug.cgi?id=1151820

"Installed RDO Juno, created two networks, one with ipv4 subnet, another once with ipv6 subnet. Then booted Cirros instance with two NICs, one per network. Cirros booted with eth0 up, with IPv4 address set properly. eth1 didn't have address set, because Cirros does not attempt to configure it (see /etc/network/interfaces). Once I did 'ifconfig eth1 up', the address was set correctly.

I suspect that's behaviour you experienced, and it's correct behaviour (as far as we may consider image behaviour as correct)."

I believe this is not-a-bug, unless more details arise.

[Livnat Peer (lpeer)]

On 10/27/2014 03:43 PM, Ihar Hrachyshka wrote:
> Quoting myself from the Red Hat counterpart bug:
> https://bugzilla.redhat.com/show_bug.cgi?id=1151820
>
> "Installed RDO Juno, created two networks, one with ipv4 subnet, another
> once with ipv6 subnet. Then booted Cirros instance with two NICs, one
> per network. Cirros booted with eth0 up, with IPv4 address set properly.
> eth1 didn't have address set, because Cirros does not attempt to
> configure it (see /etc/network/interfaces). Once I did 'ifconfig eth1
> up', the address was set correctly.
>
> I suspect that's behaviour you experienced, and it's correct behaviour
> (as far as we may consider image behaviour as correct)."
>
> I believe this is not-a-bug, unless more details arise.
>
> ** Bug watch added: Red Hat Bugzilla #1151820
> https://bugzilla.redhat.com/show_bug.cgi?id=1151820
>

Great thanks for the update, let's see if nir confirms and closes the bug.

[Ann Taraday (akamyshnikova)]

Check this with ubuntu instances. http://paste.openstack.org/show/126521/ "nova show" and "nova list" show ip addresses, although if we connect to vm via VNC there will be no ip addresses and no gateways in both cases.

[Ann Taraday (akamyshnikova)]

Check this with ubuntu instances. http://paste.openstack.org/show/126521/ "nova show" and "nova list" show ip addresses, although if we connect to vm via VNC there will be no ip addresses and no gateways in both cases.

[Ihar Hrachyshka (ihar-hrachyshka)]

@Ann, you eth1 is down. I suspect Ubuntu does not try to configure eth1, or require ipv4 addresses for all interfaces. Have you tried to make it up manually: ifconfig eth1 up?

[Sergey Belous (sbelous)]

@Ihar, I checked this for ubuntu: http://paste.openstack.org/show/184928/ and all works fine.