[lxc-users] LXD 2.12 - VXLAN configuration connected to eth1

[Stéphane Graber]

Hi,

I sent a pull request which allows oevrriding the interface in multicast mode:
    https://github.com/lxc/lxd/pull/3210

When writing that code, I did notice that in my earlier implementation I
always selected the default interface for those, so that explains why no
amount of routing trickery would help.

Stéphane

On Sun, Apr 23, 2017 at 04:36:43PM -0400, Ron Kelley wrote:
> Thanks for the speedy reply!  From my testing, the VXLAN tunnel always seems to use eth0.  After running the “ip -4 route add” command per your note below, I disabled eth1 on one of the hosts but was still able to ping between the two containers.  I re-enabled that interface and disabled eth0; the ping stopped.  It seems the VXLAN tunnel is bound to eth0.
> 
> By chance, is there a workaround to make this work properly?  I also tried using the macvlan interface type specifying a VXLAN tunnel interface and it would not work either.  For clarity, this is what I did:
> 
> ip link add vxlan500 type vxlan group 239.0.0.1 dev eth1 dstport 0 id 500
> ip route -4 add 239.0.0.1 eth1
> <edit the LXD default profile; set the nictype to “macvlan”, and the parent to “vxlan500”>
> 
> I was hoping a raw VXLAN interface would work instead of using the LXD create command.
> 
> 
> -Ron
> 
> 
> > On Apr 23, 2017, at 4:18 PM, Stéphane Graber <stgraber at ubuntu.com> wrote:
> > 
> > Hi,
> > 
> > VXLAN in multicast mode (as is used in your case), when no multicast
> > address is specified will be using 239.0.0.1.
> > 
> > This means that whatever route you have to reach "239.0.0.1" will be
> > used by the kernel for the VXLAN tunnel, or so would I expect.
> > 
> > 
> > Does:
> > ip -4 route add 239.0.0.1 dev eth1
> > 
> > Cause the VXLAN traffic to now use eth1?
> > 
> > If it doesn't, then that'd suggest that the multicast VXLAN interface
> > does in fact get tied to a particular parent interface and we should
> > therefore add an option to LXD to let you choose that interface.
> > 
> > Stéphane
> > 
> > On Sun, Apr 23, 2017 at 04:04:03PM -0400, Ron Kelley wrote:
> >> Greetings all.
> >> 
> >> Following Stéphane’s excellent guide on using multicast VXLAN with LXD (https://stgraber.org/2016/10/27/network-management-with-lxd-2-3/).  In my lab, I have setup a few servers running Ubuntu 16.04 with LXD 2.12 and multiple interfaces (eth0, eth1, eth2).  My goal is to setup a multi-tenant computing solution using VXLAN to separate network traffic.  I want to dedicate eth0 as the mgmt-only interface and use eth1 (or other additional interfaces) as customer-only interfaces. I have read a number of guides but can’t find anything that clearly spells out how to create bridged interfaces using eth1, eth2, etc for LXD.
> >> 
> >> I can get everything working using a single “eth0” interface on my LXD hosts using the following commands:
> >> -----------------------------------------------------------
> >> lxc network create vxlan100 ipv4.address=none ipv6.address=none tunnel.vxlan100.protocol=vxlan tunnel.vxlan100.id=100
> >> lxc launch ubuntu: testvm01
> >> lxc network attach vxlan100 testvm01
> >> -----------------------------------------------------------
> >> 
> >> All good so far.  I created two test containers running on separate LXD servers using the above VXLAN ID and gave each a static IP Address (i.e.: 10.1.1.1/24 and 10.1.1.2/24).  Both can ping back and forth.  100% working.
> >> 
> >> The next step is to use eth1 instead of eth0 on my LXD servers,  but I can’t find a keyword in the online docs that specify which interface to bind (https://github.com/lxc/lxd/blob/master/doc/networks.md).
> >> 
> >> Any pointers/clues?
> >> 
> >> Thanks,
> >> 
> >> -Ron
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 801 bytes
Desc: not available
URL: <http://lists.linuxcontainers.org/pipermail/lxc-users/attachments/20170423/68170e2a/attachment.sig>