On 11/23/2015 09:37 AM, Colin Walters wrote:
> On Fri, Nov 20, 2015, at 01:25 PM, Dusty Mabe wrote:
>> See attached patch for making atomic host have eth0 be the interface by
> So this is something like the addition of dracut-network by default
> for PXE-to-Live got us networking in the initramfs, which in turn
> broke the ifnames change?
I'm not sure what broke it. I'd love to know but don't have time to
look at it further.
> Your link was only talking about Vagrant, but this change will
> affect *all* uses of the cloud image, such as EC2, OpenStack,
Right, this would affect them all. As you mention below we were
intending for this ("eth0") to be the case for a long time so I
figured we should fix it. You are right that this would be a change
over how we released F23 atomic and it does have implications for
Hopefuly testing would identify any issues but you never know what
could crop up.
> Now, we were clearly *intending* to do this for a long
> time, going back to what (I believe) mattdm did with
> the original Fedora Cloud image.
> That said, while clearly a number of people just do cloud
> instances with one NIC and hence predictable = eth0 is fine,
> it's really quite valid to use multiple NICs in any kind of nontrivial
> situation. And once one does that, the rationale
> for PIN becomes apparent.
> I admit I'm only a light EC2 user, so I can't comment on
> this patch from that angle. From an OpenStack (specifically
> KVM) point of view, one really important thing to note is
> that recently virtio-net enumeration order was declared stable:
> Which definitely affects upgrades.
> It's worth noting that *currently* on KVM on Fedora 23 one gets PIN due
> to that change, and this would revert it.
> I'd like to explore this vagrant issue more to understand how PIN
> is breaking it.
Please do! I'd like other eyes looking at this. I took the
but would love for someone to make it better.