[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Proposal: enable stateless persistant network interface names



On Sun, May 17, 2015 at 08:03:54PM +0200, Bernd Zeimetz wrote:
> On 05/08/2015 09:28 AM, Josh Triplett wrote:
> > Marc Haber wrote:
> >> I have tried this just last week and have found it kind of
> >> unsatisfactory that it doesn't work in virtualized environments. For
> >> example, in a KVM VM with virtio ethernet, the network devices still
> >> end up in the system as eth0, eth1, eth2.
> > 
> > As I understand it, that's intentional and expected, for two reasons.
> > First, because on a virtual machine, the network interfaces are likely
> > to be more stable, always showing up with the same numbers.  And second,
> > because there's little else to go on when naming them.
> 
> Unfortunately that is not true for VMware.
> If you run a vm with more than three vmxnet(3) network interfaces it will depend
> on the hw version of the vm in which order they appear (due to other pci ids and
> different order on the pci bus!), it might happen that the 3rd or 4th interface
> is ordered in front of the first interface.
> Also mixing e1000 and vmxnet(3) interfaces will result in a mess.
> 
> It would be really great to fix this, as far as I can see the only way would be
> using mac address based interface names as the pci bus ids will change while
> upgrading hardware versions in vmware.

If that's a standard property of VMWare VMs, you could submit a udev
rule that improves the default naming on such systems.  I believe there
are already some rules defining policies in that area, including
VM-specific information.

- Josh Triplett


Reply to: