Le 31/03/2015 09:21, Sven Kieske a écrit : > You should file an RFE on bugzilla for this > kind of stuff and maybe raise awareness during > the weekly meeting on irc. > > this increases the chance of an implementation > taking place dramatically ;) > > HTH > > Hi Sven, I found and append this : https://bugzilla.redhat.com/show_bug.cgi?id=1045022

Hi Lior, Le 31/03/2015 12:55, Lior Vernia a écrit : > Indeed, if you're using oVirt 3.4 and up, and you supply all the NICs > whenever you create the VM (not afterwards - as part of the new VM > dialog), the NICs should receive MAC addresses according to their > ordering by names, e.g. nic1 will always get a lower MAC address than > nic2. For a newly-created VM this should guarantee that nic1 will indeed > end up the first NIC inside the guest OS and be used by PXE. In the BZ I provided above, I explain that my tests are showing that this is not true : the nics are not yet sorted according to their MAC nor their names. I'm using 3.5.1

> > If this is something you do a lot, then I'd suggest creating a new > template as such and create VMs from that. I know that could be a way, but this is not our strategy, as we prefer using kickstart and we have so many different setups that factorisation is not possible. > > On 31/03/15 11:23, Nicolas Ecarnot wrote: >> Le 31/03/2015 09:21, Sven Kieske a écrit : >>> You should file an RFE on bugzilla for this >>> kind of stuff and maybe raise awareness during >>> the weekly meeting on irc. >>> >>> this increases the chance of an implementation >>> taking place dramatically ;) >>> >>> HTH >>> >>> >> >> Hi Sven, >> >> I found and append this : >> https://bugzilla.redhat.com/show_bug.cgi?id=1045022 >>

Le 31/03/2015 13:57, Lior Vernia a écrit : > > > On 31/03/15 14:21, Nicolas Ecarnot wrote: >> Hi Lior, >> >> Le 31/03/2015 12:55, Lior Vernia a écrit : >>> Indeed, if you're using oVirt 3.4 and up, and you supply all the NICs >>> whenever you create the VM (not afterwards - as part of the new VM >>> dialog), the NICs should receive MAC addresses according to their >>> ordering by names, e.g. nic1 will always get a lower MAC address than >>> nic2. For a newly-created VM this should guarantee that nic1 will >>> indeed >>> end up the first NIC inside the guest OS and be used by PXE. >> >> In the BZ I provided above, I explain that my tests are showing that >> this is not true : the nics are not yet sorted according to their MAC >> nor their names. >> >> I'm using 3.5.1 >> > > I see. It seems MAC addresses are allocated according to the NIC name > order, I'm not sure this is exact. According to what I'm witnessing, MAC addresses are allocated in incremental order, from the first free MAC address in the MAC pool range.

> and that they are also named in the corresponding order within > the guest OS. Er.. what gives when I'm naming my interfaces ethX, and amongst that, oVirt comes trying to add some nicX???

> So the only problem here is how gPXE chooses a NIC to boot > from... I'm not familiar with the behavior of gPXE, but oVirt seems to > behave alright. I mostly agree with that, in the sense that one has to dig in which way gPXE is sorting the NICs, ie mapping the MACs to its "net0", "net1", and so on.

And then, decide whether oVirt could have a control upon this sorting, and then issue a RFE to master it from the web GUI.

Le 31/03/2015 14:48, Lior Vernia a écrit : >>> I see. It seems MAC addresses are allocated according to the NIC name >>> order, >> >> I'm not sure this is exact. >> According to what I'm witnessing, MAC addresses are allocated in >> incremental order, from the first free MAC address in the MAC pool >> range. >> > > My question is as follows: when you create a VM, and add NICs as part of > the new VM dialog (using the widget at the bottom), have you ever > encountered a situation where a "higher-named" NIC got a lower MAC > address? Absolutely YES. I just repeated this test, and it is easy to reproduce : in a MAC pool, manage to free some lower MAC addresses, and the will get allocated from the first free in the pool.

And I found this behaviour SOUND. Don't change it. >>> So the only problem here is how gPXE chooses a NIC to boot >>> from... I'm not familiar with the behavior of gPXE, but oVirt seems to >>> behave alright. >> >> I mostly agree with that, in the sense that one has to dig in which way >> gPXE is sorting the NICs, ie mapping the MACs to its "net0", "net1", and >> so on. >> > > That's the key point here for me, as it's difficult to predict the > behavior of n'importe quel piece of software... :) Thanks for your explanation, though I think the present way oVirt is assigning the MACs and naming the interfaces are not bad and should not change. The key point is how to make oVirt force gPXE to use the interface we want, whatever its name or its MAC. If possible? It seems I'm not the first one to deal with this issue : http://etherboot.org/pipermail/gpxe/2011-September/001742.html

Le 31/03/2015 15:55, Lior Vernia a écrit : > So the lower ones are allocated. But I would expect nic0 to receive a > lower MAC address than nic1 on the same VM. Have you encountered a > situation where that is not the case? Once again : YES. What I'm witnessing is that oVirt only uses the first free MAC, and does not mind the name.