[CentOS] True bond howto for Centos 6

Sun May 13 16:54:12 UTC 2012
bob <bob at bobhoffman.com>

On 5/13/2012 12:30 PM, aurfalien wrote:
> On May 13, 2012, at 12:23 PM, bob wrote:
>
>> On 5/13/2012 11:45 AM, aurfalien wrote:
>>> Hi all,
>>>
>>> Read many posts on the subject.
>>>
>>> Using 802.3ad.
>>>
>>> Few problems;
>>> Cannot ping some hosts on the network, they are all up.
>>> Cannot resolve via DNS which is one of the hosts that I cannot ping, internal nor external DNS hosts.
>>> Unplugging the NICS and plugging them back in will then not allow pining if the default d=gateway.
>>>
>>> When cold booting it somewhat works, some hosts are pingable while others are not.
>>>
>>> When restarting the network service via /etc/init.d/network, nothing is pingable.
>>>
>>> Here are my configs;
>>>
>>> ifcfg-bond0
>>> DEVICE=bond0
>>> USERCTL=no
>>> BOOTPROTO=none
>>> ONBOOT=yes
>>> IPADDR=10.0.0.10
>>> NETMASK=255.255.0.0
>>> NETWORK=10.0.0.0
>>> TYPE=Unknown
>>> IPV6INIT=no
>>>
>>> ifcfg-eth0
>>> DEVICE=eth0
>>> BOOTPROTO=none
>>> ONBOOT=yes
>>> MASTER=bond0
>>> SLAVE=yes
>>> USERCTL=no
>>>
>>> ifcfg-eth1
>>> DEVICE=eth1
>>> BOOTPROTO=none
>>> ONBOOT=yes
>>> MASTER=bond0
>>> SLAVE=yes
>>> USERCTL=no
>>>
>>> /etc/modprob.d/bonding.conf
>>> alias bond0 bonding
>>> options bond0 mode=5 miimon=100
>>>
>>> Bonding worked great in Centos 5.x, not so well for me in Centos 6.2.
>>>
>>> My goal is to get this working under bridging for KVM, I can only imagine the nightmare seeing I can't get a simple bond to work!
>>>
>>> Any guidance is golden.
>>>
>>> - aurf
>>>
>>> _______________________________________________
>>> CentOS mailing list
>>> CentOS at centos.org
>>> http://lists.centos.org/mailman/listinfo/centos
>>>
>>>
>> I spent two months on bonding two nics inside a box to a bridge in the box.
>> There is a bug, very prominent in fedora bugzillas about it.
>> You cannot do it without some modification. Libvirt loses some vms, no
>> way to make it work that I know of
>> except for those suggested changes which I did not try.
>>
>> if you look in your libvirt logs you will see xml bond errors....and
>> thus impossible to do inside of the box.
>> this only applies if the bonded nics and bridge are all in the same box,
>>
>> also, the options should no longer go in bonding.conf, but in the bridge
>> file itself.
>>
>> in all my testing all vms worked except the one assigned vnet0, that
>> always got 'lost'...
>> however, any attempt by the vm to send a signal outside to the net,
>> would cause it to be found again..
>>
>> this bug is not fixed in 6 or in latest fedora when i last
>> checked....there are self made patches in fedora bugzilla though.
> Hi Bob,
>
> WOW, ok cool.
>
> I will simply do 2 bridges and allocate some of my guests to either for a sort of manual load balancing.
>
> Good info to know.
>
> Really appreciate the feedback.
>
> - aurf
> _______________________________________________
> CentOS mailing list
> CentOS at centos.org
> http://lists.centos.org/mailman/listinfo/centos
>
>
and dont forget a lot of the bond modes require a second device like a 
switch...i think 0 and 6 were the ones I tried

I really wanted the bond, but decided to blow some ips and just use 
extra bridges to try to balance.
It was not fun at all finding this bug...lol

I think the reason they do not want to get into it is alomost everyone 
uses bond/bridges outside of the single server and it is not an 
issue...and to rewrite and debug all that for a few of us that are crazy 
to do such single box bridge bond, well, we in't gonna see that 
according to the bugzilla responses I saw...still, one can hope