[Server-devel] Notes on replacing bridging with bonding

[Jerry Vonau]

Jerry Vonau wrote:
> Martin Langhoff wrote:
>> On Mon, Sep 8, 2008 at 7:43 PM, Alexander Dupuy <alex.dupuy at mac.com> wrote:
>>> I have placed TYPE=Bonding in the ifcfg-bond0 config files, but this is not
>>> needed for Fedora 7 or later (it doesn't hurt to have it, though)
>> well, I grepped network-functions and ifup-eth and none of the TYPE
>> checks are for bonding anything. The check is done against the output
>> of modprobe. So if it's going to be ignored, I'll save the
>> corresponding bytes (and potential confusion later).
>>
> ifup does a check for type=Bridge
> 
Bonding can be handled via VLAN=yes and PHYDEV=msh0 in the ifcfg files 
for the bond device, ifup uses those values to setup the bonding in a 
hotplug environment. The slave (msh0) should have SLAVE=yes MASTER=bond0 
in its file. Hotplug works also, bond0 doesn't need to have its slave 
present to become active, plug the slave in, becomes active and bonding 
does the right thing. Bye, bye dummy interface.

>>>> # cat /etc/modprobe.d/xs-bonding
>>>> alias bond0 bonding
>>>>
>>> A modprobe.d/ directory - that's a nice trick!  I wasn't aware of this, so
>>> just added some lines to the /etc/modprobe.conf file:
>> that's what the most modern doco could find suggested. And the
>> ifup-eth src is clearly looking at the output of modprobe, so there's
>> something to it...
>>
> well, udev loads the modules on boot, for configured devices. "network" 
> checks to see if the required module for the nic is loaded, if missing 
> loads it.
> 
Think we can just dump a file in /etc/modprobe.d/ with the bonding 
options needed. I've been using bonding mode=active-backup miimon=100" 
to test with, any idea what we may be using here?

>>> everything you have looks perfectly
>>> reasonable.
>> thanks for sanity checking. One thing I have not been able to _test_
>> is that I want one of the bonding devices to have 2 IP addresses,so
>> I've setup lanbond0 and lanbond0:1 - with eth1 as a slave to lanbond0.
>>
>> lanbond0:1 comes up automatically with lanbond0 . I am not 100%
>> certain that eth1 - if / when it comes up - will get the 2 addresses.
>>
> Sorry, I'll get to mocking up the layout later.
> 
Based on the quick mockup, I'll say no, the slaves should not have an 
ipaddress assigned to them.

Jerry