initscripts, xen and bonding

[Axel Thimm]

Hi,

I'm having a hard time getting bonding to work on FC5t3. It looks like
the bond0 interface bond0 is renamed to pbond0 by xen [*] and the
enslaving does not work anymore then. It works if I log on the
terminal and ifenslave the devices to pbond0 instead of bond0 after
the boot process has finished.

But I want this to be handled automatically during reboots, so I tried
assigning the bonding slaves to pbond0 as a master in the
ifcfg-eth{0,1} scripts, but that doesn't work.

What is causing the bond0 rename and should this renaming be
considered by initscript? Is this a user error or should I file this
against bugzilla? xen or initscripts?

Thanks!

[*] The only reference I found was a Japanese site and google's
    translation of it.

   
http://translate.google.com/translate?hl=en&sl=ja&u=http://hoop.e...
-- 
Axel.Thimm at ATrpms.net

[Bill Nottingham]

Axel Thimm (Axel.Thimm(a)ATrpms.net) said: 
>  I'm having a hard time getting bonding to work on FC5t3. 
Not surprising. The way Xen does its bridging breaks bonding
fairly badly.

It's possible it could be fixed in the xen scripts, but I don't
think it's been looked at.

Bill

[Axel Thimm]

On Fri, Feb 24, 2006 at 06:28:20PM -0500, Bill Nottingham wrote:
>  Axel Thimm (Axel.Thimm(a)ATrpms.net) said: 
>  > I'm having a hard time getting bonding to work on FC5t3.
>  
>  Not surprising. The way Xen does its bridging breaks bonding
>  fairly badly.
>  
>  It's possible it could be fixed in the xen scripts, but I don't
>  think it's been looked at. 
are the xen scripts called by initscripts? E.g. will this logic be
encapsulated in such a way that an ifdown/ifup on bond0 outside the
boot sequence would do the right thing w/o initscripts knowing
anything more about xen, or would initscripts still require some
special xen handling?

I haven't looked to deeply into xen, so I may be missing the obvious
here.
-- 
Axel.Thimm at ATrpms.net

[Bill Nottingham]

Axel Thimm (Axel.Thimm(a)ATrpms.net) said: 
>  are the xen scripts called by initscripts? 
No, they are their own entirely separate pile of goo.

Bill

[Axel Thimm]

On Fri, Feb 24, 2006 at 06:28:20PM -0500, Bill Nottingham wrote:
>  Axel Thimm (Axel.Thimm(a)ATrpms.net) said: 
>  > I'm having a hard time getting bonding to work on FC5t3.
>  
>  Not surprising. The way Xen does its bridging breaks bonding
>  fairly badly.
>  
>  It's possible it could be fixed in the xen scripts, but I don't
>  think it's been looked at. 
I now jumped into the xen networking concepts. These make sense, but
there are two issues:

a) bonding should be performed onto the "physical" bonding interface,
   which xen renames to pbond0. Still the initscripts only work with
   MASTER=pbond0 if called *after* completion of the boot process.

b) even if a) is fixed and requires MASTER=pbond0 setting, the
   switching from a xen kernel to a non-xen kernel the
   configuration in the initscripts would be bogus (pointing to
   a nonexisting pbond0), and the bonding device would be w/o slaves.

I don't understand why a) happens, seems to be dependent on the order
of the steps performed. Maybe the bonding/enslaving (at boot time)
happens before xen renames the bond0 device and later on the enslaving
is wrong. Reruning the ifup scripts sees the renamed devices and
performs properly. That's a theory at least.

I'm not sure how to solve b). Ideally one would like to not have xen
specific actions in initscripts, but otherwise one risks cutting of
the system of the net completely when upgrading kernels and booting
into the wron one. Probably not something one wants when setting up
bonding to ensure network availability. ;)

Should I file a bugzilla? xen or initscripts?
-- 
Axel.Thimm at ATrpms.net

[Bill Nottingham]

Axel Thimm (Axel.Thimm(a)ATrpms.net) said: 
>  I don't understand why a) happens, seems to be dependent on the
> order
>  of the steps performed. Maybe the bonding/enslaving (at boot time)
>  happens before xen renames the bond0 device and later on the enslaving
>  is wrong. 
Correct.

>  I'm not sure how to solve b). Ideally one would like to not have
> xen
>  specific actions in initscripts, but otherwise one risks cutting of
>  the system of the net completely when upgrading kernels and booting
>  into the wron one. Probably not something one wants when setting up
>  bonding to ensure network availability. ;)
>  
>  Should I file a bugzilla? xen or initscripts? 
Ideally, xen shouldn't be mucking with networking that much. It
does it because of some supposed race.

Bill


>  -- 
>  fedora-test-list mailing list
>  fedora-test-list(a)redhat.com
>  To unsubscribe: 
>  https://www.redhat.com/mailman/listinfo/fedora-test-list

[Axel Thimm]

Hi,

I'm retrying to get bonding and xen working with FC5:

https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=189473

The summary is that

o xend at boot time does not attach anything to xenbr, the network is
  other than the added broken xenbr unaltered. So at least booting
  into kernel-xen0 doesn't hook you off the wire anymore (as it was
  with FC5t3)

o after calling xend restart the bridge is almost properly set up,
  only the slaves of the bond interface get lost in the process

Anyone know how to fix/workaround these issues?

Thanks!
-- 
Axel.Thimm at ATrpms.net

I'm using Fedora 4 and wound up fixing it manually for now, if we decide 
to keep xen we'll write a boot script to do it.

Basically, both bond0 and pbond0 must be active at the same time, so leave 
your bond0 setup as is. Copy /etc/sysconfig/network-scripts/ifcfg-bond0 to 
ifcfg-pbond0 and rename the Device value (not sure this step is 
necessary).

Reboot and on bootup, do:
sudo ifconfig pbond0 (IP address) network (255.255.255.x) broadcast (your 
address)
sudo ifenslave pbond0 eth0
sudo ifenslave pbond0 eth1

The 3 addresses above are the same as used by bond0.




Axel Thimm <Axel.Thimm(a)ATrpms.net&gt; 
Sent by: fedora-xen-bounces(a)redhat.com
04/20/06 08:15 AM
Please respond to
fedora-xen(a)redhat.com


To
fedora-xen(a)redhat.com
cc

Subject
[Fedora-xen] Re: initscripts, xen and bonding






Hi,

I'm retrying to get bonding and xen working with FC5:

https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=189473

The summary is that

o xend at boot time does not attach anything to xenbr, the network is
  other than the added broken xenbr unaltered. So at least booting
  into kernel-xen0 doesn't hook you off the wire anymore (as it was
  with FC5t3)

o after calling xend restart the bridge is almost properly set up,
  only the slaves of the bond interface get lost in the process

Anyone know how to fix/workaround these issues?

Thanks!
-- 
Axel.Thimm at ATrpms.net
--
Fedora-xen mailing list
Fedora-xen(a)redhat.com
https://www.redhat.com/mailman/listinfo/fedora-xen

[Axel Thimm]

On Thu, Apr 20, 2006 at 02:15:36PM +0200, Axel Thimm wrote:
>  Hi,
>  
>  I'm retrying to get bonding and xen working with FC5:
>  
>  https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=189473 
It's been almost a year and no response to the bugzilla. There was a
related bug on RHEL5, which was supposedly fixed in beta2.

The core issue is that renaming the bond device will lose its slaves
and they require re-enslaving in the xen scripts otherwise your
network stops working at all.

Can anyone confirm that bonding/xen really works on RHEL5? If so, can
we have the fix also for FC6?

>  The summary is that
>  
>  o xend at boot time does not attach anything to xenbr, the network is
>    other than the added broken xenbr unaltered. So at least booting
>    into kernel-xen0 doesn't hook you off the wire anymore (as it was
>    with FC5t3)
>  
>  o after calling xend restart the bridge is almost properly set up,
>    only the slaves of the bond interface get lost in the process
>  
>  Anyone know how to fix/workaround these issues?
>  
>  Thanks! 




-- 
Axel.Thimm at ATrpms.net