Xen devel - Oct 2005 - Seeing bridge issues, how do I tell what domU is using what interface?

I seeing an issue sporadically where a Dom does not see the network.
In once case it was triggered by running SPF on a Dom, in that scenario
putting the DomU''s interface in promiscuous mode fixes it. I have hit
another condition where putting the DomU''s interface in promiscuous
mode
doesn''t fly. With some sporadic failure of domU creation still floating
this can be a real pita. 


Is their a way to get mapping of domU to vif ?
Log files don''t provide it.


Opened up an rfe on this a while back as well

http://bugzilla.xensource.com/bugzilla/show_bug.cgi?id=248


Regards,
Ted


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 10/14/05, Ted Kaczmarek <tedkaz@optonline.net> wrote:
> Is their a way to get mapping of domU to vif ?
> Log files don''t provide it.

nic@wuwei:~$ sudo xm network-list 2
(1 ((backend-id 0) (mac aa:00:00:88:02:40) (handle 1) (backend
/local/domain/0/backend/vif/2/1) (tx-ring-ref 521) (rx-ring-ref 522)
(event-channel 10)))


Replaces vif-list

--
Nicholas Lee
http://stateless.geek.nz
gpg 8072 4F86 EDCD 4FC1 18EF  5BDD 07B0 9597 6D58 D70C

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On Fri, 2005-10-14 at 10:59 +1300, Nicholas Lee wrote:
> On 10/14/05, Ted Kaczmarek <tedkaz@optonline.net> wrote:
> 
> > Is their a way to get mapping of domU to vif ?
> > Log files don''t provide it.
> 
> 
> nic@wuwei:~$ sudo xm network-list 2
> (1 ((backend-id 0) (mac aa:00:00:88:02:40) (handle 1) (backend
> /local/domain/0/backend/vif/2/1) (tx-ring-ref 521) (rx-ring-ref 522)
> (event-channel 10)))
> 
> 
> Replaces vif-list
> 
> --
> Nicholas Lee
> http://stateless.geek.nz
> gpg 8072 4F86 EDCD 4FC1 18EF  5BDD 07B0 9597 6D58 D70C
Well that makes it easy :-)

xm create is not creating the interface, cut and dry.


Also appears the domid matches the bridge interface, which is good and
bad. From a snmp standpoint it really sucks when you have moving targets
of instances to deal with, takes the simple part right out of the
protocol as well as adding additional overhead. Would be best if the
domU''s could be mapped to a particular bridge interface or at least
reuse the same one.


Definetly some timing related issue also,the faster you try to bring the
domu''s up the more of this failure I see. This also harkens back to the
IO contention issue I have when too many domU''s are coming from same IO
channel. Weird thing here is the domU gets created but the bridge
interface does not. Doesn''t seem to matter what the guest OS is, I run
Centos 4.1, FC4, Debian 3.1, Gentoo 2005 and Slackware 10.3 guests, the
Centos 4.1 and FC4 are my spins the other are jailtime.org, with the
Centos 4.1 being two fatter clients.


The Dom0 is using eth0 for all the domU''s with an eth1 as the gateway.

xen_changeset : Thu Oct 13 10:26:44 2005 +0100 7369:92c6021f23e4
Tyan 2462 SMP

Anyone have some 8 gig solid state and controllers they feel like
sending me :-)

Is their something in the test suite for such an animal?

I will try to put something together myself but don''t want to spend too
much time reinventing the wheel, puls I also have hardware limitations
that would affect such a testing scenario at present.

Regards,
Ted



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On Thu, Oct 13, 2005 at 07:25:10PM -0400, Ted Kaczmarek wrote:
> On Fri, 2005-10-14 at 10:59 +1300, Nicholas Lee wrote:
> > On 10/14/05, Ted Kaczmarek <tedkaz@optonline.net> wrote:
> > 
> > > Is their a way to get mapping of domU to vif ?
> > > Log files don''t provide it.
> > 
> > 
> > nic@wuwei:~$ sudo xm network-list 2
> > (1 ((backend-id 0) (mac aa:00:00:88:02:40) (handle 1) (backend
> > /local/domain/0/backend/vif/2/1) (tx-ring-ref 521) (rx-ring-ref 522)
> > (event-channel 10)))
> > 
> > 
> > Replaces vif-list
> > 
> > --
> > Nicholas Lee
> > http://stateless.geek.nz
> > gpg 8072 4F86 EDCD 4FC1 18EF  5BDD 07B0 9597 6D58 D70C
> 
> Well that makes it easy :-)
> 
> xm create is not creating the interface, cut and dry.
> 
> 
> Also appears the domid matches the bridge interface, which is good and
> bad. From a snmp standpoint it really sucks when you have moving targets
> of instances to deal with, takes the simple part right out of the
> protocol as well as adding additional overhead. Would be best if the
> domU''s could be mapped to a particular bridge interface or at
least
> reuse the same one.
Xend''s not really involved by this point -- the problem has been passed
on to
the hotplug scripts, and any admin or distro is free to fiddle with these, of
course.  We already have vif-nat, vif-bridge, and vif-route -- there''s
nothing
to stop you modifying these or replacing them with something that does more
intelligent bridging or whatever you need.

As for the bugs in the current implementations, it''s clear that there
are some
real bugs here, and we''d certainly appreciate it if you can figure out
whether
the scripts are even running, and if they are, where and why they are failing.
I think you are right that there is a timing or race issue here, but I
don''t
have a good handle on where it is.

Ewan.

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On Fri, 2005-10-14 at 00:42 +0100, Ewan Mellor wrote:
> On Thu, Oct 13, 2005 at 07:25:10PM -0400, Ted Kaczmarek wrote:
> 
> > On Fri, 2005-10-14 at 10:59 +1300, Nicholas Lee wrote:
> > > On 10/14/05, Ted Kaczmarek <tedkaz@optonline.net> wrote:
> > > 
> > > > Is their a way to get mapping of domU to vif ?
> > > > Log files don''t provide it.
> > > 
> > > 
> > > nic@wuwei:~$ sudo xm network-list 2
> > > (1 ((backend-id 0) (mac aa:00:00:88:02:40) (handle 1) (backend
> > > /local/domain/0/backend/vif/2/1) (tx-ring-ref 521) (rx-ring-ref
522)
> > > (event-channel 10)))
> > > 
> > > 
> > > Replaces vif-list
> > > 
> > > --
> > > Nicholas Lee
> > > http://stateless.geek.nz
> > > gpg 8072 4F86 EDCD 4FC1 18EF  5BDD 07B0 9597 6D58 D70C
> > 
> > Well that makes it easy :-)
> > 
> > xm create is not creating the interface, cut and dry.
> > 
> > 
> > Also appears the domid matches the bridge interface, which is good and
> > bad. From a snmp standpoint it really sucks when you have moving
targets
> > of instances to deal with, takes the simple part right out of the
> > protocol as well as adding additional overhead. Would be best if the
> > domU''s could be mapped to a particular bridge interface or at
least
> > reuse the same one.
> 
> Xend''s not really involved by this point -- the problem has been
passed on to
> the hotplug scripts, and any admin or distro is free to fiddle with these,
of
> course.  We already have vif-nat, vif-bridge, and vif-route --
there''s nothing
> to stop you modifying these or replacing them with something that does more
> intelligent bridging or whatever you need.
> 
> As for the bugs in the current implementations, it''s clear that
there are some
> real bugs here, and we''d certainly appreciate it if you can figure
out whether
> the scripts are even running, and if they are, where and why they are
failing.
> I think you are right that there is a timing or race issue here, but I
don''t
> have a good handle on where it is.
> 
> Ewan.
> 
Will see what I can dig up.

Regards,
Ted


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel