[Pkg-xen-devel] Bug#631102: Bug#631102: xen: Xen guests on Squeeze lose networking randomly

[Kevin Bowling]

On Mon, Jun 20, 2011 at 3:14 AM, Thomas Goirand <zigo at debian.org> wrote:
> On 06/20/2011 05:50 PM, Kevin Bowling wrote:
>> Debian Squeeze Dom0, up to date.
>>
>> Networking is handled by OS scripts, such that br0 and br1 are bridge interfaces.
>>
>> Without warning, the DomU (Ubuntu 10.04 LTS) loses inbound connectivity.  It tends to happen after several hours.  It doesn't seem to be affected by throughput and is triggering on very small ammounts (5MB in and out over that time).  Having active traffic doesn't help, it still dies.
>>
>> Updated the DomU to kernel 2.6.35 with no change.  I suspect the problem lies on the Dom0 side.

Bumped to 2.6.38, no change.  Also have a Squeeze guest with same
symptoms.  It's definitely on the Dom0 side (kernel, xen, bridging).

>
> My guess is that what's happening isn't what you say. It might well be
> due to your switch forgetting about the MAC address of your domU, if
> there's no network activity on it (at least that's my guess, and it did
> happen as well with Lenny and Debian as domU). I had the issue in many
> data centers/switches, and I wrote a small python script to fix it in a
> cron job. I have attached the script to this email.

It doesn't appear so.  The switch is managed and I see the MAC on the
right port.  I've disabled MAC aging completely on it for further
testing.

arp shows the correct default gateway on the DomU.

The "forgetting" appears to be in the Dom0 networking stack.  Pinging
the DomU from the Dom0 and then pinging various hosts sometimes
revives it.  Except, packets are lost to those hosts that I tried
pinging earlier.

No such problems like this exist on the Dom0 to the outside world.

Maddening!  Any more ideas to try?