[libvirt-users] TCP reset (RST) networking issue (maybe was 'Sporadic network issues")

[Brian Fallik]

Hi,

I'm seeing a very similar problem to the thread "Sporadic networking
issues" [1].  In our setup, some TCP operations receive unexpected RST
packets when executed on the guest.  I can see these RST packets via
wireshark but can't explain their source.

We originally uncovered this when some 'git clone' operations failed
with "Broken Pipe" in the guests.  I've been able to reproduce the
same failures using scp and 'svn checkout'.  The failure occurs 80-90%
of the time and 'svn checkout' seems to be the quickest to fail.  The
same operations on the host are always successful.

My setup is a Lucid amd64 host with two guests (Lucid and Hardy, both
i386, using kvm).  Networking is configured as described in "Routed
network config".  One anomoly I notice is that our configuration is
missing the 'mode' and 'dev' attributes of the 'forward' node.  Not
sure if that makes a difference, especially since networking seems to
mostly work.

I've tried this with both virtio and e1000 drivers with no difference.

I tried Jeremy's suggestion to limit the host to start a single guest
but that did not seem to have an impact.  One thing I wonder is if I'm
resetting the libvirt correctly to make this change.  I did not reboot
the host, but I did shutdown the guests, shutdown the default network,
verified that no kvm was running and no tap/bridge devices existed,
undefined the Hardy domain, then restarted the default network and
started the Lucid kvm.  Does that sound correct?

Anyone have suggestions for further debugging or pointer(s) to
existing related bugs?  My search didn't turn up anything obvious.
I'm very eager to resolve this issue since it's blocking our ability
to use the VMs.

Thanks,
brian

1 - https://www.redhat.com/archives/libvirt-users/2010-August/msg00012.html