[lxc-devel] Cloud instance with reduced MTU, container uses 1500, connections stall

[Serge Hallyn]

Quoting Andreas Hasenack (andreas at canonical.com):
> Hi,
> 
> I'm not sure where or how this should be fixed.
> 
> I have a case where openstack was juju deployed with neutron networking,
> and we needed to set instance-mtu = 1454 in the neutron-gateway charm.
> 
> Instances launched in the cloud get that 1454 MTU set in their eth0 devices
> and life is good.
> 
> If, however, I create a container inside an instance (for example, via juju
> deploy --to lxc:0), that container gets eth0 set with an MTU of 1500. And
> that makes almost all connections stall, and the deployment fails.
> 
> Now, who should set the default MTU for containers to be 1454 in this case?
> Or, more explicitly, to mimic the MTU of the "host" (in this case, the
> instance)? juju? lxc-create?

Hi,

Does adding

lxc.network.mtu = 1454

to /etc/lxc/default.conf on the instance before the containers are
created fix the issue?

Setting the mtu automatically in lxc is in general a good idea, but it
won't catch all cases and may not catch yours.  Are you using lxcbr0, or
a br0 with eth0 bridged to it?  (Actually if using lxcbr0 then I'd like
to think that the kernel would fragment the lxcbr0 traffic has it hits
your eth0 which would just slow your traffic down)

-serge