Open Stack

On Jan 12, 2014, at 8:29 AM, Zhuang Yuxin <yuxin.zhuang at gmail.com> wrote:

> Dear all,
> 
> With ML2 plugin, multiple mechanism drivers can be used simultaneously. Each mech driver may require different vif type implementation, e.g. tap device as backend or sriov VF passthru. If I take it correct, it's possible for two hosts on a network to use different mech drivers, supporting different type of vif. What if a vm is migrated to a host where the mech driver and vif type are different from the source host? I'm wondering if there is any deployment requirement to avoid this kind of situation or any code to check this.
> 
I’m not sure if this is currently what happens, but what *should* happen is
that if port binding succeeds on the remote host, then the migration will
also succeed. For port binding to succeed, a MechanismDriver on the
remote host will need to satisfy the binding requirements (e.g. segment
type and ID, possibly security groups, etc.). I don’t think this has been
tested yet, however, but it’s something which would be good to try
and have tests for in Tempest I think.

Thanks,
Kyle

> Thanks so much,
> Yuxin