On Mon, Oct 15, 2012 at 03:29:40PM -0700, Maru Newby wrote:
> I'm running XCP 1.6 (which ships with ovs 1.4.2), and noticed that
> the ovs-xapi-sync script removes data from the Interface table for a
> standalone tap.  Specifically, the 'iface-id' and 'iface-status'
> keys are removed from external-ids if a tap does not have a
> corresponding vif.  Reviewing the script, it is not obvious why this
> modification is required.

It's because we've never seen a standalone tap device under XCP at
Nicira.  Standalone vifs, sure, but I was under the impression that
XenServer/XCP never created a tap without a vif.

> For background, I'm trying to run Devstack+Quantum under XCP, and
> specifically the Quantum dhcp agent running under dom0.  This
> requires adding a standalone tap port on the Quantum bridge and
> running a dhcp server on that port.  Then Quantum (specifically the
> L2 agent) needs to be able to identify the tap port via the iface-id
> so as to be able to apply a vlan tag to limit the scope of the dhcp
> server.  Identification of the port is not possible if the iface-id
> is cleared by the sync script.

So what causes the tap to be standalone?  Does XCP 1.6 never create
vif/tap pairs, only a vif or a tap?  Is this DHCP agent a special case
somehow?  Does the DHCP agent run in a VM?  (And if not, where does
the iface-id come from?)

> Would it be possible to simply ignore standalone tap ports?  Or, if
> the current processing of the sync script needs to be maintained,
> would it be reasonable to mark a given interface as 'do-not-sync' to
> prevent it from being modified?

Let's figure out what makes sense, then do that.
_______________________________________________
discuss mailing list
[email protected]
http://openvswitch.org/mailman/listinfo/discuss

Reply via email to