[ovs-dev] VTEP gateway and OVN workflow

[Darrell Ball]

On 12/23/15, 1:34 PM, "dev on behalf of Guru Shetty" <dev-bounces at openvswitch.org on behalf of guru at ovn.org> wrote:

>Hello All,
> I just looked at the OVN workflow for implementing VTEP schema (L2 only)
>and at first glance it feels wrong. There is possibly a reason for the way
>it has been implemented, but this is how I see it.
>
>The current workflow (from a test case in ovn-controller-vtep.at):
>
>1. In the VTEP GW device, create a logical switch (via vtep-ctl).
>
>2.  In VTEP GW device, create a binding that associates vlan+physical_port
>to that logical switch. This creates an entry in ovn-sb's chassis table's
>'vtep_logical_switches' column for that VTEP GW.
>
>3. You are then supposed to create a logical switch in OVN NB and create a
>logical port for that switch and then set options:vtep-logical-switch to
>point to the logical switch that you created in the VTEP GW.
>
>IMO the above is wrong. The correct workflow should just have been:
>
>1. In OVN NB, create a logical port (for a particular VTEP physical port),
>specify the VTEP GW chassis to which this logical port belongs to and also
>specify the vlan.

IMO
The OVN NB logic/DB should not link a logical port/endpoint to a vlan or any other
“physical construct” and other “physical constructs" are possible; the OVN NB logic should remain at the logical level.

The mapping of logical endpoints to physical endpoints should remain at the chassis layer.



>
>The above information should go down to VTEP GW's database and it should
>create a logical switch record and also create the vlan bindings from
>physical port to logical switch.
>_______________________________________________
>dev mailing list
>dev at openvswitch.org
>http://openvswitch.org/mailman/listinfo/dev