[ovs-discuss] OVS Faucet tutorial broken

Brad Cowie brad at cowie.nz
Tue Aug 21 23:08:02 UTC 2018


Hi Ben,

On 21 August 2018 at 04:41, Ben Pfaff <blp at ovn.org> wrote:

> > Since the tutorial recommends ovs-sandbox, ports will always be down due
> to
> > the use of the netdev-dummy datapath which skips the linux network stack.
> > This is fine though as we can use "ovs-appctl
> netdev-dummy/set-admin-state
> > up"
> > to force the ports to come up and this will keep faucet happy.
>
> Hmm.  I wonder whether there should be a different default for dummy
> ports.  I don't know a good reason to force users to "up" them by hand.
> Do you have a preference?
>

I think the right thing to do for netdev_dummy is to default to "up" on
dummy ports
and the user can admin them down with the ovs-appctl command if they want.


> > I am still seeing some issues where even when the ports are up the
> > ofproto/trace -generate commands don't seem to be triggering packet_ins
> > towards faucet so learning isn't happening.
>
> Do you have an example?  I'd be pleased to investigate.
>

I finally tracked this down to this commit:

https://github.com/openvswitch/ovs/commit/d39ec23de38464ee35b3098b9f6c5f06d5191015

So I think if we send to controller and pass the packet along the OpenFlow
pipeline
OVS will optimise away the send to controller (slow path) action.

I see there is a "debug_slow" action that can be used to restore the old
behaviour
but this seems a little icky for us to implement in faucet.

I also note someone already reported this issue here:

https://github.com/openvswitch/ovs-issues/issues/145

Brad
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mail.openvswitch.org/pipermail/ovs-discuss/attachments/20180822/6983fa17/attachment.html>


More information about the discuss mailing list