[ovs-discuss] Behaviour change in OpenVSwitch

[Tony van der Peet]

The test is basic.OutputExact in oftest, which is on GitHub at:

https://github.com/floodlight/oftest/blob/master/tests-1.3/basic.py

Yes, it does have a barrier between adding the flow and verifying packets,
so your idea would seem to be not only a total fix, but also the most
elegant way to do it.

To answer Alex in the same email:

1 - Yes, a flow remove would also fix it, although I'm not sure whether
this would make the timing of deleting the dpif flows any faster. But since
we're trying to use oftest without too many mods, and *not* doing the flow
remove actually tests something else (that changing a flow actually changes
the flow), which may in fact be the actual test case we are testing, we
really want this to work without the flow remove.

2 - Not sure exactly where I put the sleep in all cases (I intend to fix
our copy of oftest next week at work and may come up with a different
method anyway), but I think from memory that it's after the call to flow
add.

Tony

On Sat, Jun 6, 2015 at 3:18 AM, Ben Pfaff <blp at nicira.com> wrote:

> On Fri, Jun 05, 2015 at 07:02:13PM +1200, Tony van der Peet wrote:
> > Turns out that the only behaviour change is a slight timing change. If I
> > put a 0.2s sleep into the outer loop of the test I can make is pass
> > reliably. In other words, the flows are still being deleted correctly,
> it's
> > just that the test is so quick in changing the flows and expecting to be
> > able to test the new flow that it's seeing a failure.
> >
> > In the real world, there would of course be much less chance of this sort
> > of coordination between data and flow changes, so I am satisfied that
> > OpenVSwitch continues to operate in the same high quality way that I have
> > always experienced.
>
> Does the test have an OpenFlow barrier between the flow table change and
> the packet?  I'm considering a change to OVS that would make OVS wait
> until datapath flows are up-to-date before replying to a barrier
> request.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20150607/8152dc25/attachment-0002.html>