[ovs-dev] [PATCH v2] ofp-actions: Add "ingress" and "egress" options to "sample" action.

Ben Pfaff blp at ovn.org
Wed Nov 30 16:50:59 UTC 2016


On Mon, Nov 28, 2016 at 04:47:30PM +0100, Simon Horman wrote:
> On Sat, Nov 26, 2016 at 01:36:00PM -0800, Ben Pfaff wrote:
> > On Fri, Nov 25, 2016 at 03:46:55PM +0100, Simon Horman wrote:
> > > Hi Ben,
> > > 
> > > On Wed, Nov 23, 2016 at 11:18:14PM -0800, Ben Pfaff wrote:
> > > > Before Open vSwitch 2.5.90, IPFIX reports from Open vSwitch didn't include
> > > > whether the packet was ingressing or egressing the switch.  Starting in
> > > > OVS 2.5.90, this information was available but only accurate if the action
> > > > included a port number that indicated a tunnel.  Conflating these two does
> > > > not always make sense (not every packet involves a tunnel!), so this patch
> > > > makes it possible for the sample action to simply say whether it's for
> > > > ingress or egress.
> > > > 
> > > > This is difficult to test, since the "tests" directory of OVS does not have
> > > > a proper IPFIX listener.  This passes those tests, plus a couple that just
> > > > verify that the actions are properly parsed and formatted.  Benli did test
> > > > it end-to-end in a VMware use case.
> > > 
> > > I am wondering if it is necessary to add NXAST_RAW_SAMPLE3 instead
> > > of just adding a direction field to NXAST_RAW_SAMPLE2 and updating
> > > branch-2.6. Is it too late for this change to change the behaviour of
> > > NXAST_RAW_SAMPLE2?
> > 
> > I'm wary of doing that because 2.6.0 was released months ago and because
> > trying to use this new with an older OVS would cause it to be silently
> > ignored.  (If we had been careful with the definition of the action, as
> > I try to be these days, then it would have provoked an
> > OFPERR_NXBRC_MUST_BE_ZERO error, and I'd be happy to just extend the
> > action.)
> > 
> > There's not too much cost to inventing NXAST_RAW_SAMPLE3, so I'm
> > comfortable with it.
> 
> Understood. I agree it makes sense to be cautious.
> 
> Acked-by: Simon Horman <simon.horman at netronome.com>

Thanks a lot for the review.  I applied this to master.


More information about the dev mailing list