On Wed, Apr 12, 2017 at 11:01:50AM -0700, Joe Stringer wrote: > On 12 April 2017 at 10:53, Flavio Leitner <[email protected]> wrote: > > On Wed, Apr 12, 2017 at 10:04:50AM -0700, Joe Stringer wrote: > >> On 12 April 2017 at 08:22, Flavio Leitner <[email protected]> wrote: > >> > On Fri, Apr 07, 2017 at 04:13:08PM +0300, Roi Dayan wrote: > >> >> From: Paul Blakey <[email protected]> > >> >> > >> >> Usage: > >> >> # to dump all datapath flows (default): > >> >> ovs-dpctl dump-flows > >> >> > >> >> # to dump only flows that in kernel datapath: > >> >> ovs-dpctl dump-flows type=ovs > >> >> > >> >> # to dump only flows that are offloaded: > >> >> ovs-dpctl dump-flows type=offloaded > >> > > >> > > >> > It should return an error if the type= is wrong/unavailable. > >> > >> I wonder if we should also add a piece in verbose mode (-m) at the end > >> of each flow that says something like "type=software", "type=hardware" > >> (which should match the name of the ovs-dpctl dump-flows type > >> argument)? > > > > Makes sense. Perhaps offloaded isn't the correct term, neither > > software or hardware, because it seems TC can decide to not offload to > > the HW and that is not visible to OVS. As a result, you could pass > > "type=offloaded" and see rules that are actually in TC software. > > Perhaps I missed something. > > wrt 'offloaded', seems fine to me. If it's offloaded from regular OVS > path to TC path, even in software, this is still more accurately > describing what happened than to say 'hardware' since the tc-policy > could be sw-only. So 'ovs' and 'offloaded' seem OK.
OK, well, I thought about saying 'tc' instead to be clear of what is going on, but then I realized this wouldn't work if we enable offloading in userspace DP. I guess you're right and 'offloaded' seems OK. > The main point I was highlighting above was that if you just dumped > the flows in regular mode, you don't have any indication where the > flows are being executed. Maybe that's fine, but if ovs-dpctl > dump-flows ends up printing both sources it might be easier to notice > something odd going on if there were an indication where the flow came > from. The tradeoff is we end up printing even more information on the > commandline, hence why I figured it might be better to only display it > in verbose mode. Yup, I agree with you. It seems useful. -- Flavio _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
