On Wed, Jul 27, 2016 at 1:44 AM, Babu Shanmugam <bscha...@redhat.com> wrote:
> > > On Wednesday 27 July 2016 06:43 AM, Russell Bryant wrote: > > On Tue, Jul 26, 2016 at 6:46 AM, <bscha...@redhat.com> wrote: > >> From: Babu Shanmugam <bscha...@redhat.com> >> >> Commit 263064a (Convert binding_run to incremental processing.) removed >> the usage >> of all_lports from binding_run, but it is infact used in the context of >> the caller, >> especially by update_ct_zones(). >> >> Without this change, update_ct_zones operates on an empty set always. >> >> Signed-off-by: Babu Shanmugam <bscha...@redhat.com> >> > > Ouch. This is a really bad regression. If I understand correctly, we're > not setting a ct zone ID for any logical ports. All are just using the > default zone of 0. > > Yes Russell, your understanding is correct. > > We should think about a good way to test OVN's use of conntrack zones to > ensure that entries end up in separate zones for separate ports. A good > test for that may require userspace conntrack support, though. > Another test we could do now would be looking at the flows in table 0 and > ensuring that the input flow for each port has a different conntrack zone > ID assigned. That feels like kind of a hack, though. > > I agree that we need more test cases. I could not spend much time to > figure out a proper approach for a test case. I will have a look at it. > FYI, the patch doesn't apply for some reason. It's small enough that I can do it manually, though. I can take a look at this tomorrow to add the missing bits that I pointed out unless you beat me to it. I know you've got a few other things you're working on, as well. -- Russell Bryant _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
