Guru Shetty <guru....@gmail.com> wrote on 07/26/2016 10:22:30 PM: > From: Guru Shetty <guru....@gmail.com> > To: Ryan Moats/Omaha/IBM@IBMUS > Cc: Guru Shetty <g...@ovn.org>, ovs dev <dev@openvswitch.org> > Date: 07/26/2016 10:22 PM > Subject: Re: [ovs-dev] [PATCH] ovn-controller: squelch expected > duplicate flow warnings > > > On Jul 26, 2016, at 5:30 PM, Ryan Moats <rmo...@us.ibm.com> wrote:
> Guru Shetty <g...@ovn.org> wrote on 07/26/2016 06:05:47 PM: > > > From: Guru Shetty <g...@ovn.org> > > To: Ryan Moats/Omaha/IBM@IBMUS > > Cc: ovs dev <dev@openvswitch.org> > > Date: 07/26/2016 06:06 PM > > Subject: Re: [ovs-dev] [PATCH] ovn-controller: squelch expected > > duplicate flow warnings > > > > On 26 July 2016 at 15:54, Ryan Moats <rmo...@us.ibm.com> wrote: > > > > > > > > Guru Shetty <g...@ovn.org> wrote on 07/26/2016 03:54:29 PM: > > > > > From: Guru Shetty <g...@ovn.org> > > > To: Ryan Moats/Omaha/IBM@IBMUS > > > Cc: ovs dev <dev@openvswitch.org> > > > Date: 07/26/2016 03:54 PM > > > Subject: Re: [ovs-dev] [PATCH] ovn-controller: squelch expected > > > duplicate flow warnings > > > > > > On 24 July 2016 at 10:07, Ryan Moats <rmo...@us.ibm.com> wrote: > > > In the physical processing of ovn-controller, there are two > > > sets of OF flows that are still fully recalculated every cycle: > > > > > > Flows that aren't associated with any logical flow, and > > > Flows calculated based on multicast groups > > > > > > Because these flows are recalculated fully each cycle, full > > > duplicates of existing OF flows are created and the OF management > > > code in ovn-controller pollutes the logs with false positive > > > warnings about repeated duplicates. > > > > > > As a short term measure, ignore full duplicates for both of > > > these types of flows, but still warn if the action changes > > > (as that is not expected and may be indicative of a problem). > > > > > > Signed-off-by: Ryan Moats <rmo...@us.ibm.com> > > > > > > I also noticed that "commit 70c7cfef188b5ae9940abd5 (ovn-controller: > > > Add incremental processing to lflow_run and physical_run)" causes > > > load balancing system unit tests to fail. A little debugging shows > > > that groups are getting deleted when new flows are added. My hunch > > > is that this is likely because 'desired_groups' in ofctl_put gets > > > deleted in every run. But in the next run, it does not get updated > > > as we no longer process all flows. > > > > That's going to take persisting the desired_groups data. > > > > I can take a shot if you'd like, just give me the link to the > > patch set that includes the load balancing system unit tests > > and I'll see what I can do to make it right ... > > > > It already exists in the OVN repo. tests/system-ovn.at > > Ack and verified that it is failing - I'll take a deeper look > later tonight/tomorrow and see what I can make work. > > Thanks much. > > (Just to make sure you have the environment right, you should have > the right kernel modules with conntrack support installed on your > machine. On master, it will only work on pre 4.6 kernels if there is > no ovs kernel module already instslled from upstream kernel. To make > it work, you should either remove upstream kernel modules or install > a /etc/depmod.d/openvswitch.conf to override upstream one. On 4.6 > and above it should not matter as upstream kernel module has > conntrack support. > > You can make sure that you get the tests working before the said > commit so that you dont go on a wild goose chase.) Mitigation patch is at http://patchwork.ozlabs.org/patch/653068/ for review. In my previous message, I incorrectly stated that the above patch didn't handle flow modifications correctly. It actually does. _______________________________________________ dev mailing list dev@openvswitch.org http://openvswitch.org/mailman/listinfo/dev
