Thanks for the quick follow up Ben, So we'll indeed try against latest versions to rule out the possibility of a bug that has been fixed already although I could not find any commit with such mention. We'll report back here.
At this moment, we can reproduce over and over within minutes. We've nailed down -- or at least we think -- to something related to race condition or memory overwrite between the time connection negotiation happens and the flow mod packet arrives. We've been able to run the stress test for hours when we used --flow-format=NXM+table_id as an argument to ovs-ofctl but eventually, we've hit the same error. Sagar is spending more time debugging this issue so maybe he'll be able to provide more information. On Fri, Jan 13, 2017 at 1:45 PM, Ben Pfaff <b...@ovn.org> wrote: > On Thu, Jan 12, 2017 at 03:54:42PM -0500, Samuel Jean via dev wrote: > > It seems that shelling out to ovs-ofctl very quickly can lead to bug > where > > it reports an OFPT_ERROR. > > > > We were able to constantly reproduce within minutes of running the above > > flow modifications on Unbutu. > > > > Any help, hints or guidance would be appreciated. I'd be happy to pursue > > some debugging that would be required to nail down the issue here. > > Thanks for the bug report and especially for the detailed reproduction > advice. > > I've now tried running this reproduction case against Open vSwitch from > latest master and against the latest versions from the 2.6.x and 2.5.x > branches, and I can't see any failures even after letting the script run > for a few minutes. > > Maybe you should try 2.5.1 or the latest from branch-2.5 and see if it > fixes the problem? And if not, then we'll have to figure out what's > different between your setup and mine. > _______________________________________________ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev
