On Wed, Jan 29, 2025 at 11:20:49AM -0500, Aaron Conole wrote: > Vipul Ashri via dev <[email protected]> writes: > > > Hi Adrián, > > > > Thanks for reviewing and reaching the depth of this issue. > > > > Right now addressing your first two comments with [PATCH v2], > > > > but as per third comment I agree we should have right reimplementation > > of flush API which need some aggressive changes but I find > > reimplementation is optional for current stability and can be deferred > > as we already calling right APIs to smartly cleaning right dp flows > > before dustruct() Apis e.g. close_dpif_backer() etc. > >
I don't see that, maybe I'm missing something. close_dpif_backer() does not seem to remove any dp flow. What it does is close the dpif, i.e: stop revalidator/handler threads, delete ukeys, etc. And that is very intentional because we want to be able to gracefully restart OVS without interrupting existing traffic. Your change, in essense, breakes the "--cleanup" in "ovs-appctl exit [--cleanup]". Thanks. > > We have also tested this patch with our inhouse deployments, and we > > currently started using it as downstream patch with our latest > > releases. We find no issue with the patch so far. > > I don't think this is a very helpful response. Except in the most > extreme cases, we should be able to add tests for specific scenarios. > "Worked on my system when I tested it," doesn't apply to everyone's > systems. > > Also, the CC list in my client showed: > CC: <[email protected]> > > Adrian's email should be > [email protected] > > So I fixed it in this reply. > Thanks > > BR > > Vipul > > > > _______________________________________________ > > dev mailing list > > [email protected] > > https://mail.openvswitch.org/mailman/listinfo/ovs-dev > _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
