On 9/28/26 2:07 AM, Jacob Tanenbaum via dev wrote: > On restart, ovn-controller monitored the whole cluster's > Advertised_Route table at startup, and only ran its kernel route > sync once all of it had been read. The whole-cluster pull added > measurable startup latency in large deployments, and the guard was > coarser than it needed to be: the controller only ever deletes > routes that it installed itself, and it only needs the advertised > routes of the local route-exchange datapaths to know which ones > those are. > > Monitor no Advertised_Route rows at startup. Once local datapaths > are known the monitor condition is already scoped to them, so the > dump the controller waits for is small. > > Defer the route-exchange kernel sync until the controller can be > confident that dump is complete: a local route-exchange datapath > exists, the current Advertised_Route condition covers those > datapaths, and the condition seqno shows the dump was delivered. > The kernel routes persist across the deferral, so none are lost. > The datapath set changing or a southbound reconnect re-arms the > gate. > > The condition covering the local datapaths is what keeps this > safe. Waiting only on the condition seqno could arm the sync > against the initial empty condition, and the first sync would > then delete the routes installed before the restart. > > The dynamic-routing system test now checks that the deferred sync > is enabled after a restart and that the installed OVN routes > survive a --restart of the controller. > > Reported-at: https://redhat.atlassian.net/browse/FDP-3992 > Assisted-by: Qwen3.8-27B-FP8, opencode > Signed-off-by: Jacob Tanenbaum <[email protected]> > ---
Hi Jacob, I just wanted to point out that CI is very red with this patch, e.g.: https://github.com/ovsrobot/ovn/actions/runs/36362128048/job/108743883752 ERROR: All 120 tests were run, 22 failed unexpectedly. ## -------------------------------------- ## ## system-dpdk-testsuite.log was created. ## ## -------------------------------------- ## Please send `tests/system-dpdk-testsuite.log' and all information you think might help: To: <[email protected]> Subject: [ovn 26.09.90] system-dpdk-testsuite: 214 216 218 220 222 224 226 228 230 232 233 234 264 268 270 272 295 298 300 302 304 306 failed It seems to be breaking VRF management from ovn. I'm marking this as "changes requested" in patchwork. Regards, Dumitru _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
