On 7/2/25 11:48 AM, Xavier Simonart via dev wrote: > If BFD is down between two gateways in a chassis group, while both gateways > are alive, both ovn-controller were fighting to claim the router port. > There was a 500 msec, hardcoded, grace time preventing both controller > to run at 100% CPU. > Depending of sb dynamics, either gw1 or gw2 ended up claiming the port > most of the time. > For instance: > - gw1 claims the port at time T > - gw2 claims the port at time T (+ few micros) > - gw1 postpone the claims. > - gw1 claims the port at time T+500. > - gw2 claims the port at time T+500 (+ few micros) > - gw1 postpone the claims > However, computes still send their traffic towards the highest priority > gateway. > > This patch guarantees that, if the highest priority gateway is alive, it > always claims > the port. > This patch does not change the behavior if the highest priority chassis > "dies". > > Note: computes send their traffic towards ha gateways based on BFD status and > priority, i.e. > without ovn-controller and sb handling. > We do not want to base the traffic on which gateway claims the router port, > as this would require > ovn-controller & sb handling, potentially delaying migration of the traffic. > > Signed-off-by: Xavier Simonart <[email protected]> > ---
Recheck-request: github-robot-_Build_and_Test CI infra issue: Err:21 http://archive.ubuntu.com/ubuntu noble/main amd64 gcc-multilib amd64 4:13.2.0-7ubuntu1 Unable to connect to archive.ubuntu.com:http: [IP: 91.189.91.82 80] _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
