Hi all,

I'd like to ask for review of two related fixes for BFD status correctness
in OVN, submitted week ago. Both address ways the NB/SB BFD status column
can permanently disagree with reality, which impacts ECMP static-route
nexthop selection.

northd: Mark BFD sessions down when no chassis can run them.
Issue: https://github.com/ovn-org/ovn/issues/320
PR:    https://github.com/ovn-org/ovn/pull/321
When the only chassis with the bridge mapping for a gateway port goes away,
no ovn-controller runs the BFD session and nothing ever invalidates the
stale "up" — traffic keeps hashing to the dead nexthop indefinitely. northd
now derives session ownership the same way ovn-controller does
(chassisredirect binding) and marks the session down when no owner exists.
Also fixes SB bfd.chassis_name being permanently empty for distributed
gateway ports.

controller: Reconcile BFD status with local session state.
Issue: https://github.com/ovn-org/ovn/issues/322
PR:    https://github.com/ovn-org/ovn/pull/323
pinctrl only writes status on state-machine transitions, so a stale or
externally-modified value is never corrected until a real flap. Adds a
guarded, rate-limited reconciliation in bfd_monitor_run().

Both reproduce on 24.03 and current main;

Thanks,
Premysl
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev

Reply via email to