Hi Alexandra,
Unfortunately, we have introduced a regression in 26.03 since
1b4058b9162c ("northd: Process external arps on ha chassis.") [0].
The problem is that now we drop all ARP requests if they're received
from a localnet port that's part of a switch connected to a router with
a DGP on a chassis where the DGP is not bound.
That makes sense for router owned IPs but it breaks functionality when
the targeted IP is actually owned by a logical switch port bound to the
chassis where we process the ARP request.
I think this topology depicts the situation the best:
(fabric) -- localnet-port -- LS --- DGP(chassis2) -- LR
|
|
VM (chassis1)
If on chassis1 (where the VM actually runs and is bound) we receive an
ARP request for VM's IP, we currently drop it because DGP is bound to
chassis2 and REGBIT_EXT_ARP == 1.
More details in [1].
Would you happen to have some time to look into this issue?
Thanks,
Dumitru
[0] https://github.com/ovn-org/ovn/commit/1b4058b
[1] https://redhat.atlassian.net/browse/FDP-3767
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
