[ovs-discuss] possible bug when falling back to the normal switch

[tim huang]

Hi there,

I am doing some test for my controller to handle the situation when some of
the switches in the topology lose the connection to the controller.  The
part of the topology will be something like this:

S1(DPID:11,port4)——(DPID:23,port3)S13(DPID:23,port2)—— (DPID:22,port3)S12

At first, all the switches are connected to the controller. All the
switches controlled by controller send LLDP and special broadcast to all
their ports.Then, I set the S13 to be standalone mode, using command to
delete the controller for S13. S1, S12 will still send LLDP and  special
broadcast for all their ports from controller’s packet_out . After falling
back, I found the S13 can receive the special broadcast for both sides, but
will only forward one side broadcast to the other. For example, S13 can
receive S12’s broadcast and forward it to the S1, and it also can receive
the broadcast from S1 but won’t forward it to S12. I used the wireshark to
monitor the interfaces to get the conclusion. It doesn’t happen every
time.  But the frequency is very high.

What’s more interesting is if I fall back the switch and delete the
controller before I start the controller,  it will not happen at all.

Is there any idea about this?

BTW: I used 1.9.3 and also tried with 2.3.1, the situation is the same.


-- 
Thanks
Tim
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://openvswitch.org/pipermail/ovs-discuss/attachments/20150117/c253f287/attachment-0002.html>