On Mon, Jul 28, 2014 at 9:44 PM, Ben Pfaff <[email protected]> wrote: > On Fri, Jul 25, 2014 at 07:29:24PM +0400, Andrey Korolyov wrote: >> Assuming we have trivial case: libvirt, its QoS limits (for which >> appropriate rules are executed for each port) and more than one VM >> sharing the same OVS bridge. Port configurations are initially empty >> and addition going through libvirt` internal command execution by >> calling ovs-vsctl binary, so no QoS configuration is performed through >> existing vswitchd mechanisms. >> >> Since OVS decides to flush all disciplines every time when port added >> or removed from the switch, one will not be able to keep more than one >> port with externally set tc disciplines. >> >> At least this bug affects 0fe1d7f39de9836fea01c560a6fdbfd1405096ea >> (2.1.3 effectively) in Openflow mode, though probably cover area is >> wider. FAQ suggests tc as a fallback method for complex >> configurations, so I can guess that the bug is definitely a bug but >> not an expected behavior. > > It does sound like a bug. Do you have time to fix it?
Hi Ben, I`ll take a detailed look a bit later. The problem probably is in incorrect netdev infrastructure calling in the tc_add_del_ingress_qdisc() CCing colleague who may issue the fix earlier than me. _______________________________________________ discuss mailing list [email protected] http://openvswitch.org/mailman/listinfo/discuss
