On 2/26/24 17:40, Mikhail Sennikovsky wrote: > Hi Ilya and everyone, > > I'm interested in obtaining further insight into the technical > rationale behind the automatic removal of ingress qdiscs for OVS > internal ports. > I.e. it's no doubt that ovs internal ports are managed by ovs, however > since internal ports are something different than external ports, I > would expect they could/should be managed differently. > > From the "outside the ovs" standpoint how would the the ovs internal > port network interface differ from a veth pair network interface whose > peer is enslaved in the ovs (via external port)? > > I.e. why should qdisc configuration for the ovs internal port network > interface have some additional limitations compared to that of the > veth interface whose peer is enslaved in ovs?
OVS removes qdiscs from veth interfaces the same way it removes them from internal ports. There is no special treatment, AFAIK. That was the point of my reply. Ingress qdiscs will be removed from any port attached to OVS. Egress qdiscs will also be removed, unless linux-noop QoS type is configured. Best regards, Ilya Maximets. > > Kind regards, > Mikhail > > >> On 2/22/24 10:03, Reshma Sreekumar wrote: >>> Hello, >>> >>> I see similar behaviour for *ovs-internal* ports as well. My understanding >>> is that, >>> ovs-internal ports aren't /managed/ by Openvswitch (Please correct me if >>> I'm wrong) ? >> >> It would be strange if ovs-internal ports weren't managed by OVS. >> So, they are managed. >> >>> >>> In this case, is it expected for the *ingress qdisc on ovs-internal ports* >>> as well >>> to be managed by ovs ? >> >> It is expected for any ports attached to OVS. >> >> Basically, if you see an interface in the output of 'ovs-vsctl show', >> OVS will manage qdiscs on that interface. >> >> Bets regards, Ilya Maximets. _______________________________________________ discuss mailing list disc...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-discuss
