On Mon, Dec 8, 2025 at 1:40 PM Han Zhou <[email protected]> wrote: > > > > On Mon, Dec 8, 2025 at 10:22 AM Han Zhou <[email protected]> wrote: > > > > > > > > On Wed, Nov 26, 2025 at 1:02 AM Ales Musil <[email protected]> wrote: > > > > > > > > > > > > On Mon, Nov 24, 2025 at 3:47 PM Ales Musil <[email protected]> wrote: > > >> > > >> > > >> > > >> On Mon, Nov 24, 2025 at 9:45 AM Ales Musil <[email protected]> wrote: > > >>> > > >>> > > >>> > > >>> On Mon, Nov 17, 2025 at 6:15 AM Han Zhou <[email protected]> wrote: > > >>>> > > >>>> This patch introduces flow-based tunnels as an alternative to > > >>>> traditional port-based tunnels, significantly reducing tunnel port > > >>>> count > > >>>> in large deployments. > > >>>> > > >>>> Flow-based tunnels use shared ports (one per tunnel type) with > > >>>> options:local_ip=flow and options:remote_ip=flow. OpenFlow flows > > >>>> dynamically set tunnel endpoints using set_field actions, reducing port > > >>>> count to O(T) where T is the number of tunnel types. > > >>>> > > >>>> The feature is experimental, and controlled by > > >>>> external_ids:ovn-enable-flow-based-tunnels (default: false). > > >>>> > > >>>> Some known limitations: > > >>>> - IPsec is not supported > > >>>> - BFD between tunnel endpoints is not supported, thus HA chassis not > > >>>> supported. > > >>>> > > >>>> Assisted-by: Cursor, with model: Claude Sonnet 4.5 > > >>>> Signed-off-by: Han Zhou <[email protected]> > > >>>> --- > > >>> > > >>> > > >>> Hi Han, > > >>> > > >>> thank you for the v3. Seems like ovn-k jobs are failing for this > > >>> patch. Let's try to re-run them again. Other than that it looks good. > > >>> > > >>> Recheck-request: github-robot-_ovn-kubernetes > > > > > > > > > > > > Hi Han, > > > > > > the recheck failed too, I'm afraid there is regression could you please > > > check > > > that out? > > > > > > > Indeed. I debugged and finally found a bug which was introduced by > > refactoring the function add_tunnel_ingress_flows in physical.c. An extra > > ofpbuf_clear(ofpacts) was added, thus the actions for the flows that handle > > ICMP packet-too-large were wrong, leading to e2e case "Pod to pod TCP with > > low MTU" failing. I fixed it with the below one line change: > > ---------------- > > diff --git a/controller/physical.c b/controller/physical.c > > index f6de77430..43e1c2dbc 100644 > > --- a/controller/physical.c > > +++ b/controller/physical.c > > @@ -369,7 +369,6 @@ add_tunnel_ingress_flows(const struct chassis_tunnel > > *tun, > > ofpacts, hc_uuid); > > > > /* Set allow rx from tunnel bit */ > > - ofpbuf_clear(ofpacts); > > put_load(1, MFF_LOG_FLAGS, MLF_RX_FROM_TUNNEL_BIT, 1, ofpacts); > > put_resubmit(OFTABLE_CT_ZONE_LOOKUP, ofpacts); > > ----------------- > > > > With this fix (and also rebased on top of main), github actions have passed. > > I can merge it with this oneline change, or please let me know if v4 is > > necessary. > > > I just realized that I didn't get your official Ack yet, and I just sent v4: > https://patchwork.ozlabs.org/project/ovn/patch/[email protected]/ > > cc Mark and Dumitru
Hi Han, I'm unlikely to get to it today, but I will have a look Tuesday or Wednesday. Thanks! _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
