Hi Ales
Thanks for the patch.
Patch looks good to me.
I am just wondering whether the "Fixes" tag is the most appropriate. It
looks to me that the first part fixes 47203af82aba ("mac-cache: Send
broadcast re-ARP probes after a while.")
And the second part fixes 47203af82aba ("mac-cache: Send broadcast re-ARP
probes after a while.").
Other than that:
Acked-by: Xavier Simonart <[email protected]>
Thanks
Xavier
On Mon, Sep 14, 2026 at 3:34 PM Ales Musil via dev <[email protected]>
wrote:
> The probing tests used fixed delays and absolute probe counts around the
> MAC binding aging deadline. Under load, controller statistics could
> cross the deadline before a reply was processed, causing an active
> binding to be removed.
>
> Synchronize with stale, used MAC cache flows while the binding still
> exists, and compare probe counts relative to their observed values.
> Reply to each neighbor as soon as its probe arrives so that serial
> waits cannot consume the aging window.
>
> Fixes: 1e4fa9f174f1 ("mac-cache: Do not send probes for inactive MAC
> Bindings.")
> Assisted-by: GPT-5.6-sol, OpenCode
> Signed-off-by: Ales Musil <[email protected]>
> ---
> tests/ovn.at | 47 ++++++++++++++++++++++++++++++-----------------
> 1 file changed, 30 insertions(+), 17 deletions(-)
>
> diff --git a/tests/ovn.at b/tests/ovn.at
> index 8acfecb7b..448e0818b 100644
> --- a/tests/ovn.at
> +++ b/tests/ovn.at
> @@ -37738,14 +37738,12 @@ dump_arp 1 00:00:00:00:10:00 00:00:00:00:10:1b
> 10.10.10.1 10.10.10.101 00:00:00:
> dump_ns 00:00:00:00:10:1b 00:00:00:00:10:00 fd11::65 fd11::1 fd11::65 >
> ucast_v6_65.pkt
> OVS_WAIT_UNTIL([test $($PYTHON "$ovs_srcdir/utilities/ovs-pcap.in"
> hv1/vif1-tx.pcap | \
> grep -Fc "$(cat ucast_v4_101.pkt)") -ge 2])
> +v4_ts=$(fetch_column Mac_Binding timestamp ip=10.10.10.101)
> +send_garp hv1 vif1 2 00:00:00:00:10:1b 00:00:00:00:10:00 10.10.10.101
> 10.10.10.1
> +
> OVS_WAIT_UNTIL([test $($PYTHON "$ovs_srcdir/utilities/ovs-pcap.in"
> hv1/vif1-tx.pcap | \
> grep -Fc "$(cat ucast_v6_65.pkt)") -ge 2])
> -
> -# The neighbours answer, refreshing the rows in place (resetting
> -# arp_attempts). Confirm the timestamps advanced.
> -v4_ts=$(fetch_column Mac_Binding timestamp ip=10.10.10.101)
> v6_ts=$(fetch_column Mac_Binding timestamp ip=\"fd11::65\")
> -send_garp hv1 vif1 2 00:00:00:00:10:1b 00:00:00:00:10:00 10.10.10.101
> 10.10.10.1
> send_na hv1 vif1 00:00:00:00:10:1b 00:00:00:00:10:00 fd11::65 fd11::1
> OVS_WAIT_UNTIL([test $(fetch_column Mac_Binding timestamp
> ip=10.10.10.101) -gt $v4_ts])
> OVS_WAIT_UNTIL([test $(fetch_column Mac_Binding timestamp
> ip=\"fd11::65\") -gt $v6_ts])
> @@ -37869,6 +37867,25 @@ dump_icmp() {
> echo $packet
> }
>
> +mac_cache_flow_is_stale() {
> + local flow=$1
> + local stats field n_packets idle_age
> +
> + stats=$(as hv1 ovs-ofctl dump-flows br-int \
> + "table=OFTABLE_MAC_CACHE_USE,$flow" | grep 'n_packets=')
> + n_packets=
> + idle_age=
> + for field in $stats; do
> + field=${field%,}
> + case $field in
> + n_packets=*) n_packets=${field#n_packets=} ;;
> + idle_age=*) idle_age=${field#idle_age=} ;;
> + esac
> + done
> +
> + test "$n_packets" -gt 0 && test "$idle_age" -ge "$aging_th"
> +}
> +
> aging_th=5
> net_add n1
> sim_add hv1
> @@ -37928,17 +37945,15 @@ OVS_WAIT_UNTIL([$(ovs-ofctl dump-flows br-int
> table=OFTABLE_MAC_BINDING | \
>
> send_icmp_echo_req hv1 public 00:00:00:00:10:00 00:00:00:00:10:1a
> 192.168.20.2 192.168.10.100
>
> -# Wait until ovn-controller sends the ARP request.
> -OVS_WAIT_UNTIL([test $(grep -c "Sending ARP/ND.*ip: 192.168.20.2"
> hv1/ovn-controller.log) -eq 1])
> -send_icmp_echo_req hv1 public 00:00:00:00:10:00 00:00:00:00:10:1a
> 192.168.20.2 192.168.10.100
> -
> # Now drop ICMP echo reply in order to force OVN to arp the mac binding
> entry
> # for the distributed router lr.
> check ovn-nbctl --wait=hv acl-add join from-lport 1000 'inport ==
> "join-lr" && icmp' drop
>
> -sleep $((aging_th / 2))
> +OVS_WAIT_UNTIL([mac_cache_flow_is_stale 'ip,nw_src=192.168.20.2'])
> +
> +probe_count=$(grep -c "Sending ARP/ND.*ip: 192.168.20.2"
> hv1/ovn-controller.log)
> send_icmp_echo_req hv1 public 00:00:00:00:10:00 00:00:00:00:10:1a
> 192.168.20.2 192.168.10.100
> -OVS_WAIT_UNTIL([test $(grep -c "Sending ARP/ND.*ip: 192.168.20.2"
> hv1/ovn-controller.log) -eq 2])
> +OVS_WAIT_UNTIL([test "$(grep -c "Sending ARP/ND.*ip: 192.168.20.2"
> hv1/ovn-controller.log)" -gt "$probe_count"])
>
> check ovn-nbctl --wait=hv acl-del join
> wait_row_count mac_binding 0 ip="192.168.10.100" logical_port="gw-public"
> @@ -37949,17 +37964,15 @@ wait_row_count mac_binding 1
> mac=\"00:00:00:00:30:00\" ip=\"fd12::2\"
>
> send_icmp6_echo_req hv1 public 00:00:00:00:10:00 00:00:00:00:50:01
> fd12::2 fd11::64
>
> -# Wait until ovn-controller sends the NS request.
> -OVS_WAIT_UNTIL([test $(grep -c "Sending ARP/ND.*ip: fd12::2"
> hv1/ovn-controller.log) -eq 1])
> -send_icmp6_echo_req hv1 public 00:00:00:00:10:00 00:00:00:00:50:01
> fd12::2 fd11::64
> -
> # Now drop ICMPv6 echo reply in order to force OVN to send NS for the lr
> mac binding entry.
> check ovn-nbctl --wait=hv acl-add join from-lport 1010 'inport ==
> "join-lr" && ip6 && icmp6.type == 0x87' allow
> check ovn-nbctl --wait=hv acl-add join from-lport 1000 'inport ==
> "join-lr" && icmp' drop
>
> -sleep $((aging_th / 2))
> +OVS_WAIT_UNTIL([mac_cache_flow_is_stale 'ipv6,ipv6_src=fd12::2'])
> +
> +probe_count=$(grep -c "Sending ARP/ND.*ip: fd12::2"
> hv1/ovn-controller.log)
> send_icmp6_echo_req hv1 public 00:00:00:00:10:00 00:00:00:00:50:01
> fd12::2 fd11::64
> -OVS_WAIT_UNTIL([test $(grep -c "Sending ARP/ND.*ip: fd12::2"
> hv1/ovn-controller.log) -eq 2])
> +OVS_WAIT_UNTIL([test "$(grep -c "Sending ARP/ND.*ip: fd12::2"
> hv1/ovn-controller.log)" -gt "$probe_count"])
>
> wait_row_count mac_binding 0 ip=\"fd12::2\" logical_port="gw-join"
>
> --
> 2.55.0
>
> _______________________________________________
> dev mailing list
> [email protected]
> https://mail.openvswitch.org/mailman/listinfo/ovs-dev
>
>
_______________________________________________
dev mailing list
[email protected]
https://mail.openvswitch.org/mailman/listinfo/ovs-dev
