[ovs-discuss] ovs-dpdk unicast flooding issue

Satish Patel satish.txt at gmail.com
Thu Oct 28 04:28:52 UTC 2021


Folks,

I have deployed ovs-dpdk using legacy neutron-openvswitch-agent. My
openstack version is wallaby and the OVS version follows. My issue is
I have multiple VMs running on this compute node and interestingly if
I run tcpdump on vm-A then I can see traffic of vm-B which is hosted
on the same hypervisor. look like OVS not learning arp properly and
flooding frames to all ports.

I have two compute nodes compute-A and compute-B.

compute-A has vm-A1 and vm-A2
compute-B has vm-B1 and vm-B2

I am running tcpdump on vm-A1 and trying to ping vm-B1 from vm-A2 then
I can see that icmp traffic on vm-A1 (that means ovs flooding frame to
each ports) and this is happening with only outgoing traffic.

when i look into fdb tables then found mac age is always zero so look
like constantly trying to refresh mac age.

$ ovs-appctl fdb/show br-vlan | grep fa:16:3e:95:f5:26
    4  2002  fa:16:3e:95:f5:26    0

I have tried following but no luck

$ ovs-vsctl set bridge br-vlan  other_config:mac-aging-time=1800

$ ovs-ofctl dump-flows br-vlan
 cookie=0xd86865ad3ff15101, duration=553359.704s, table=0,
n_packets=18043751, n_bytes=5862441509,
priority=4,in_port="phy-br-vlan",dl_vlan=12
actions=mod_vlan_vid:2002,NORMAL
 cookie=0xd86865ad3ff15101, duration=700758.555s, table=0,
n_packets=0, n_bytes=0, priority=2,in_port="phy-br-vlan" actions=drop
 cookie=0xd86865ad3ff15101, duration=700758.560s, table=0,
n_packets=83925014, n_bytes=9304779835, priority=0 actions=NORMAL

I found this bug but not sure this is my issue or not -
https://bugs.launchpad.net/neutron/+bug/1732067

followings are my versions:

$ uname -a
Linux compute-A  5.4.0-88-generic #99-Ubuntu SMP Thu Sep 23 17:29:00
UTC 2021 x86_64 x86_64 x86_64 GNU/Linux

$ ovs-vswitchd -V
ovs-vswitchd (Open vSwitch) 2.13.3
DPDK 19.11.7


More information about the discuss mailing list