unicast ARP problem

[Dominik Mostowiec]

On R2.20 we noticed strange unicast ARP behaviour.
2 VMs on separated compute nodes.
VM_SRC: 10.189.85.11(02:86:45:5e:06:38), VM_DST: 10.189.85.10(02:25:6b:67:72:e9)

tcpdump on VM_SRC:
tcpdump -i eth0 -e arp
13:41:41.694854 02:25:6b:67:72:e9 (oui Unknown) > 02:86:45:5e:06:38
(oui Unknown), ethertype ARP (0x0806), length 42: Request who-has
10.189.85.11 tell 10.189.85.10, length 28
13:41:42.696888 02:25:6b:67:72:e9 (oui Unknown) > 02:86:45:5e:06:38
(oui Unknown), ethertype ARP (0x0806), length 42: Request who-has
10.189.85.11 tell 10.189.85.10, length 28
13:41:43.698889 02:25:6b:67:72:e9 (oui Unknown) > 02:86:45:5e:06:38
(oui Unknown), ethertype ARP (0x0806), length 42: Request who-has
10.189.85.11 tell 10.189.85.10, length 28
13:41:45.690163 02:25:6b:67:72:e9 (oui Unknown) > Broadcast, ethertype
ARP (0x0806), length 42: Request who-has 10.189.85.11 tell
10.189.85.10, length 28
13:41:45.690203 02:86:45:5e:06:38 (oui Unknown) > 02:25:6b:67:72:e9
(oui Unknown), ethertype ARP (0x0806), length 42: Reply 10.189.85.11
is-at 02:86:45:5e:06:38 (oui Unknown), length 28

3 x request to mack 02:86:45:5e:06:38 who-has 10.189.85.11 and no response.
And after that 1 request to all "who-has 10.189.85.11" and proper response.

On VM_DST:
tcpdump -i eth0 -e arp
13:41:41.886815 02:86:45:5e:06:38 (oui Unknown) > 02:25:6b:67:72:e9
(oui Unknown), ethertype ARP (0x0806), length 42: Reply 10.189.85.11
is-at 02:86:45:5e:06:38 (oui Unknown), length 28
13:41:42.888807 02:86:45:5e:06:38 (oui Unknown) > 02:25:6b:67:72:e9
(oui Unknown), ethertype ARP (0x0806), length 42: Reply 10.189.85.11
is-at 02:86:45:5e:06:38 (oui Unknown), length 28
13:41:43.890797 02:86:45:5e:06:38 (oui Unknown) > 02:25:6b:67:72:e9
(oui Unknown), ethertype ARP (0x0806), length 42: Reply 10.189.85.11
is-at 02:86:45:5e:06:38 (oui Unknown), length 28

3x response on previous 3 requests.
There is no request, only responses.
If vrouter send the response of request 'who-has 10.189.85.11' - the
response is on wrong VM.
If VM_DST OS should send this - on OS we don't see any request.

I think the response is send to wrong VM.

Additional info:

Contrail agent version:
 dpkg -l | grep contrail-vrouter-agent
ii contrail-vrouter-agent 2.20-64 amd64 OpenContrail vrouter agent

In attachements: video where you can see this behaviour

Screen from another network(compute nodes) with ping (first tab), tcpdump on SRC VM (second tab), tcpdump on DST VM (third tab): https://dysk.onet.pl/link/TKgXi

[OpenContrail Admin (ci-admin-f)]

bug update...

[Hari Prasad Killi (haripk)]

Based on the addresses shown, the response is going to the correct VM. Please explain the problem seen.