ARP packets received on physical interface loop back out on the same interface with Open vSwitch 2.6.1 in Red Hat OpenStack Platform 10

Issue

ARP packets received on physical interface loop back out on the same interface with Open vSwitch 2.6.1 in Red Hat OpenStack Platform 10

Open vSwitch "leaks" incoming ARP requests on bond1 back out on bond1. This happens randomly. In an environment with ca. 50 compute nodes, this issue can be observed several times per hour in the entire cloud network.

We came to the conclusion that outgoing faulty packets are likely actually first received by the compute then sent back to the switch.

A tcpdump with direction out give the following outbound capture:

No.     Time                          Source                Destination           Protocol Length Info
      1 2019-06-12 12:11:43,742681    IETF-VRRP-VRID_8e     HewlettP_<MAC>     ARP      64     10.109.142.126 is at 00:00:5e:00:01:8e
      2 2019-06-12 12:11:43,824268    IETF-VRRP-VRID_8e     HewlettP_<MAC>     ARP      64     10.109.142.126 is at 00:00:5e:00:01:8e

When actually we received the same packets few milli/micro seconds earlier as shown by a tcpdump with direction in on the same interface:

 No.     Time                          Source                Destination           Protocol Length Info
  69926 2019-06-12 12:11:43,742520    IETF-VRRP-VRID_8e     HewlettP_<MAC>     ARP      64     10.109.142.126 is at 00:00:5e:00:01:8e
  69943 2019-06-12 12:11:43,824243    IETF-VRRP-VRID_8e     HewlettP_<MAC>     ARP      64     10.109.142.126 is at 00:00:5e:00:01:8e