In our lab environment, we found these weird log in /var/log/messages

kernel: vxlan: non-ECT from IP_ADDRESS with TOS=0x2

This is kind of annoying messages and also, the vxlan tunnel networking would go wrong randomly. e.g. router: 192.168.100.1 vm1: 192.168.100.7 vm1: 192.168.100.8 private subnet connection between VMs is good (192.168.100.7 <---> 192.168.100.8)

but VM to router is no good ( 192.168.100.7 x 192.168.100.1) ( 192.168.100.8 x 192.168.100.1)

I've checked overlay network ip at compute node & network node is good (at least icmp) & neutron-l3-agent is up & running( other router can work normally)

So this is weird to us, why these troubled VMs could reach other but router?

Any advise is much appreciated.