CSIT Sporadic failures - First port on compute (DHCP Server MAC) has ELAN drop rule instead of output to tunnel

https://jenkins.opendaylight.org/releng/job/netvirt-patch-test-current-boron/229/

The failure is getting DHCP server IP from the console log:
https://logs.opendaylight.org/releng/jenkins092/netvirt-csit-1node-openstack-newton-nodl-v2-gate-stateful-boron/119/archives/log.html.gz#s1-s1-s1-t7

This is just a side effect of metadata failing, causing the DHCP server IP output to be delayed.
The actual cause is the metadata server (DHCP server) MAC address on compute 2 in the ELAN pipeline has an output:drop rule, instead of outputting to the tunnel.

cookie=0x803138a, duration=337.546s, table=51, n_packets=220, n_bytes=15264, priority=20,metadata=0x138a000000/0xffff000000,dl_dst=fa:16:3e:70:f7:f0 actions=drop
cookie=0x803138b, duration=317.800s, table=51, n_packets=100, n_bytes=9592, priority=20,metadata=0x138b000000/0xffff000000,dl_dst=fa:16:3e:6b:e0:64 actions=set_field:0x2->tun_id,output:3

Note that on the same compute, there are other rules pointing to that tunnel, implying that only the first port on the compute suffers from this, probably relates to auto tunnel configuration.

(30.0.0.2 fa:16:3e:70:f7:f0 - fails)
(40.0.0.2 fa:16:3e:6b:e0:64 - works)

Note that pings also fail later from the DHCP to these VMs.