restart of quantum ovs plugin and network node reboot

Three Node Setup, Quantum + OVS + GRE, simple Network, no router no floating, one network + subnet.

Initial network creating works fine, Instances get an IP, can ping each other and so on. After restarting the quantum-plugin-openvswitch-agent on the network node, I cant connect to existing VMs anymore. Trying to create new instances result in dhcp not providing an IP.

Running tcpdump against all the qbr- qvo- qvb-<something> devices showed that arp requests do get answered by the compute node, but never reach the network node. Packets seem to be dropped by the "action=drop" line and do not match the other flow any longer.

# ovs-ofctl dump-flows br-tun
NXST_FLOW reply (xid=0x4):
 cookie=0x0, duration=5313.915s, table=0, n_packets=886, n_bytes=41676, priority=3,tun_id=0x1,dl_dst=01:00:00:00:00:00/01:00:00:00:00:00 actions=mod_vlan_vid:1,output:1
 cookie=0x0, duration=5313.974s, table=0, n_packets=1644, n_bytes=205097, priority=4,in_port=1,dl_vlan=1 actions=set_tunnel:0x1,NORMAL
 cookie=0x0, duration=5313.856s, table=0, n_packets=182, n_bytes=28539, priority=3,tun_id=0x1,dl_dst=fa:16:3e:12:d1:f3 actions=mod_vlan_vid:1,NORMAL
 cookie=0x0, duration=6213.217s, table=0, n_packets=1427, n_bytes=149242, priority=1 actions=drop

A workaround was to add another default flow via ovs-ofctl to br-tun.

# ovs-ofctl add-flow br-tun action=Normal

This only works in case packets get lost on their way back via br-tun, but it seems that there are other issues after restart of network node / quantum ovs plugin. Sometimes it only helps to recreate the whole netconfig from scratch.

Any idea what goes wrong here and how to get to the core of that issue?.