Comment 3 for bug 1681979

The agent_boot_time is untouched, so default of 180. This is well more than the duration of OVS agent boot to stabilize/all ports to be re-updated.

I think the problem here is, there are not too many overall ports on this system, but there are at least a couple of networks. This node is not a hypervisor - just used for Glance image lib and network node services (l3 agent, dhcp/dnsmasq).

Therefore the only port updates received after OVS restart are for router ports and dhcp ports. We also see this regularly on compute hosts where there are not too many VMs, but more than 1 or 2 networks (4 networks, around 1-2 VMs per network). Some networks get no flows, some get only UC flows as per examples above.

Increasing agent_boot_time to something obscenely large doesn't seem like a proper fix.