Open Stack

Hi all,

I testing upgrade from Havana to Icehouse.  This is changing a number
of networking related things so not sure which is my problem.  My goal
is to maintain my current production state jumbo frames network.

In Havana production I'm using openvswitch plugin, but not
LibvirtHybridOVSBridgeDriver, so tap devices are directly on ovs
bridges.  I have veth_mtu set in [agent] section of
ovs_neutron_plugin.ini and all devices get created with the correct
MTU.

In Icehouse I've migrated to ML2, still with ovs and added
OVSHybridIptablesFirewallDriver so now I have the qbr|qvb|qvo devices
as well.  I copied the veth_mtu into the [agent] section of
ml2_conf.ini (which is what openvswitch-agent init file is sourcing on
startup) but all interfaces are being created with standard 1500 MTU.
This include the tap devices on the network node for dhcp-agents
(which are associated with qbr devices) and the tap,qbr,qvb,qvo
devices associated with instances on compute nodes (the OVS devices
int-br, eth1-br, etc all have jumbo MTU).

Is this my configuration, a bug, or expected behaviour?  Strongly
hoping for the first option...

Thanks,
-Jon