Nicira NVP plugin create extension lswitches when port number on a bridged (flat) network reaches upper limit (configured by max_lp_per_bridged_ls in nvp.ini).
However, in Havana version, when extension lswitches are created for "flat" networks, it use wrong transport zone binding.
Expected: use same transport zone binding as specified by the network.
Result: use default transport zone binding configured by default_tz_uuid and default_transport_type in nvp.ini.
This bug is introduced in Havana version.
Root cause:
in Havana, tz_uuid was generated in a new function _convert_to_nvp_transport_zones(), and it checks if mpnet.SEGMENTS is not set for the network, the default tz_uuid and transport type are used:
if (network and not attr.is_attr_set(
network.get(mpnet.SEGMENTS))):
return [{"zone_uuid": cluster.default_tz_uuid,
"transport_type": cfg.CONF.NVP.default_transport_type}]
For a bridged network without VLAN, this mpnet.SEGMENTS is not set. So the function returns with the default binding rather than use the tz binding of the specified network.
Fixed in master/juno already by https:/