[Openstack] instance networking fails after upgrade to Juno

Harm Weites harm at weites.com
Sun Oct 26 12:44:33 UTC 2014


After upgrading from Icehouse to Juno all networking fails, neutron and
nova are building hybrid bridges so it seems. This is RDO on CentOS 7.

Situation on Icehouse:
One controller node and one compute node with nova-compute and
neutron-openvswitch-agent. There is only Openvswitch/ML2 using GRE, no
linuxbridges. All instances (tap interfaces) are patched to the vswitch
and there is regular connectivity and metadata access.

After the upgrade to Juno:
Openvswitch/ML2 is creating a vswitch but besides that there also are
several linuxbridges created. And since those are not patched on to the
vswitch, instances are unreachable.

My controller node does have several tap devices on the vswitch, no

Nova is responsible for creating the linuxbridges, but I have no clue
why it suddenly wants to go hybrid - and apparently fails doing that.
Nova is calling plug_ovs_hybrid.

>From nova.conf on controller and compute:
 linuxnet_interface_driver = nova.network.linux_net.LinuxOVSInterfaceDriver
 libvirt_vif_driver = nova.virt.libvirt.vif.LibvirtGenericVIFDriver

[root at computenode ~]# brctl show
bridge name     bridge id               STP enabled     interfaces
qbr68933310-a9          8000.000000000000       no
qbr9c0fa825-f9          8000.000000000000       no

The tap device is patched to one of those, but those bridges aren't
listed with the vswitch.


More information about the Openstack mailing list