Ask Your Question
0

Virtual interfaces created by nova-network are faulty

asked 2014-07-11 10:05:15 -0500

updated 2014-07-31 03:54:17 -0500

Hi,

I'm running nova-network VlanManager mode on a Debian compute node, connected to a VMware vCenter cluster.

The physical internal interface (referenced in /etc/nova/nova.conf as vlan_interface) is connected to a trunk port-group. As requested, this interface as no IP configuration, and is in promiscious mode. Nova-network successfully create port-group on the VMware vSwitch for each vlans.

As you know, when the first instance of a network is deployed, nova-network create a bridge and a virtual interface respectively labelled brIDVLAN & vlanIDVLAN. The virtual interface is enslaved to the bridge, so we have the following bridge table : C:\fakepath\Capture.PNG

The route table is also updated automatically : C:\fakepath\route table.PNG

However though, the compute node can't ping the VMs. The bridge interface has an Ip configuration wich act as a gateway for the vlan, but the virtual interface has none. When I replace the virtual interface by one manually created with a proper IP configuration, it works. I can ping both ways, and instances get dhcp leases. This led me to think that the creation of virtual interface by nova-network must be faulty. The bridge creation is fine thought.

I have the 8021Q module enabled in the core, but I don't have the metadata api service installed on the compute node (can't install it because in debian it has the neutron-ovs-plugin as a dependency. Talk about logic). Do I need that for networking functions ? I can't find enough information on this service.

I really don't know what could be the root of this problem, but it is ruining the automation.
Maybe some debian specific issues ?

I would really like some help here.

Regards.

edit retag flag offensive close merge delete

1 answer

Sort by ยป oldest newest most voted
0

answered 2014-07-17 11:32:33 -0500

Alright, I have been hacking around, and my conclusion so far is the following :

"On my system (Debian Wheezy x64), virtual interfaces created with ip link add doesn't work, whereas the ones created with vconfig does".

I have no explainations for this. At first it lead me to think that there was something wrong with the ones generated by nova-network, because I could ping my instance when I created a proper one my self.
But it occured that even manual virtual interfaces created with ip link add aren't working at all. I didn't found any bugs related to this on the Debian side. To be continued I guess ....

Anyway, if you happen to be in the same situation, the solution is the following :

  1. Let nova-network create the bridge and virtual interfaces on first instance boot.
  2. Delete the virtual interface created by nova-network (named something like vlanID)
  3. Create a new one with vconfig add physical_interface vlanID (and no you don't need any IP configuration on it in order to ping your instances)
  4. Rename it as originally named by nova-network (you can use ip link set virt_int name vlanID)

That way you can kind off cheat nova-network and establish a proper communication between yours instances and your compute node.

Note that if your manually created interface has a different name, nova-network will fail booting with the following error : "Stderr: 'Cannot find device 'vlanID' ".

Hope this can help a fellow Debian stacker out.

edit flag offensive delete link more

Your Answer

Please start posting anonymously - your entry will be published after you log in or create a new account.

Add Answer

Get to know Ask OpenStack

Resources for moderators

Question Tools

1 follower

Stats

Asked: 2014-07-11 10:05:15 -0500

Seen: 377 times

Last updated: Jul 31 '14