At the point nova tries to start the LXC instance, there are no directories except for proc, dev and tmp on the rootfs, causing the LXC container to fail to start. If I manually set up an nbd device from the qcow image, mount it and start the instance using virsh, it boots correctly. This has worked in previous versions.
Based on past experience, when booting an LXC instance I would expect compute to get a root image, create a qcow image from it, create an nbd device from the qcow and mount the nbd as the instance rootfs, as seen in the below extract from the log files from a previous working nova version:
Running cmd (subprocess): qemu-img create -f qcow2 -o cluster_size=2M,backing_file=/var/lib/nova/instances/_base/da4b9237bacccdf19c0760cab7aec4a8359010b0 /var/lib/nova/instances/instance-00000003/disk
Running cmd (subprocess): sudo qemu-nbd -c /dev/nbd13 /var/lib/nova/instances/instance-00000003/disk
Running cmd (subprocess): sudo mount /dev/nbd13 /var/lib/nova/instances/instance-00000003//rootfs
instance instance-00000003: is running
In the version currently available in Ubuntu and the trunk ppa I see the creation of the qcow and immediately after that the boot of the instance:
Running cmd (subprocess): qemu-img create -f qcow2 -o cluster_size=2M,backing_file=/var/lib/nova/instances/_base/356a192b7913b04c54574d18c28d46e6395428ab /var/lib/nova/instances/instance-00000001/disk
instance instance-00000001: is running
Last known working version: 2011.3~d4~20110814.1433-0ubuntu0ppa1~oneiric1
Not Working versions: 2011.3~rc~20110909.r1155-0ubuntu0ppa1~oneiric1 and 2012.1~e1~20110913.1567-0ubuntu0ppa1~oneiric2
Do you try to start your instance without key or network file or metadata to inject ?
It looks as a duplicated bug: https:/