Comment 48 for bug 1023755

I will be back working on this (though realistically this will not be before next week). From the various dumps I have been looking at until now it looks like the problem is somewhere in the way the storage stack is set up and possibly a slightly higher memory footprint.
Basically cpus are in schedule waiting for some io to complete and I wonder whether everything on the io path has at least some emergency buffers in place for cases where io has to be kept running to make room for more io.

John, would you be able to experiment a bit with the setup in order to narrow this a bit down? What I would like to check would be a setup where the volume group resides on a real partition or disk (so excluding the loop device and the filesystem io).
At some point it will also be good to re-check against a recent mainline kernel. I will be able to do that, I just would likely need more time to figure out how to re-configure devstack to use a vg based on a raw block device.