When using shared storage across hypervisors, libvirt driver resize/migrate operations can result in a loss of instance data. This is happening because many of the operations to create a copy of the instance are done within a try/except block. Thus, if any operations fail, you're into the exception which does the following:
=== code ===
except Exception:
with excutils.save_and_reraise_exception():
self._cleanup_remote_migration(dest, inst_base,
inst_base_resize)
def _cleanup_remote_migration(self, dest, inst_base, inst_base_resize):
"""Used only for cleanup in case migrate_disk_and_power_off fails."""
try:
if os.path.exists(inst_base_resize):
utils.execute('rm', '-rf', inst_base)
utils.execute('mv', inst_base_resize, inst_base)
utils.execute('ssh', dest, 'rm', '-rf', inst_base)
except Exception:
pass
=== end ===
It doesn't take looking at this code for long to see why this is going to be a problem with shared storage. In effect, the last ssh operation in the block above is going to blow away the original copy of the instance directory.
The issue can be easily reproduced by issuing a resize of an instance with a large root disk. In the middle of the resize, kill the ssh process created from the following call (https://github.com/openstack/nova/blob/master/nova/virt/libvirt/driver.py#L3508) and observe the exception handler destroying everything.
I've got a patch ready to be submitted.