Salt keeps creating a half-complete staging instance on control, which stops staging from upgrading

Tyler and I found a folder on control in /srv/lava/instances/staging which was blocking the cron job which upgrades staging to HEAD.

Traceback (most recent call last):
  File "/srv/salt/scripts/lava-upgrade", line 22, in <module>
    m_ret, w_ret = lava_salt.upgrade(client, args.instance, args.dryrun)
  File "/srv/salt/scripts/lava_salt.py", line 93, in upgrade
    assert not master, 'Detected multiple master instances in LAVA deployment'
AssertionError: Detected multiple master instances in LAVA deployment
Wed Aug 21 10:00:12 BST 2013

We removed that folder and the upgrade worked fine.

I've just checked today and found:

matthew-hart@control:~$ find /srv/lava/instances/staging/
/srv/lava/instances/staging/
/srv/lava/instances/staging/etc
/srv/lava/instances/staging/etc/lava-dispatcher
/srv/lava/instances/staging/etc/lava-dispatcher/device-types
/srv/lava/instances/staging/etc/lava-dispatcher/device-types/vexpress-tc2-benchmark.conf
/srv/lava/instances/staging/etc/lava-dispatcher/device-types/rtsm_ve-a15x1-a7x1.conf
/srv/lava/instances/staging/etc/lava-dispatcher/device-types/arndale-uefi.conf
/srv/lava/instances/staging/etc/lava-dispatcher/device-types/vexpress-a9.conf
/srv/lava/instances/staging/etc/lava-dispatcher/device-types/vexpress-a5.conf
/srv/lava/instances/staging/etc/lava-dispatcher/device-types/rtsm_ve-a15x4-a7x4.conf
/srv/lava/instances/staging/etc/lava-dispatcher/device-types/.lava
/srv/lava/instances/staging/etc/lava-dispatcher/urlmappings.txt
/srv/lava/instances/staging/etc/lava-server
/srv/lava/instances/staging/etc/lava-server/settings.conf

So it looks like control is getting this folder populated by salt somehow.