Provisioning failed if inactive md device had been found

It's hard to reproduce. But the possibility is a bit higher if cluster was re-deployed

Sometimes fuel-agent fails with

2014-11-07 12:45:20.795 4899 CRITICAL fuel-agent [-] KeyError: 'devices'
2014-11-07 12:45:20.795 4899 TRACE fuel-agent Traceback (most recent call last):
2014-11-07 12:45:20.795 4899 TRACE fuel-agent File "/usr/bin/provision", line 10, in <module>
2014-11-07 12:45:20.795 4899 TRACE fuel-agent sys.exit(provision())
2014-11-07 12:45:20.795 4899 TRACE fuel-agent File "/usr/lib/python2.6/site-packages/fuel_agent/cmd/agent.py", line 37, in provision
2014-11-07 12:45:20.795 4899 TRACE fuel-agent main(['do_provisioning'])
2014-11-07 12:45:20.795 4899 TRACE fuel-agent File "/usr/lib/python2.6/site-packages/fuel_agent/cmd/agent.py", line 67, in main
2014-11-07 12:45:20.795 4899 TRACE fuel-agent getattr(mgr, action)()
2014-11-07 12:45:20.795 4899 TRACE fuel-agent File "/usr/lib/python2.6/site-packages/fuel_agent/manager.py", line 296, in do_provisioning
2014-11-07 12:45:20.795 4899 TRACE fuel-agent self.do_partitioning()
2014-11-07 12:45:20.795 4899 TRACE fuel-agent File "/usr/lib/python2.6/site-packages/fuel_agent/manager.py", line 123, in do_partitioning
2014-11-07 12:45:20.795 4899 TRACE fuel-agent mu.mdcreate(md.name, md.level, *md.devices)
2014-11-07 12:45:20.795 4899 TRACE fuel-agent File "/usr/lib/python2.6/site-packages/fuel_agent/utils/md_utils.py", line 97, in mdcreate
2014-11-07 12:45:20.795 4899 TRACE fuel-agent set(reduce(lambda x, y: x + y, [md['devices'] for md in mds], [])):
2014-11-07 12:45:20.795 4899 TRACE fuel-agent KeyError: 'devices'
2014-11-07 12:45:20.795 4899 TRACE fuel-agent

The actual reason is that '/proc/mdstat' contains information about inactive md devices too.
Eg.:
Personalities :
md127 : inactive vdc3[2](S)
      204736 blocks

unused devices: <none>

`mdadm --details` can't get any information about inactive device
mdadm: md device /dev/md127 does not appear to be active.