CephCollector sending badly-formatted pool stat names
Crash signature (v1):
Crash signature (v2):
_publish_cluster_stats was passing the 'name' part of the tuple returned from flatten_dictionary directly to
publish_*, rather than first
_PATH_SEP.join() ing it. This resulted in network messages like
"ceph.cluster.['ceph', 'pool.0', u'num_objects_unfound'] 0 1388805994\n"
which were certainly not going to work.