The hadoop-glusterfs shim has some experimental "quick slave i/o" code.  the code was meant to read directly from the raw local brick if it existed on the system.

this is problematic as there is no guarantee to the consistency of the raw local file.  reading and writing to the file through the supported FUSE I/O path guarantees file consistency across the cluster.

in other words, someone may be still writing to the file and the change is coming over the network.  by going directly to the local file system we bypass any checks or blocking that FUSE would do that would prevent the file from being read until it was fully consistent. 

bug: remove the quick slave I/O code from the shim layer.