Oversubscription of storage may not be desirable. Having users create 6 20TB images on a 60TB brick is a disaster waiting to happen. Yes, sharding may help mitigate that, but it would be nice to have allocation guarantees. It seems to me that if you added an option to the posix translator where any ftruncate would immediately be followed with an fallocate for that same size, that guarantee could be met.
Migrated to github: https://github.com/gluster/glusterfs/issues/577 Please follow the github issue for further updates on this bug.