Description of problem: If M has a geo-rep replica S, and M becomes inaccessible/unreliable/corrupt, it should be possible to restore it to an operational state from S in a controlled way (failback), while, at the same time, downtime should be minimized by migrating applications over to S (failover).
I guess this feature is already present in codebase. Can you please add 'Doc Text' to this bug and close it?
Feature is implemented by http://review.gluster.com/3541. Feature also relies on checkpointing, cf. https://bugzilla.redhat.com/show_bug.cgi?id=826512