Thanks Sergey Shelukhin! Correct, no change in logic, just refactoring.
As for the AtomicBoolean, I originally thought the same and was going to remove it. The problem though is that on the Region server, it is also passed to the ReplicationSourceManager, and then in turn passed to replication sources (Check out Replication.java). I can see removing it in a future patch, and passing around ReplicationStateInterfaces instead, but I wanted to keep this first patch small. I can see doing a bunch of clean up like that once we have the three interfaces broken out.
I'll address your other comments, investigate some of the other failures and post another patch.