Kind of like
HBASE-4462 but for replication. If while replicating you get a socket timeout, the last thing you want to do is to retry it right away. Since we can't fail the replication thread, the best I can think of is to sleep a really long amount of time.
Planning to bring this to all branches.