1. backup seems dead here but it doesn't exit the JVM process. Why not we simply exit it?

Well with replication we do stop the JVM (check master), this is because we move files around and we want the user to manually intervene with the cleaning of the journals before restarting. with shared store the backup once more becomes a backup server

2. Would it be better if the liver server is back, the current live reduces itself to backup again?

This is what happens with shared store