The replication tables had the Master_Id column as the primary key. 
Despite the suggestive name, it was not supposed to store a server_id
which is  used to uniquely identify a server, including the master.
The column was just an internal identifier. 
             
Unfortunately, a bug made the slave's server_id be stored instead of a 
const number such as "zero" thus causing confusion and leading to this
bug report. 
             
To fix the problem, we have done what follows: 
. Made master_host and master_port slave_master_info's primary key. 
. Renamed slave_relay_log_info's Master_Id to Id. 
             
We also had taken the chance to improve the code as follows: 
. Removed duplicated code from Rpl_info_factory. 
. Refactored Rpl_info_handler's interface to avoid passing the 
  primary key as parameter in all the functions. 
. Refactored the code to allow one to redefine the primary keys 
  without changing the code.