Proposal to deprecate COM_REFRESH packet

In the MySQL team we are proposing to deprecate the COM_REFRESH packet in favor of specific queries to execute FLUSH commands. To provide a bit of context:

• The MySQL server protocol allows for clients to speak API commands via both a query and binary protocol interface. The set of the API commands can be seen in the MySQL Client/Server Protocol internals documentation, or very simply as they appear in a single switch statement:

  # ./sql/sql_parse.cc:1009 (simplified view)
    switch (command) {
    case COM_REGISTER_SLAVE:
    {
      /* do stuff */
      break;
    }
    case COM_QUERY:
    {
      /* parse query, do stuff */
      break;
    }
    case COM_REFRESH:
    {
     /* equivalent to running a FLUSH command */
      break;
    }
    case COM_SHUTDOWN:
    {
      kill_mysql();
      break;
    }
  }
  

• The historical advantage of having a binary protocol meant that less bytes needed to be transfered across the network as well as less processing on the server, since parsing of statements is not required.

• The disadvantage of the binary protocol is that it is less extensible. In the case of COM_REFRESH it was designed to accept only a 1 byte payload to indicate which flush command to run. A quick look at the MySQL manual shows that there are now more than eight (1 byte payload = 8 possible flags) flush commands available. This leads to API inconsistency as some commands are available via the binary interface and others are not.

Our proposed solution to this problem is to deprecate the binary interface for refresh commands and promote usage of the query interface (FLUSH LOGS, FLUSH TABLES, FLUSH HOSTS etc) instead. While we have no immediate plans to remove the COM_REFRESH binary interface (and thus break older clients) we are seeking feedback on the direction of this change. Please leave a comment, or get in touch!