7.1.2 The Optimizer Code

This diagram shows the structure of the function handle_select() in /sql/sql_select.cc (the server code that handles a query):

handle_select()
   mysql_select()
     JOIN::prepare()
       setup_fields()
     JOIN::optimize()            /* optimizer is from here ... */
       optimize_cond()
       opt_sum_query()
       make_join_statistics()
         get_quick_record_count()
         choose_plan()
           /* Find the best way to access tables */
           /* as specified by the user.          */
           optimize_straight_join()
             best_access_path()
           /* Find a (sub-)optimal plan among all or subset */
           /* of all possible query plans where the user    */
           /* controls the exhaustiveness of the search.   */
           greedy_search()
             best_extension_by_limited_search()
               best_access_path()
           /* Perform an exhaustive search for an optimal plan */
           find_best()
       make_join_select()        /* ... to here */
     JOIN::exec()

The indentation in the diagram shows what calls what. Thus you can see that handle_select() calls mysql_select() which calls JOIN::prepare() which calls setup_fields(), and so on. The first part of mysql_select() is JOIN::prepare() which is for context analysis, metadata setup, and some subquery transformations. The optimizer is JOIN::optimize() and all its subordinate routines. When the optimizer finishes, JOIN::exec() takes over and does the job that JOIN::optimize() decides upon.

Although the word “JOIN” appears, these optimizer routines are applicable to all query types.

The optimize_cond() and opt_sum_query() routines perform transformations. The make_join_statistics() routine puts together all the information it can find about indexes that might be useful for accessing the query's tables.