Update: MariaDB is now on github. I have created my own fork, and pushed my changes on to the selfTuningOptimizer branch.

In the previous blog post, I had stated that I will start coding the things discussed in it. I have set up a branch on launchpad and have pushed a commit which I will be discussing in this post.

I first needed to make a table in the mysql database, from where we would be reading constants and later updating them too. The sql commands for creating a table in the mysql database are written in the file scripts/mysql_system_tables.sql. I have added a table named all_constants which currently has only two columns const_name and const_value.

-- Tables for Self Tuning Cost Optimizer

CREATE TABLE IF NOT EXISTS all_constants (const_name varchar(64) NOT NULL, const_value double NOT NULL, PRIMARY KEY (const_name) ) ENGINE=MyISAM CHARACTER SET utf8 COLLATE utf8_bin comment='Constants for optimizer';

-- Remember for later if all_constants table already existed
set @had_all_constants_table= @@warning_count != 0;

I also added the constants READ_TIME_FACTOR and SCAN_TIME_FACTOR in the all_constants table. Here is the corresponding code from scripts/mysql_system_tables_data.sql.

CREATE TEMPORARY TABLE tmp_all_constants LIKE all_constants;
INSERT INTO tmp_all_constants VALUES ('READ_TIME_FACTOR', 1.0);
INSERT INTO tmp_all_constants VALUES ('SCAN_TIME_FACTOR', 1.0);
INSERT INTO all_constants SELECT * FROM tmp_all_constants WHERE @had_all_constants_table=0;
DROP TABLE tmp_all_constants;

Next I needed to read these constants from the all_constants table. The file sql/statistics.cc has functions to update persistent statistical tables and to read from them. I have created two files sql/opt_costmodel.h and sql/opt_costmodel.cc which will have functions to read and update all_constants table. I have created All_constants class. Its methods allow us to read constants from the all_constants table. I have still not written the code for updating the table. Let us have a look at the function read_constant_from_table.

static double read_constant_from_table(THD *thd, const LEX_STRING const_name)      
  TABLE_LIST table_list;
  Open_tables_backup open_tables_backup;                                           
  if (open_table(thd, &table_list, &open_tables_backup, FALSE))                    
    return 1.0;                                                                    
  All_constants all_constants(table_list.table, const_name);                       
  close_system_tables(thd, &open_tables_backup);                                   
  return all_constants.get_const_value();                                          

open_table is a helper function which opens the all_constants table for reading/writing. set_key_fields sets the value of const_name_field, which is the primary key and read_const_value searches for the constant having that name in all_constants table and fills that value in const_value variable.

void set_key_fields()
  const_name_field->store(const_name.str, const_name.length, system_charset_info);
bool find_const()
  uchar key[MAX_KEY_LENGTH];
  key_copy(key, record[0], key_info, key_length);
  return !all_constants_file->ha_index_read_idx_map(record[0], key_idx, key,
                                                    HA_WHOLE_KEY, HA_READ_KEY_EXACT);
void read_const_value()
  if (find_const())
    Field *const_field= all_constants_table->field[ALL_CONSTANTS_CONST_VALUE];
      const_value= const_field->val_real();

So to read READ_TIME_FACTOR from the all_constants table, a call to read_constant_from_table is made with const_name set as READ_TIME_FACTOR. Similar is the case for SCAN_TIME_FACTOR.

Next big thing was how to test the code which I had just written. ./mtr was showing a lot of failed tests because I had added an extra table in the system database. So I decided to write a simple test of mine own in mysql-test/t/costmodel.test.


This test just creates a table with one field and runs a select query. To test my code, I added calls to get_read_time_factor and get_scan_time_factor in sql/sql_select.cc. Then I ran ./mtr --gdb costmodel to run the test-case with gdb. Using gdb I set up breakpoints on these functions and confirmed that the correct values were indeed being read.

Now that we have functions which will read the constants from the database, I need to figure out where all I need to call this functions. I made a list of places where handler::read_time() and handler::scan_time() were being called. The problem is that get_read_time_factor and get_scan_time_factor requires a pointer to current thread as an argument. I will tackle this problem in the coming week. Also I need an example query to add to my test case, which will take different execution plans on changing the values of READ_TIME_FACTOR and SCAN_TIME_FACTOR.

Suggestions/review on the code are welcome. Again, this is the link to the commit. I will be pushing to that branch regularly from now on.

Anshu Avinash

blog comments powered by Disqus