Uploaded image for project: 'MariaDB Server'
  1. MariaDB Server
  2. MDEV-410

EXPLAIN shows type=range, while SHOW EXPLAIN and userstat show full table scan is used

Details

    • Bug
    • Status: Closed (View Workflow)
    • Major
    • Resolution: Fixed
    • None
    • 10.0.0
    • None
    • None

    Description

      For the following query

      SELECT MIN(b) FROM ( SELECT * FROM t1, t2, t3 WHERE d = b ) AS alias1 
      WHERE SLEEP(0.2) OR c < 'p' OR b = ( SELECT MIN(b) FROM t2 )

      usual EXPLAIN produces

      id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
      1	PRIMARY	t1	system	NULL	NULL	NULL	NULL	1	
      1	PRIMARY	t2	range	b	b	5	NULL	2	Using index condition; Using where
      1	PRIMARY	t3	ref	d	d	5	test.t2.b	2	Using where; Using index
      3	SUBQUERY	NULL	NULL	NULL	NULL	NULL	NULL	NULL	Select tables optimized away

      while SHOW EXPLAIN says

      id	select_type	table	type	possible_keys	key	key_len	ref	rows	Extra
      1	PRIMARY	t1	system	NULL	NULL	NULL	NULL	1	
      1	PRIMARY	t2	ALL	b	NULL	NULL	NULL	2	Using where
      1	PRIMARY	t3	ref	d	d	5	test.t2.b	2	Using where; Using index
      3	SUBQUERY	NULL	NULL	NULL	NULL	NULL	NULL	NULL	Select tables optimized away

      As always, SLEEP in the query is not essential, it just makes the query a bit longer and allows to catch it by SHOW EXPLAIN.

      bzr version-info (5.5-show-explain-test1)

      revision-id: psergey@askmonty.org-20120719215203-m2p9cbqb37n0th7n
      date: 2012-07-20 01:52:03 +0400
      build-date: 2012-07-24 04:43:23 +0400
      revno: 3456

      Test case:

       
      CREATE TABLE t1 (a VARCHAR(3) PRIMARY KEY) ENGINE=MyISAM;
      INSERT INTO t1 VALUES ('USA');
       
      CREATE TABLE t2 (b INT, c VARCHAR(52), KEY(b)) ENGINE=MyISAM;
      INSERT INTO t2 VALUES (3813,'United States'),(3940,'Russia');
       
      CREATE TABLE t3 (d INT, KEY(d)) ENGINE=MyISAM;
      INSERT INTO t3 VALUES (12),(22),(9),(45);
       
      EXPLAIN
      SELECT MIN(b) FROM ( SELECT * FROM t1, t2, t3 WHERE d = b ) AS alias1 
      WHERE SLEEP(0.2) OR c < 'p' OR b = ( SELECT MIN(b) FROM t2 );
      --echo #---------------
      --echo # SHOW EXPLAIN output:
       
      --connect (con1,localhost,root,,)
      --let $con_id = `SELECT CONNECTION_ID()`
       
      --let $trials = 50
       
      --disable_query_log
       
      while ($trials)
      {
        --dec $trials
        --let $run = 1000
       
        send
        SELECT MIN(b) FROM ( SELECT * FROM t1, t2, t3 WHERE d = b ) AS alias1 
        WHERE SLEEP(0.2) OR c < 'p' OR b = ( SELECT MIN(b) FROM t2 );
       
       
        --connection default
        while ($run)
        {
          --error 0,1932
          eval SHOW EXPLAIN FOR $con_id;
          --dec $run
          if (!$mysql_errno)
          {
            --let $run = 0
            --let $trials = 0
            --let $found = 1
          }
        }
       
        --disable_result_log
        --connection con1
        --reap
        --enable_result_log
       
      }
       
      if (!$found)
      {
        --echo ########### Could not catch the query by SHOW EXPLAIN, try again  #############
      }
       
      DROP TABLE t1, t2, t3;

      Attachments

        Issue Links

          Activity

            It seems, SHOW EXPLAIN's output is correct, and EXPLAIN is wrong.

            Let's create a copy of t2 so that accesses to t2-outside-the-subquery and t2-inside-the-subquery are counted separately:

            create table t2a like t2;
            insert into t2a select * from t2;

            Then I run:
            SELECT MIN(b) FROM ( SELECT * FROM t1, t2, t3 WHERE d = b ) AS alias1 WHERE SLEEP(0.2) OR c < 'p' OR b = ( SELECT MIN(b) FROM t2a );

            and then I see:

            MariaDB [j12]> show index_statistics;
            -------------------------------------------+

            Table_schema Table_name Index_name Rows_read

            -------------------------------------------+

            j12 t2a b 1

            -------------------------------------------+
            1 row in set (0.01 sec)

            MariaDB [j12]> show table_statistics;
            --------------------------------------------------------------------

            Table_schema Table_name Rows_read Rows_changed Rows_changed_x_#indexes

            --------------------------------------------------------------------

            j12 t2a 1 0 0
            j12 t2 2 0 0
            j12 t1 2 0 0

            --------------------------------------------------------------------
            3 rows in set (0.01 sec)

            As one can see, there were 2 rows read from table t2, and they were not read through an index.

            psergei Sergei Petrunia added a comment - It seems, SHOW EXPLAIN's output is correct, and EXPLAIN is wrong. Let's create a copy of t2 so that accesses to t2-outside-the-subquery and t2-inside-the-subquery are counted separately: create table t2a like t2; insert into t2a select * from t2; Then I run: SELECT MIN(b) FROM ( SELECT * FROM t1, t2, t3 WHERE d = b ) AS alias1 WHERE SLEEP(0.2) OR c < 'p' OR b = ( SELECT MIN(b) FROM t2a ); and then I see: MariaDB [j12] > show index_statistics; ------------- ---------- ---------- ----------+ Table_schema Table_name Index_name Rows_read ------------- ---------- ---------- ----------+ j12 t2a b 1 ------------- ---------- ---------- ----------+ 1 row in set (0.01 sec) MariaDB [j12] > show table_statistics; ------------- ---------- --------- ------------ ------------------------ Table_schema Table_name Rows_read Rows_changed Rows_changed_x_#indexes ------------- ---------- --------- ------------ ------------------------ j12 t2a 1 0 0 j12 t2 2 0 0 j12 t1 2 0 0 ------------- ---------- --------- ------------ ------------------------ 3 rows in set (0.01 sec) As one can see, there were 2 rows read from table t2, and they were not read through an index.

            Debugged SELECT and its EXPLAIN concurrently. The difference is in make_join_statistics(), here:

            /*
            We plan to scan all rows.
            Check again if we should use an index.
            We could have used an column from a previous table in
            the index if we are using limit and this is the first table
            */

            if (!tab->table->is_filled_at_execution() &&
            ((cond && (!tab->keys.is_subset(tab->const_keys) && i > 0)) ||
            (!tab->const_keys.is_clear_all() && i == join->const_tables &&
            join->unit->select_limit_cnt <
            join->best_positions[i].records_read &&
            !(join->select_options & OPTION_FOUND_ROWS))))
            {

            psergei Sergei Petrunia added a comment - Debugged SELECT and its EXPLAIN concurrently. The difference is in make_join_statistics(), here: /* We plan to scan all rows. Check again if we should use an index. We could have used an column from a previous table in the index if we are using limit and this is the first table */ if (!tab->table->is_filled_at_execution() && ((cond && (!tab->keys.is_subset(tab->const_keys) && i > 0)) || (!tab->const_keys.is_clear_all() && i == join->const_tables && join->unit->select_limit_cnt < join->best_positions [i] .records_read && !(join->select_options & OPTION_FOUND_ROWS)))) {

            The difference comes from different value of tab->const_keys(), which comes from update_ref_and_keys() , from add_key_fields() call for this part of the WHERE condtion:

            (`j14`.`t2`.`b` = (select min(`j14`.`t2`.`b`) from `j14`.`t2`))

            Here, the right part is an Item_singlerow_subselect, and it has item->const_item() == FALSE for the EXPLAIN and TRUE for the SELECT.
            the difference comes from different value of their Item::const_item_cache

            psergei Sergei Petrunia added a comment - The difference comes from different value of tab->const_keys(), which comes from update_ref_and_keys() , from add_key_fields() call for this part of the WHERE condtion: (`j14`.`t2`.`b` = (select min(`j14`.`t2`.`b`) from `j14`.`t2`)) Here, the right part is an Item_singlerow_subselect, and it has item->const_item() == FALSE for the EXPLAIN and TRUE for the SELECT. the difference comes from different value of their Item::const_item_cache

            Verified that the problem with EXPLAIN being inconsistent with userstat profiling exists on the current mariadb-5.5 ( sanja@montyprogram.com-20120724145006-17q5hre8fwp1toql )

            psergei Sergei Petrunia added a comment - Verified that the problem with EXPLAIN being inconsistent with userstat profiling exists on the current mariadb-5.5 ( sanja@montyprogram.com-20120724145006-17q5hre8fwp1toql )

            Figured out this is a problem in EXPLAIN, not in SHOW EXPLAIN.
            Debugged, fixed, pushed into 5.5
            Merged into the 5.5-show-explain-test1 tree

            psergei Sergei Petrunia added a comment - Figured out this is a problem in EXPLAIN, not in SHOW EXPLAIN. Debugged, fixed, pushed into 5.5 Merged into the 5.5-show-explain-test1 tree

            People

              psergei Sergei Petrunia
              elenst Elena Stepanova
              Votes:
              0 Vote for this issue
              Watchers:
              4 Start watching this issue

              Dates

                Created:
                Updated:
                Resolved:

                Git Integration

                  Error rendering 'com.xiplink.jira.git.jira_git_plugin:git-issue-webpanel'. Please contact your Jira administrators.