Uploaded image for project: 'MariaDB Server'
  1. MariaDB Server
  2. MDEV-7316

Assertion `join->best_read < double(...) failed in bool greedy_search with optimizer_use_condition_selectivity>2, InnoDB

Details

    Description

      The problem appeared on 10.0 tree with the following revision:

      revno: 4465 [merge]
      		 
      revision-id: igor@askmonty.org-20141028233126-zt6nb214qnyqfuy8
      		 
      parent: jplindst@mariadb.org-20141027145816-cuwsbq6d5yber97u
      		 
      parent: igor@askmonty.org-20141028213331-qrdbjknh6i7nzsuf
      		 
      committer: Igor Babaev <igor@askmonty.org>
      		 
      branch nick: maria-10.0-trunk
      		 
      timestamp: Tue 2014-10-28 16:31:26 -0700
      		 
      message:
      		 
        Merge
      		 
          ------------------------------------------------------------
      		 
          revno: 4446.1.1
      		 
          revision-id: igor@askmonty.org-20141028213331-qrdbjknh6i7nzsuf
      		 
          parent: psergey@askmonty.org-20141016185808-n8cw47wxrn3a9yxm
      		 
          committer: Igor Babaev <igor@askmonty.org>
      		 
          branch nick: maria-10.0
      		 
          timestamp: Tue 2014-10-28 14:33:31 -0700
      		 
          message:
      		 
            Fixed bug mdev-6325.
      		 
            
            Field::selectivity should be set for all fields used in range conditions.
      		 

      --source include/have_innodb.inc
      		 
       
      		 
      CREATE TABLE t1 (a VARCHAR(16), b INT, c INT, PRIMARY KEY(a), KEY(c)) ENGINE=InnoDB;
      		 
      INSERT INTO t1 VALUES ('USAChinese',1,10),('USAEnglish',2,20),('USAFrench',3,30);
      		 
       
      		 
      CREATE TABLE t2 (i int) ENGINE=InnoDB;
      		 
      INSERT INTO t2 VALUES (1),(2),(3),(4),(5),(6),(7),(8),(9),(10),(1),(2),(3),(4);
      		 
       
      		 
      SET use_stat_tables=PREFERABLY, optimizer_use_condition_selectivity=3;
      		 
      ANALYZE TABLE t1, t2;
      		 
       
      		 
      SELECT * FROM t1, t2 WHERE b IN ( SELECT 2 UNION SELECT 3 ) AND a <> 'USARussian' AND c IS NULL;
      		 

      10.0/sql/sql_select.cc:6950: bool greedy_search(JOIN*, table_map, uint, uint, uint): Assertion `join->best_read < double(1.79769313486231570815e+308L)' failed.
      		 
      141213 22:37:26 [ERROR] mysqld got signal 6 ;

      Stack trace from 10.0 revno 4543

      		 
      #6  0x00007fbaa3e056f1 in *__GI___assert_fail (assertion=0xf11688 "join->best_read < double(1.79769313486231570815e+308L)", file=<optimized out>, line=6950, function=0xf14680 "bool greedy_search(JOIN*, table_map, uint, uint, uint)") at assert.c:81
      		 
      #7  0x00000000006c17e5 in greedy_search (join=0x7fba91229bd0, remaining_tables=3, search_depth=62, prune_level=1, use_cond_selectivity=3) at 10.0/sql/sql_select.cc:6950
      		 
      #8  0x00000000006c0d6f in choose_plan (join=0x7fba91229bd0, join_tables=3) at 10.0/sql/sql_select.cc:6519
      		 
      #9  0x00000000006ba73d in make_join_statistics (join=0x7fba91229bd0, tables_list=..., conds=0x7fba912297f0, keyuse_array=0x7fba91229ed8) at 10.0/sql/sql_select.cc:4038
      		 
      #10 0x00000000006b13a3 in JOIN::optimize_inner (this=0x7fba91229bd0) at 10.0/sql/sql_select.cc:1339
      		 
      #11 0x00000000006b033a in JOIN::optimize (this=0x7fba91229bd0) at 10.0/sql/sql_select.cc:1024
      		 
      #12 0x00000000006b7ebf in mysql_select (thd=0x7fba9d9ee070, rref_pointer_array=0x7fba9d9f26e0, tables=0x7fba91341300, wild_num=1, fields=..., conds=0x7fba912297f0, og_num=0, order=0x0, group=0x0, having=0x0, proc_param=0x0, select_options=2147748608, result=0x7fba91229bb0, unit=0x7fba9d9f1d80, select_lex=0x7fba9d9f2468) at 10.0/sql/sql_select.cc:3294
      		 
      #13 0x00000000006ae549 in handle_select (thd=0x7fba9d9ee070, lex=0x7fba9d9f1cb8, result=0x7fba91229bb0, setup_tables_done_option=0) at 10.0/sql/sql_select.cc:373
      		 
      #14 0x0000000000683244 in execute_sqlcom_select (thd=0x7fba9d9ee070, all_tables=0x7fba91341300) at 10.0/sql/sql_parse.cc:5261
      		 
      #15 0x000000000067b542 in mysql_execute_command (thd=0x7fba9d9ee070) at 10.0/sql/sql_parse.cc:2545
      		 
      #16 0x00000000006859bd in mysql_parse (thd=0x7fba9d9ee070, rawbuf=0x7fba91341088 "SELECT * FROM t1, t2 WHERE b IN ( SELECT 2 UNION SELECT 3 ) AND a <> 'USARussian' AND c IS NULL", length=95, parser_state=0x7fbaa5fc5610) at 10.0/sql/sql_parse.cc:6407
      		 
      #17 0x00000000006787b2 in dispatch_command (command=COM_QUERY, thd=0x7fba9d9ee070, packet=0x7fba986b9071 "SELECT * FROM t1, t2 WHERE b IN ( SELECT 2 UNION SELECT 3 ) AND a <> 'USARussian' AND c IS NULL", packet_length=95) at 10.0/sql/sql_parse.cc:1299
      		 
      #18 0x0000000000677b57 in do_command (thd=0x7fba9d9ee070) at 10.0/sql/sql_parse.cc:996
      		 
      #19 0x0000000000794877 in do_handle_one_connection (thd_arg=0x7fba9d9ee070) at 10.0/sql/sql_connect.cc:1375
      		 
      #20 0x00000000007945ca in handle_one_connection (arg=0x7fba9d9ee070) at 10.0/sql/sql_connect.cc:1289
      		 
      #21 0x0000000000ccda5e in pfs_spawn_thread (arg=0x7fba986b10f0) at 10.0/storage/perfschema/pfs.cc:1860
      		 
      #22 0x00007fbaa5bffb50 in start_thread (arg=<optimized out>) at pthread_create.c:304
      		 
      #23 0x00007fbaa3eb620d in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:112

      Attachments

        Issue Links

          relates to

          Bug - A problem which impairs or prevents the functions of the product. MDEV-7413 optimizer_use_condition_selectivity > 2 crashes 10.0.15+maria-1~wheezy

          • Major - Major loss of function.
          • Closed

          Activity

            Ascending order - Click to sort in descending order
            psergei Sergei Petrunia added a comment -

            The crash doesn't repeat if one re-runs the query. However, running the whole example reliably produces the crash.

            The crash is caused by the first call to table_cond_selectivity() returning inf.

            psergei Sergei Petrunia added a comment - The crash doesn't repeat if one re-runs the query. However, running the whole example reliably produces the crash. The crash is caused by the first call to table_cond_selectivity() returning inf.
            psergei Sergei Petrunia added a comment -

            The first call for table_cond_selectivity(s=t1, idx=0).

            Selectivity becomes infinite here:

                          sel /= table->field[fldno]->cond_selectivity;

            The field in question is t1.c.

            psergei Sergei Petrunia added a comment - The first call for table_cond_selectivity(s=t1, idx=0). Selectivity becomes infinite here: sel /= table->field[fldno]->cond_selectivity; The field in question is t1.c.
            psergei Sergei Petrunia added a comment -

            The WHERE clause has:

            c IS NULL

            This equality is used to construct ref access. Then, table_cond_selectivity attempts to discount the selectivity of "c IS NULL". This is where it divides by zero.

            psergei Sergei Petrunia added a comment - The WHERE clause has: c IS NULL This equality is used to construct ref access. Then, table_cond_selectivity attempts to discount the selectivity of "c IS NULL". This is where it divides by zero.
            psergei Sergei Petrunia added a comment -

            The data in table t1:

            INSERT INTO t1 VALUES 
              ('USAChinese',1,10),
            		 
              ('USAEnglish',2,20),
            		 
              ('USAFrench', 3,30);

            There are 3 rows in total, 0 rows that have t1.c=NULL.

            get_column_range_cardinality(field="c") executes this code:

                if (nulls_incl)
            		 
                {
            		 
                  /* This is null single point range */
            		 
                  res= col_nulls;

            col_nulls=0, which is true. But then, we attempt to divide by zero to discount the selectivity.

            psergei Sergei Petrunia added a comment - The data in table t1: INSERT INTO t1 VALUES ('USAChinese',1,10), ('USAEnglish',2,20), ('USAFrench', 3,30); There are 3 rows in total, 0 rows that have t1.c=NULL. get_column_range_cardinality(field="c") executes this code: if (nulls_incl) { /* This is null single point range */ res= col_nulls; col_nulls=0, which is true. But then, we attempt to divide by zero to discount the selectivity.
            psergei Sergei Petrunia added a comment -

            A possible fix: http://lists.askmonty.org/pipermail/commits/2015-February/007456.html . igor, please review.

            psergei Sergei Petrunia added a comment - A possible fix: http://lists.askmonty.org/pipermail/commits/2015-February/007456.html . igor , please review.
            psergei Sergei Petrunia added a comment -

            A different fix for this bug has actually been pushed into the 10.0 bzr tree: 

                revno: 4544.1.1
            		 
                revision-id: igor@askmonty.org-20150206040908-23e1eht6sj59qiz3
            		 
                parent: knielsen@knielsen-hq.org-20141218105908-7n3szfd3xb5bm8gs
            		 
                committer: Igor Babaev <igor@askmonty.org>
            		 
                branch nick: maria-10.0
            		 
                timestamp: Thu 2015-02-05 20:09:08 -0800
            		 
                message:
            		 
                  Fixed bug mdev-7316.
            		 
                  The function table_cond_selectivity() should take into account that condition selectivity
            		 
                  for some fields can be set to 0.

            this fix hasn't been merged into the git tree, yet.

            psergei Sergei Petrunia added a comment - A different fix for this bug has actually been pushed into the 10.0 bzr tree: revno: 4544.1.1 revision-id: igor@askmonty.org-20150206040908-23e1eht6sj59qiz3 parent: knielsen@knielsen-hq.org-20141218105908-7n3szfd3xb5bm8gs committer: Igor Babaev <igor@askmonty.org> branch nick: maria-10.0 timestamp: Thu 2015-02-05 20:09:08 -0800 message: Fixed bug mdev-7316. The function table_cond_selectivity() should take into account that condition selectivity for some fields can be set to 0. this fix hasn't been merged into the git tree, yet.
            psergei Sergei Petrunia added a comment -

            Actually it has been merged already. After 10.0.16

            psergei Sergei Petrunia added a comment - Actually it has been merged already. After 10.0.16

            People

              igor Igor Babaev (Inactive)
              elenst Elena Stepanova
              Votes:
              0 Vote for this issue
              Watchers:
              3 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.