Uploaded image for project: 'MariaDB Server'
  1. MariaDB Server
  2. MDEV-5581

Server crashes in in JOIN::prepare on 2nd execution of PS with materialization+semijoin

        Details

        • Type: Bug
        • Status: Closed
        • Priority: Major
        • Resolution: Fixed
        • Affects Version/s: 5.5.35, 10.0.7, 5.3.13
        • Fix Version/s: 5.5.36, 10.0.9, 5.3.13
        • Component/s: None
        • Labels:
          None

          Description

          #3  <signal handler called>
#4  0x0000000000697d83 in JOIN::prepare (this=0x7feea6059998, rref_pointer_array=0x7feea61cf2f8, tables_init=0x7feea61cfc58, wild_num=0, conds_init=0x7feea6348d18, og_num=0, order_init=0x0, skip_order_by=false, group_init=0x0, having_init=0x0, proc_param_init=0x0, select_lex_arg=0x7feea61cf088, unit_arg=0x7feea61cf3e8) at /home/elenst/bzr/10.0/sql/sql_select.cc:768
#5  0x00000000008ecbd6 in subselect_single_select_engine::prepare (this=0x7feea63479b8) at /home/elenst/bzr/10.0/sql/item_subselect.cc:3483
#6  0x00000000008e38cc in Item_subselect::fix_fields (this=0x7feea6347828, thd_param=0x7feea8f3d070, ref=0x7feea6348180) at /home/elenst/bzr/10.0/sql/item_subselect.cc:257
#7  0x00000000008ebed2 in Item_in_subselect::fix_fields (this=0x7feea6347828, thd_arg=0x7feea8f3d070, ref=0x7feea6348180) at /home/elenst/bzr/10.0/sql/item_subselect.cc:3166
#8  0x000000000087fea1 in Item_in_optimizer::fix_fields (this=0x7feea63480e8, thd=0x7feea8f3d070, ref=0x7feeb0922700) at /home/elenst/bzr/10.0/sql/item_cmpfunc.cc:1515
#9  0x0000000000610738 in setup_tables (thd=0x7feea8f3d070, context=0x7feea61cc880, from_clause=0x7feea61cc9a0, tables=0x7feea61cd6d0, leaves=..., select_insert=false, full_table_list=false) at /home/elenst/bzr/10.0/sql/sql_base.cc:7822
#10 0x00000000006108a1 in setup_tables_and_check_access (thd=0x7feea8f3d070, context=0x7feea61cc880, from_clause=0x7feea61cc9a0, tables=0x7feea61cd6d0, leaves=..., select_insert=false, want_access_first=1, want_access=1, full_table_list=false) at /home/elenst/bzr/10.0/sql/sql_base.cc:7876
#11 0x0000000000697b89 in JOIN::prepare (this=0x7feea60592e0, rref_pointer_array=0x7feea61ccaa0, tables_init=0x7feea61cd6d0, wild_num=0, conds_init=0x7feea6348de0, og_num=0, order_init=0x0, skip_order_by=false, group_init=0x0, having_init=0x0, proc_param_init=0x0, select_lex_arg=0x7feea61cc830, unit_arg=0x7feea61cc150) at /home/elenst/bzr/10.0/sql/sql_select.cc:703
#12 0x00000000006a0ac5 in mysql_select (thd=0x7feea8f3d070, rref_pointer_array=0x7feea61ccaa0, tables=0x7feea61cd6d0, wild_num=0, fields=..., conds=0x7feea6348de0, og_num=0, order=0x0, group=0x0, having=0x0, proc_param=0x0, select_options=2416184064, result=0x7feea63479f8, unit=0x7feea61cc150, select_lex=0x7feea61cc830) at /home/elenst/bzr/10.0/sql/sql_select.cc:3272
#13 0x00000000006972a7 in handle_select (thd=0x7feea8f3d070, lex=0x7feea61cc090, result=0x7feea63479f8, setup_tables_done_option=0) at /home/elenst/bzr/10.0/sql/sql_select.cc:372
#14 0x000000000066c3a1 in execute_sqlcom_select (thd=0x7feea8f3d070, all_tables=0x7feea61cd6d0) at /home/elenst/bzr/10.0/sql/sql_parse.cc:5250
#15 0x0000000000664815 in mysql_execute_command (thd=0x7feea8f3d070) at /home/elenst/bzr/10.0/sql/sql_parse.cc:2571
#16 0x0000000000686e10 in Prepared_statement::execute (this=0x7feea613b470, expanded_query=0x7feeb0923b30, open_cursor=false) at /home/elenst/bzr/10.0/sql/sql_prepare.cc:3971
#17 0x0000000000685eeb in Prepared_statement::execute_loop (this=0x7feea613b470, expanded_query=0x7feeb0923b30, open_cursor=false, packet=0x0, packet_end=0x0) at /home/elenst/bzr/10.0/sql/sql_prepare.cc:3626
#18 0x00000000006841ee in mysql_sql_stmt_execute (thd=0x7feea8f3d070) at /home/elenst/bzr/10.0/sql/sql_prepare.cc:2777
#19 0x0000000000664843 in mysql_execute_command (thd=0x7feea8f3d070) at /home/elenst/bzr/10.0/sql/sql_parse.cc:2581
#20 0x000000000066eb20 in mysql_parse (thd=0x7feea8f3d070, rawbuf=0x7feea6059088 "EXECUTE stmt", length=12, parser_state=0x7feeb0924630) at /home/elenst/bzr/10.0/sql/sql_parse.cc:6394
#21 0x00000000006619da in dispatch_command (command=COM_QUERY, thd=0x7feea8f3d070, packet=0x7feea8f33071 "EXECUTE stmt", packet_length=12) at /home/elenst/bzr/10.0/sql/sql_parse.cc:1296
#22 0x0000000000660d7c in do_command (thd=0x7feea8f3d070) at /home/elenst/bzr/10.0/sql/sql_parse.cc:993
#23 0x000000000077a4bd in do_handle_one_connection (thd_arg=0x7feea8f3d070) at /home/elenst/bzr/10.0/sql/sql_connect.cc:1379
#24 0x000000000077a210 in handle_one_connection (arg=0x7feea8f3d070) at /home/elenst/bzr/10.0/sql/sql_connect.cc:1293
#25 0x0000000000a9d1f1 in pfs_spawn_thread (arg=0x7feea9775df0) at /home/elenst/bzr/10.0/storage/perfschema/pfs.cc:1853
#26 0x00007feeb0671b50 in start_thread (arg=<optimized out>) at pthread_create.c:304
#27 0x00007feeaf1c0a7d in clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:112

          revision-id: bar@mnogosearch.org-20140127091540-1ij1ogw9l9v8je7h
revno: 3744
branch-nick: 5.3
BUILD/compile-pentium-debug-max-no-ndb

          SET optimizer_switch = 'materialization=on,semijoin=on';

CREATE TABLE t1 (a INT);
INSERT INTO t1 VALUES (2),(3);

CREATE TABLE t2 (b INT);
INSERT INTO t2 VALUES (8),(9);

CREATE TABLE t3 (c INT, INDEX(c));
INSERT INTO t2 VALUES (5),(6);

PREPARE stmt FROM "SELECT * FROM t1 WHERE ( 9, 5 ) IN ( SELECT b, COUNT(*) FROM t2 WHERE 1 IN ( SELECT MIN(c) FROM t3 ) )";
EXECUTE stmt;
EXECUTE stmt;

            Gliffy Diagrams

              Attachments

                Activity

                Ascending order - Click to sort in descending order
                Hide
                Permalink
                psergey Sergei Petrunia added a comment -

                The query plan:

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

                id select_type table type possible_keys key key_len ref rows Extra

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

                1 PRIMARY NULL NULL NULL NULL NULL NULL NULL Impossible WHERE
                2 MATERIALIZED NULL NULL NULL NULL NULL NULL NULL Impossible WHERE
                3 MATERIALIZED NULL NULL NULL NULL NULL NULL NULL No matching min/max row

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

                Both subqueries are non-merged semi-joins (aka JTBMs). Both are degenerate, i.e. they have no rows in the materialized table.
                First EXECUTE creates and frees dummy temporary tables.

                The crash happens on the second execute, when we try to run this code:

                706 tbl->table->maybe_null= 1;

                and the problem is that tbl->table == NULL, the temp.table hasn't been created yet.

                Show
                psergey Sergei Petrunia added a comment - The query plan: - ---------- --- -- ----------- -- ----- -- -- ----------------------+ id select_type table type possible_keys key key_len ref rows Extra - ---------- --- -- ----------- -- ----- -- -- ----------------------+ 1 PRIMARY NULL NULL NULL NULL NULL NULL NULL Impossible WHERE 2 MATERIALIZED NULL NULL NULL NULL NULL NULL NULL Impossible WHERE 3 MATERIALIZED NULL NULL NULL NULL NULL NULL NULL No matching min/max row - ---------- --- -- ----------- -- ----- -- -- ----------------------+ Both subqueries are non-merged semi-joins (aka JTBMs). Both are degenerate, i.e. they have no rows in the materialized table. First EXECUTE creates and frees dummy temporary tables. The crash happens on the second execute, when we try to run this code: 706 tbl->table->maybe_null= 1; and the problem is that tbl->table == NULL, the temp.table hasn't been created yet.
                Hide
                Permalink
                psergey Sergei Petrunia added a comment -

                This bug is not fixed by fix for MDEV-5582.

                Show
                psergey Sergei Petrunia added a comment - This bug is not fixed by fix for MDEV-5582 .
                Hide
                Permalink
                psergey Sergei Petrunia added a comment -

                I'm debugging the second EXECUTE.

                I see that setup_tables() is invoked and does some processing for TABLE_LISTs
                that are non-merged semi-joins (aka JTBMs). However, it doesn't create TABLE
                objects for them.

                TABLE objects are created in setup_jtbm_semi_joins(), which is called from
                parent join's JOIN::optmize().

                However, the code in JOIN::prepare()

                /*
                If the query uses implicit grouping where the select list contains both
                aggregate functions and non-aggregate fields, any non-aggregated field
                may produce a NULL value. Set all fields of each table as nullable before
                semantic analysis to take into account this change of nullability.

                Note: this loop doesn't touch tables inside merged semi-joins, because
                subquery-to-semijoin conversion has not been done yet. This is intended.
                */
                if (mixed_implicit_grouping)
                {
                tbl->table->maybe_null= 1;

                tries to access TABLE* and crashes.

                Show
                psergey Sergei Petrunia added a comment - I'm debugging the second EXECUTE. I see that setup_tables() is invoked and does some processing for TABLE_LISTs that are non-merged semi-joins (aka JTBMs). However, it doesn't create TABLE objects for them. TABLE objects are created in setup_jtbm_semi_joins(), which is called from parent join's JOIN::optmize(). However, the code in JOIN::prepare() /* If the query uses implicit grouping where the select list contains both aggregate functions and non-aggregate fields, any non-aggregated field may produce a NULL value. Set all fields of each table as nullable before semantic analysis to take into account this change of nullability. Note: this loop doesn't touch tables inside merged semi-joins, because subquery-to-semijoin conversion has not been done yet. This is intended. */ if (mixed_implicit_grouping) { tbl->table->maybe_null= 1; tries to access TABLE* and crashes.
                Hide
                Permalink
                psergey Sergei Petrunia added a comment -

                Possible solutions:
                1. Postpone the action done in JOIN::prepare() until there is a TABLE* object.
                For example, remember in TABLE_LIST somewhere that we should set
                TABLE::maybe_null to true and do it once we have the TABLE object.

                2. Perform the actions of setup_jtbm_semi_joins() earlier, at prepare phase.
                Note that we can't call the whole function earlier, because it calls optimize() for the subqueries.

                Show
                psergey Sergei Petrunia added a comment - Possible solutions: 1. Postpone the action done in JOIN::prepare() until there is a TABLE* object. For example, remember in TABLE_LIST somewhere that we should set TABLE::maybe_null to true and do it once we have the TABLE object. 2. Perform the actions of setup_jtbm_semi_joins() earlier, at prepare phase. Note that we can't call the whole function earlier, because it calls optimize() for the subqueries.
                Hide
                Permalink
                psergey Sergei Petrunia added a comment -

                Doing #2 is rather difficult, because there are two ways a temporary table is created
                C1. subquery is not degenerate, subquery optimization process creates a tmp.table
                C2. subquery is degenerate, setup_jtbm_semi_joins() calls create_dummy_tmp_table().

                It is possible to factor out temp.table creation code out of C1 and C2, but this is a significant change.

                Show
                psergey Sergei Petrunia added a comment - Doing #2 is rather difficult, because there are two ways a temporary table is created C1. subquery is not degenerate, subquery optimization process creates a tmp.table C2. subquery is degenerate, setup_jtbm_semi_joins() calls create_dummy_tmp_table(). It is possible to factor out temp.table creation code out of C1 and C2, but this is a significant change.

                  People

                  • Assignee:
                    psergey Sergei Petrunia
                    Reporter:
                    elenst Elena Stepanova
                  • Votes:
                    0 Vote for this issue
                    Watchers:
                    3 Start watching this issue

                    Dates

                    • Created:
                      Updated:
                      Resolved: