Phase 3: Implementation of re-optimization of subqueries with injected predicates
and cost comparison between Materialization and IN->EXISTS strategies.
The commit contains the following known problems:
- The implementation of EXPLAIN has not been re-engineered to reflect the
changes in subquery optimization. EXPLAIN for subqueries is called during
the execute phase, which results in different code paths during JOIN::optimize
and thus in differing EXPLAIN messages for constant/system tables.
- There are some valgrind warnings that need investigation
- Several EXPLAINs with minor differences need to be reconsidered after fixing
the EXPLAIN problem above.
This patch also adds one extra optimizer_switch: 'in_to_exists' for complete
manual control of the subquery execution strategies.
* Fixed a "crack" between semijoin analysis and materialization analysis
where semijoin didn't set the correct strategy for the IN predicate.
* Cosmetic changes in the code/comments.
1. Changed the lazy optimization for subqueries that can be
materialized into bottom-up optimization during the optimization of
the main query.
The main change is implemented by the method
Item_in_subselect::setup_engine.
All other changes were required to correct problems resulting from
changing the order of optimization. Most of these problems followed
the same pattern - there are some shared structures between a
subquery and its parent query. Depending on which one is optimized
first (parent or child query), these shared strucutres may get
different values, thus resulting in an inconsistent query plan.
2. Changed the code-generation for subquery materialization to be
performed in runtime memory for each (re)execution, instead of in
statement memory (once per prepared statement).
- Item_in_subselect::setup_engine() no longer creates materialization
related objects in statement memory.
- Merged subselect_hash_sj_engine::init_permanent and
subselect_hash_sj_engine::init_runtime into
subselect_hash_sj_engine::init, which is called for each
(re)execution.
- Fixed deletion of the temp table accordingly.
mysql-test/r/subselect_mat.result:
Adjusted changed EXPLAIN because of earlier optimization of subqueries.
libmysqld/Makefile.am:
The new file added.
mysql-test/r/index_merge_myisam.result:
subquery_cache optimization option added.
mysql-test/r/myisam_mrr.result:
subquery_cache optimization option added.
mysql-test/r/subquery_cache.result:
The subquery cache tests added.
mysql-test/r/subselect3.result:
Subquery cache switched off to avoid changing read statistics.
mysql-test/r/subselect3_jcl6.result:
Subquery cache switched off to avoid changing read statistics.
mysql-test/r/subselect_no_mat.result:
subquery_cache optimization option added.
mysql-test/r/subselect_no_opts.result:
subquery_cache optimization option added.
mysql-test/r/subselect_no_semijoin.result:
subquery_cache optimization option added.
mysql-test/r/subselect_sj.result:
subquery_cache optimization option added.
mysql-test/r/subselect_sj_jcl6.result:
subquery_cache optimization option added.
mysql-test/t/subquery_cache.test:
The subquery cache tests added.
mysql-test/t/subselect3.test:
Subquery cache switched off to avoid changing read statistics.
sql/CMakeLists.txt:
The new file added.
sql/Makefile.am:
The new files added.
sql/item.cc:
Expression cache item (Item_cache_wrapper) added.
Item_ref and Item_field fixed for correct usage of result field and fast resolwing in SP.
sql/item.h:
Expression cache item (Item_cache_wrapper) added.
Item_ref and Item_field fixed for correct usage of result field and fast resolwing in SP.
sql/item_cmpfunc.cc:
Subquery cache added.
sql/item_cmpfunc.h:
Subquery cache added.
sql/item_subselect.cc:
Subquery cache added.
sql/item_subselect.h:
Subquery cache added.
sql/item_sum.cc:
Registration of subquery parameters added.
sql/mysql_priv.h:
subquery_cache optimization option added.
sql/mysqld.cc:
subquery_cache optimization option added.
sql/opt_range.cc:
Fix due to subquery cache.
sql/opt_subselect.cc:
Parameters of the function cahnged.
sql/procedure.h:
.h file guard added.
sql/sql_base.cc:
Registration of subquery parameters added.
sql/sql_class.cc:
Option to allow add indeces to temporary table.
sql/sql_class.h:
Item iterators added.
Option to allow add indeces to temporary table.
sql/sql_expression_cache.cc:
Expression cache for caching subqueries added.
sql/sql_expression_cache.h:
Expression cache for caching subqueries added.
sql/sql_lex.cc:
Registration of subquery parameters added.
sql/sql_lex.h:
Registration of subqueries and subquery parameters added.
sql/sql_select.cc:
Subquery cache added.
sql/sql_select.h:
Subquery cache added.
sql/sql_union.cc:
A new parameter to the function added.
sql/sql_update.cc:
A new parameter to the function added.
sql/table.cc:
Procedures to manage temporarty tables index added.
sql/table.h:
Procedures to manage temporarty tables index added.
storage/maria/ha_maria.cc:
Fix of handler to allow destoy a table in case of error during the table creation.
storage/maria/ha_maria.h:
.h file guard added.
storage/myisam/ha_myisam.cc:
Fix of handler to allow destoy a table in case of error during the table creation.
- Item_in_subselect::init_left_expr_cache() should not try to
guess whether the left expression is accessed "over the grouping operation"
(i.e. the subselect is evaluated after the grouping while the left_expr is
an Item_ref that wraps an expression from before the grouping). Instead,
let new_Cached_item not to try accessing item->real_item() when creating
left expr cache.
- fix a bug in LooseScan strategy execution code (exposed by changing costs/QEP)
- Do set join_tab->sorted=TRUE for JOIN_TABs that use LooseScan (partitioning
handler cares about "sorted" parameter of h->index_init() call)
* remove handler::index_read_last()
* create handler::keyread_read_time() (was get_index_only_read_time() in opt_range.cc)
* ha_show_status() allows engine's show_status() to fail
* remove HTON_FLUSH_AFTER_RENAME
* fix key_cmp_if_same() to work for floats and doubles
* set table->status in the server, don't force engines to do it
* increment status vars in the server, don't force engines to do it
mysql-test/r/status_user.result:
correct test results - innodb was wrongly counting internal
index searches as handler_read_* calls.
sql/ha_partition.cc:
compensate for handler incrementing status counters -
we want to count only calls to underlying engines
sql/handler.h:
inline methods moved to sql_class.h
sql/key.cc:
simplify the check
sql/opt_range.cc:
move get_index_only_read_time to the handler class
sql/sp.cc:
don't use a key that's stored in the record buffer -
the engine can overwrite the buffer with anything, destroying the key
sql/sql_class.h:
inline handler methods that need to see THD and TABLE definitions
sql/sql_select.cc:
no ha_index_read_last_map anymore
sql/sql_table.cc:
remove HTON_FLUSH_AFTER_RENAME
sql/table.cc:
set HA_CAN_MEMCMP as appropriate
sql/tztime.cc:
don't use a key that's stored in the record buffer -
the engine can overwrite the buffer with anything, destroying the key
storage/myisam/ha_myisam.cc:
engines don't need to update table->status or use ha_statistic_increment anymore
storage/myisam/ha_myisam.h:
index_read_last_map is no more
Fixed some bugs introduced in 5.1.47
Disabled some tests until we have merged with latest Xtradb
configure.in:
Added testing if valgrind/memcheck.h exists
storage/pbxt/src/ha_pbxt.cc:
LOCK_plugin is not anymore locked in init
- INSERT with RAND() doesn't require row based logging again
- Some bugs fixed in opt_range() where we table->key_read was wrongly used
.bzrignore:
Ignore new xtstat binary
mysql-test/r/index_merge_myisam.result:
Update results (old result was wrong)
mysql-test/suite/binlog/r/binlog_stm_binlog.result:
Added drop table first
mysql-test/suite/binlog/r/binlog_stm_unsafe_warning.result:
Added test for when RAND() requires row based logging
mysql-test/suite/binlog/t/binlog_stm_binlog.test:
Added drop table first
mysql-test/suite/binlog/t/binlog_stm_unsafe_warning.test:
Added test for when RAND() requires row based logging
scripts/make_binary_distribution.sh:
Removed type from last commit
sql/item_create.cc:
Don't require row based logging when using RAND() with INSERT
sql/opt_range.cc:
Revert wrong patch from Oracle:
- As QUICK_RANGE_SELECT uses it's own 'file' handler to the tables, one can't use 'table->key_read' as a flag to detect if index only read (keyread) is used or not
- Don't set keyread if keyread is already enabled
- Don't disable key read, if we didn't enable it ourselves
- Simplify code (and ensure that we do proper cleanup of index only read)
sql/opt_range.h:
Added flags to detect if the range optimizer enabled index only read (key read) or not
sql/opt_sum.cc:
Use our more optimized macros
sql/sql_lex.h:
Added 'readable' function to check if we are in a sub query function or not (not normal query or sub query in FROM clause)
sql/sql_select.cc:
Use our more optimized keyread macros
Added ASSERTS early
Simplify code on eliminate_item_equal()
Fixed that substitute_for_best_equal_field() doesn't core dump in case of out of memory conditions.
Removed not needed test for 'field->maybe_null()'
Replaced master_unit()->item with is_subquery_function() (More readable)
sql/sql_update.cc:
Use our more optimized keyread macros
sql/table.cc:
Use our more optimized keyread macros
sql/table.h:
Use separate functions to enable/disable Index only reads
- Safer, more readable, better logging and faster.
WHERE predicates containing references to empty tables in a
subquery were handled incorrectly by the optimizer when
executing EXPLAIN. As a result, the optimizer could try to
evaluate such predicates rather than just stop with
"Impossible WHERE noticed after reading const tables" as
it would do in a non-subquery case. This led to valgrind
errors and crashes.
Fixed the code checking the above condition so that subqueries
are not excluded and hence are handled in the same way as top
level SELECTs.
mysql-test/r/explain.result:
Added a test case for bug #48419.
mysql-test/r/ps.result:
Updated test results to take the new (and more correct)
"Extra" comments in execution plans.
mysql-test/t/explain.test:
Added a test case for bug #48419.
sql/sql_select.cc:
There is no point in excluding subqueries from checking
for identically false WHERE conditions.
The problem is that we can not use make_cond_for_table().
This function relies on used_tables() condition
which is not set properly for subqueries.
As result subquery is not filtered out.
The fix is to use remove_eq_conds() function instead
of make_cond_for_table() func. 'remove_eq_conds()'
algorithm relies on const_item() value and it allows
to handle subqueries in right way.
mysql-test/r/having.result:
test case
mysql-test/t/having.test:
test case
sql/sql_select.cc:
The fix is to use remove_eq_conds() function instead
of make_cond_for_table() function.
- Add Item_in_subselect::get_identifier() that returns subquery's id
- Change select_describe() to produce output in new format
- Update test results (checked)
The crash is the result of an attempt made by JOIN::optimize to evaluate
the WHERE condition when no records have been actually read.
The fix is to remove erroneous 'outer_join' variable check.
mysql-test/r/join.result:
test result
mysql-test/t/join.test:
test case
sql/sql_select.cc:
removed erroneous 'outer_join' variable check.
The crash happens because greedy_serach
can not determine best plan due to
wrong inner table dependences. These
dependences affects join table sorting
which performs before greedy_search starting.
In our case table which has real 'no dependences'
should be put on top of the list but it does not
happen as inner tables have no dependences as well.
The fix is to exclude RAND_TABLE_BIT mask from
condition which checks if table dependences
should be updated.
mysql-test/r/join.result:
test result
mysql-test/t/join.test:
test case
sql/sql_select.cc:
RAND_TABLE_BIT mask should not be counted as it
prevents update of inner table dependences.
For example it might happen if RAND() function
is used in JOIN ON clause.
mysql-test/r/union.result:
Update test results (checked, it should be "DEPENDENT SUBQUERY" as it will actually run subquery's join multiple times. There is no reason to do that but executor will still do it)
The problem is that when we make conditon for
grouped result const part of condition is cut off.
It happens because some parts of 'having' condition
which refer to outer join become const after
make_join_statistics. These parts may be lost
during further having condition transformation
in JOIN::exec. The fix is adding 'having'
condition check for const tables after
make_join_statistics is performed.
mysql-test/r/having.result:
test case
mysql-test/t/having.test:
test result
sql/sql_select.cc:
added 'having' condition check for const tables
after make_join_statistics is performed.
The crash happens because of discrepancy between values of
conts_tables and join->const_table_map(make_join_statisctics).
Calculation of conts_tables used condition with
HA_STATS_RECORDS_IS_EXACT flag check. Calculation of
join->const_table_map does not use this flag check.
In case of MERGE table without union with index
the table does not become const table and
thus join_read_const_table() is not called
for the table. join->const_table_map supposes
this table is const and later in make_join_select
this table is used for making&calculation const
condition. As table record buffer is not populated
it leads to crash.
The fix is adding a check if an engine supports
HA_STATS_RECORDS_IS_EXACT flag before updating
join->const_table_map.
mysql-test/r/merge.result:
test result
mysql-test/t/merge.test:
test case
sql/sql_select.cc:
adding a check if an engine supports
HA_STATS_RECORDS_IS_EXACT flag before updating
join->const_table_map.
Fix two problems:
1. Let optimize_semijoin_nests() reset sj_nest->sjmat_info irrespectively
of value of optimizer_flag. We need this in case somebody has turned optimization
off between reexecutions of the same statement.
2. Do not pull out constant tables out of semi-join nests. The problem is that pullout
operation is not undoable, and if a table is constant because it is 1/0-row table it
may cease to be constant on the next execution. Note that tables that are constant
because of possible eq_ref(const) access will still be pulled out as they are
considered functionally-dependent.
BUG#50019: Wrong result for IN-subquery with materialization
- Fix equality substitution in presense of semi-join materialization, lookup and scan variants
(started off from fix by Evgen Potemkin, then modified it to work in all cases)
tmp_table_size can now be set to 0 (to disable in memory internal temp tables)
Improved speed for internal Maria temp tables:
- Don't use packed keys, except with long text fields.
- Don't copy key all accessed pages during key search.
Some new benchmark tests to sql-bench (for group by)
BUILD/compile-pentium64-gcov:
Update script to use same pentium_config flags as other tests
BUILD/compile-pentium64-gprof:
Update script to use same pentium_config flags as other tests
include/my_sys.h:
Added count of my_sync calls
mysql-test/r/variables.result:
tmp_table_size can now be set to 0
sql-bench/test-select.sh:
Added some new test for GROUP BY on a not key field and group by with different order by
sql/mysqld.cc:
Added count of my_sync calls
tmp_table_size can now be set to 0 (to disable in memory internal temp tables)
sql/sql_select.cc:
If tmp_table_size is 0, don't use in memory temp tables (good for benchmarking MyISAM/Maria temp tables)
Don't pack keys for Maria tables; The 8K page size makes packed keys too slow for temp tables.
storage/maria/ma_key_recover.h:
Moved definition to maria_def.h
storage/maria/ma_page.c:
Moved code used to simplify comparing of identical Maria tables to own function (page_cleanup())
Fixed that one can read a page with a read lock.
storage/maria/ma_rkey.c:
For not exact key reads, cache the page where we found key (to speed up future read-next/read-prev calls)
storage/maria/ma_search.c:
Moved code to cache last key page to separate function.
Instead of copying pages, only get a link to the page. This notable speeds up key searches on bigger tables.
storage/maria/ma_write.c:
Added comment
storage/maria/maria_def.h:
Moved page_cleanup() to separate function.
SunStudio
SunStudio compilers of late warn about methods that might hide
methods in base classes due to the use of overloading combined
with overriding. SunStudio also warns about variables defined
in local socpe or method arguments that have the same name as
a member attribute of the class.
This patch renames methods that might hide base class methods,
to make it easier both for humans and compilers to see what is
actually called. It also renames variables in local scope.
sql/field.cc:
Local scope variable or method argument same as class
attribute.
sql/item_cmpfunc.cc:
Local scope variable or method argument same as class
attribute.
sql/item_create.cc:
Renaming base class create() to create_func().
sql/item_create.h:
Renaming base class create() to create_func().
sql/protocol.cc:
Local scope variable or method argument same as class
attribute.
sql/sql_profile.cc:
Local scope variable or method argument same as class
attribute.
sql/sql_select.cc:
Local scope variable or method argument same as class
attribute.
sql/sql_yacc.yy:
Renaming base class create() to create_func().
storage/federated/ha_federated.cc:
Local scope variable or method argument same as class
attribute.
storage/myisammrg/ha_myisammrg.cc:
Local scope variable or method argument same as class
attribute.
The problem is that not all column names retrieved from a SELECT
statement can be used as view column names due to length and format
restrictions. The server failed to properly check the conformity
of those automatically generated column names before storing the
final view definition on disk.
Since columns retrieved from a SELECT statement can be anything
ranging from functions to constants values of any format and length,
the solution is to rewrite to a pre-defined format any names that
are not acceptable as a view column name.
The name is rewritten to "Name_exp_%u" where %u translates to the
position of the column. To avoid this conversion scheme, define
explict names for the view columns via the column_list clause.
Also, aliases are now only generated for top level statements.
mysql-test/include/view_alias.inc:
Add test case for Bug#40277
mysql-test/r/compare.result:
Bug#40277: SHOW CREATE VIEW returns invalid SQL
mysql-test/r/group_by.result:
Bug#40277: SHOW CREATE VIEW returns invalid SQL
mysql-test/r/ps.result:
Bug#40277: SHOW CREATE VIEW returns invalid SQL
mysql-test/r/subselect.result:
Bug#40277: SHOW CREATE VIEW returns invalid SQL
mysql-test/r/subselect3.result:
Bug#40277: SHOW CREATE VIEW returns invalid SQL
mysql-test/r/type_datetime.result:
Bug#40277: SHOW CREATE VIEW returns invalid SQL
mysql-test/r/union.result:
Bug#40277: SHOW CREATE VIEW returns invalid SQL
mysql-test/r/view.result:
Add test case result for Bug#40277
mysql-test/r/view_alias.result:
Add test case result for Bug#40277
mysql-test/t/view_alias.test:
Add test case for Bug#40277
sql/sql_view.cc:
Check if auto generated column names are conforming. Also, the
make_unique_view_field_name function is not used as it uses the
original name to construct a new one, which does not work if the
name is invalid.
- The problem was that DuplicateWeedout strategy setup code wasn't aware of the
fact that join buffering will be used and applied optimization that doesn't work
together with join buffering. Fixed by making DuplicateWeedout setup code to have
a pessimistic check about whether there is a chance that join buffering will be
used.
- Make JOIN_CACHE_BKA::init() correctly process Copy_field elements that denote saving
current rowids in the join buffer.
mysql-test/r/subselect_sj2.result:
Update test results
mysql-test/r/subselect_sj2_jcl6.result:
Update test results
mysql-test/r/subselect_sj_jcl6.result:
Testcase
mysql-test/t/subselect_sj2.test:
Update test results
mysql-test/t/subselect_sj_jcl6.test:
Testcase
sql/opt_subselect.cc:
- The problem was that DuplicateWeedout strategy setup code wasn't aware of the
fact that join buffering will be used and applied optimization that doesn't work
together with join buffering. Fixed by making DuplicateWeedout setup code to have
a pessimistic check about whether there is a chance that join buffering will be
used.
sql/sql_join_cache.cc:
Make JOIN_CACHE_BKA::init() correctly process Copy_field elements that denote saving current rowids in the join buffer.
sql/sql_select.cc:
Added a question note
Save no-records constant tables in JOIN::const_table_map before we invoke
eliminate_tables(). Failure to do so caused crash when the same table was
marked as constant two times
mysql-test/r/table_elim.result:
LPBUG#524025 Running RQG outer_join test leads to crash
- Testcase
mysql-test/t/table_elim.test:
LPBUG#524025 Running RQG outer_join test leads to crash
- Testcase
sql/sql_select.cc:
LPBUG#524025 Running RQG outer_join test leads to crash
Save no-records constant tables in JOIN::const_table_map before we invoke eliminate_tables(). Failure to do so caused crash when the same table was marked as constant two times.
consider clustered primary keys
Choosing a shortest index for the covering index scan,
the optimizer ignored the fact, that the clustered primary
key read involves whole table data.
The find_shortest_key function has been modified to
take into account that fact that a clustered PK has a
longest key of possible covering indices.
mysql-test/r/innodb_mysql.result:
Test case for bug #39653.
mysql-test/t/innodb_mysql.test:
Test case for bug #39653.
sql/sql_select.cc:
Bug #39653: find_shortest_key in sql_select.cc does not
consider clustered primary keys
The find_shortest_key function has been modified to
take into account that fact that a clustered PK has a
longest key of possible covering indices.
