The wrong result set returned by the left join query from
the bug test case happened due to several inconsistencies
and bugs of the legacy mysql code.
The bug test case uses an execution plan that employs a scan
of a materialized IN subquery from the WHERE condition.
When materializing such an IN- subquery the optimizer injects
additional equalities into the WHERE clause. These equalities
express the constraints imposed by the subquery predicate.
The injected equality of the query in the test case happens
to belong to the same equality class, and a new equality
imposing a condition on the rows of the materialized subquery
is inferred from this class. Simultaneously the multiple
equality is added to the ON expression of the LEFT JOIN
used in the main query.
The inferred equality of the form f1=f2 is taken into account
when optimizing the scan of the rows the temporary table
that is the result of the subquery materialization: only the
values of the field f1 are read from the table into the record
buffer. Meanwhile the inferred equality is removed from the
WHERE conditions altogether as a constraint on the fields
of the temporary table that has been used when filling this table.
This equality is supposed to be removed from the ON expression
when the multiple equalities of the ON expression are converted
into an optimal set of equality predicates. It supposed to be
removed from the ON expression as an equality inferred from only
equalities of the WHERE condition. Yet, it did not happened
due to the following bug in the code.
Erroneously the code tried to build multiple equality for ON
expression twice: the first time, when it called optimize_cond()
for the WHERE condition, the second time, when it called
this function for the HAVING condition. When executing
optimize_con() for the WHERE condition a reference
to the multiple equality of the WHERE condition is set
in the multiple equality of the ON expression. This reference
would allow later to convert multiple equalities of the
ON expression into equality predicates. However the
the second call of build_equal_items() for the ON expression
that happened when optimize_cond() was called for the
HAVING condition reset this reference to NULL.
This bug fix blocks calling build_equal_items() for ON
expressions for the second time. In general, it will be
beneficial for many queries as it removes from ON
expressions any equalities that are to be checked for the
WHERE condition.
The patch also fixes two bugs in the list manipulation
operations and a bug in the function
substitute_for_best_equal_field() that resulted
in passing wrong reference to the multiple equalities
of where conditions when processing multiple
equalities of ON expressions.
The code of substitute_for_best_equal_field() and
the code the helper function eliminate_item_equal()
were also streamlined and cleaned up.
Now the conversion of the multiple equalities into
an optimal set of equality predicates first produces
the sequence of the all equalities processing multiple
equalities one by one, and, only after this, it inserts
the equalities at the beginning of the other conditions.
The multiple changes in the output of EXPLAIN
EXTENDED are mainly the result of this streamlining,
but in some cases is the result of the removal of
unneeded equalities from ON expressions. In
some test cases this removal were reflected in the
output of EXPLAIN resulted in disappearance of
“Using where” in some rows of the execution plans.
This bug happened because the executor tried to use a wrong
TABLE REF object when building access keys. It constructed
keys from fields of a materialized table from a ref object
created to construct keys from the fields of the underlying
base table. This could happen only when materialized table
was created for a non-correlated IN subquery and only
when the materialized table used for lookups.
In this case we are guaranteed to be able to construct the
keys from the fields of tables that would be outer tables
for the tables of the IN subquery.
The patch makes sure that no ref objects constructed from
fields of materialized lookup tables are to be used.
This is a backport of the fix for:
Bug#13633549 HANDLE_FATAL_SIGNAL IN TEST_IF_SKIP_SORT_ORDER/CREATE_SORT_INDEX
Don't invoke the range optimizer for a NULL select.
Some queries with the "SELECT ... FROM DUAL" nested subqueries
failed with an assertion on debug builds.
Non-debug builds were not affected.
There were a few different issues with similar assertion
failures on different queries:
1. The first problem was related to the incomplete propagation
of the "non-constant" item status from underlying subquery
items to the outer item tree: in some cases non-constants were
interpreted as constants and evaluated at the preparation stage
(val_int() calls withing fix_fields() etc).
Thus, the default implementation of Item_ref::const_item() from
the Item parent class didn't take into account the "const_item"
status of the referenced item tree -- it used the insufficient
"used_tables() == 0" check instead. This worked in most cases
since our "non-constant" functions like RAND() and SLEEP() set
the RAND_TABLE_BIT in the used table map, so they aren't
non-constant from Item_ref's "point of view". However, the
"SELECT ... FROM DUAL" subquery may have an empty map of used
tables, but at the same time subqueries are never "constant" at
the context analysis stage (preparation, view creation etc).
So, the non-contantness of such subqueries was missed.
Fix: the Item_ref::const_item() function has been overloaded to
take into account both (*ref)->const_item() status and tricky
Item_ref::used_tables() return values, since the only
(*ref)->const_item() call is not enough there.
2. In some cases instead of the const_item() call we check a
value of the Item::with_subselect field to recognize items
with nested subqueries. However, the Item_ref class didn't
propagate this value from the referenced item tree.
Fix: Item::has_subquery() and Item_ref::has_subquery()
functions have been backported from 5.6. All direct
references to the with_subselect fields of nested items have
been with the has_subquery() function call.
3. The Item_func_regex class didn't propagate with_subselect
as well, since it overloads the Item_func::fix_fields()
function with insufficient fix_fields() implementation.
Fix: the Item_func_regex::fix_fields() function has been
modified to gather "constant" statuses from inner items.
4. The Item_func_isnull::update_used_tables() function has
a special branch for the underlying item where the maybe_null
value is false: in this case it marks the Item_func_isnull
as a "const_item" and sets the cached_value to false.
However, the Item_func_isnull::val_int() was not in sync with
update_used_tables(): it didn't take into account neither
const_item_cache nor cached_value for the case of
"args[0]->maybe_null == false optimization".
As far as such an Item_func_isnull has "const_item() == true",
it's ok to call Item_func_isnull::val_int() etc from outer
items on preparation stage. In this case the server tried to
call Item_func_isnull::args[0]->isnull(), and if the args[0]
item contained a nested not-nullable subquery, it failed
with an assertion.
Fix: take the value of Item_func_isnull::const_item_cache into
account in the val_int() function.
5. The auxiliary Item_is_not_null_test class has a similar
optimization in the update_used_tables() function as the
Item_func_isnull class has, and the same issue in the val_int()
function.
In addition to that the Item_is_not_null_test::update_used_tables()
doesn't update the const_item_cache value, so the "maybe_null"
optimization is useless there. Thus, we missed some optimizations
of cases like these (before and after the fix):
< <is_not_null_test>(a),
---
> <cache>(<is_not_null_test>(a)),
or
< having (<is_not_null_test>(a) and <is_not_null_test>(a))
---
> having 1
etc.
Fix: update Item_is_not_null_test::const_item_cache in
update_used_tables() and take in into account in val_int().
The previous fix for MDEV-3992 was incomplete, because it still computed
incorrectly the number of keyparts of the extended secondary key in the
case when columns of the PK participate in the secondary key.
This patch by Monty corrects the above problem.
Analysis:
---------
When the server is out of memory, an error is raised
to indicate the same. Handling the error requires
more memory to be allocated which fails, hence the
error handling loops in a recursion and causes the
server to crash.
Fix:
---
a) Prevents pushing the 'out of memory' error condition
to the diagnostic area as it requires memory allocation.
GET DIAGNOSTICS, SHOW WARNINGS and SHOW ERRORS statements
will not show information about this error. However the
'out of memory' error is returned to the client.
b) It sets the ME_FATALERROR flag when 'out of memory' errors
are reported (for places where the flag is not already set).
This flag prevents activation of SP error handlers which also
require memory allocation and therefore are likely to fail.
WITH A VARIABLE AND ORDER BY
Bug#16035412 MYSQL SERVER 5.5.29 WRONG SORTING USING COMPLEX INDEX
This is a fix for a regression introduced by Bug#12667154:
Bug#12667154 attempted to fix a performance problem with subqueries
that did filesort. For doing filesort, the optimizer creates a quick
select object to use when building the sort index. This quick select
object was deleted after the first call to create_sort_index(). Thus,
for queries where the subquery was executed multiple times, the quick
object was only used for the first execution. For all later executions
of the subquery, filesort used a complete table scan for building the
sort index. The fix for Bug#12667154 tried to fix this by not deleting
the quick object after the first execution of create_sort_index() so
that it would be re-used for building the sort index by the following
executions of the subquery.
This regression introduced in Bug#12667154 is that due to not deleting
the quick select object after building the sort index, the quick
object could in some cases be used also during the second phase of the
execution of the subquery instead of using the created sort
index. This caused wrong results to be returned.
The fix for this issue is to delete the reference to the select object
after it has been used in create_sort_index(). In this way the select
and quick objects will not be available when doing the second phase
of the execution of the select operation. To ensure that the select
object can be re-used for the following executions of the subquery
we make a copy of the select pointer. This is used for restoring the
select object after the select operation is completed.
mysql-test/suite/innodb/r/innodb_mysql.result:
Changed explain output: The explain now contains "Using where" since we
have restored the select pointer after doing the filesort operation.
sql/sql_select.cc:
Change create_sort_index() so that it always sets the pointer to
the select object to NULL. This is done in order to avoid that the
select->quick object can be used when execution the main part of
the select operation.
sql/sql_select.h:
New member in JOIN_TAB: saved_select. Used by create_sort_index to
make a backup copy of the select pointer.
Analysis:
The crash is a result of incorrect analysis of whether a secondary key
can be extended with a primary in order to compute ORDER BY. The analysis
is done in test_if_order_by_key(). This function doesn't take into account
that the primary key may in fact index the same columns as the secondary
key. For the test query test_if_order_by_key says that there is an extended
key with total 2 keyparts.
At the same time, the condition
if (pkinfo->key_part[i].field->key_start.is_set(nr))
in test_if_cheaper_oredring() becomes true for (i == 0), which results in
an invalid access to rec_per_key[-1].
Solution:
The best solution would be to reuse KEY::ext_key_parts that is already computed
by open_binary_frm(), however after detailed analysis the conclusion is that
the change would be too intrusive for a GA release.
The solution for 5.5 is to add a guard for the case when the 0-th key part is
considered, and to assume that all keys will be scanned in this case.
WITH A VARIABLE AND ORDER BY
Bug#16035412 MYSQL SERVER 5.5.29 WRONG SORTING USING COMPLEX INDEX
This is a fix for a regression introduced by Bug#12667154:
Bug#12667154 attempted to fix a performance problem with subqueries
that did filesort. For doing filesort, the optimizer creates a quick
select object to use when building the sort index. This quick select
object was deleted after the first call to create_sort_index(). Thus,
for queries where the subquery was executed multiple times, the quick
object was only used for the first execution. For all later executions
of the subquery, filesort used a complete table scan for building the
sort index. The fix for Bug#12667154 tried to fix this by not deleting
the quick object after the first execution of create_sort_index() so
that it would be re-used for building the sort index by the following
executions of the subquery.
This regression introduced in Bug#12667154 is that due to not deleting
the quick select object after building the sort index, the quick
object could in some cases be used also during the second phase of the
execution of the subquery instead of using the created sort
index. This caused wrong results to be returned.
The fix for this issue is to delete the reference to the select object
after it has been used in create_sort_index(). In this way the select
and quick objects will not be available when doing the second phase
of the execution of the select operation. To ensure that the select
object can be re-used for the following executions of the subquery
we make a copy of the select pointer. This is used for restoring the
select object after the select operation is completed.
mysql-test/suite/innodb/r/innodb_mysql.result:
Changed explain output: The explain now contains "Using where" since we
have restored the select pointer after doing the filesort operation.
sql/sql_select.cc:
Change create_sort_index() so that it always sets the pointer to
the select object to NULL. This is done in order to avoid that the
select->quick object can be used when execution the main part of
the select operation.
sql/sql_select.h:
New member in JOIN_TAB: saved_select. Used by create_sort_index to
make a backup copy of the select pointer.
The bug could lead to a wrong estimate of the number of expected rows
in the output of the EXPLAIN commands for queries with GROUP BY.
This could be observed in the test case for LP bug 934348.
KILL now breaks locks inside InnoDB
Fixed possible deadlock when running INNODB STATUS
Added ha_kill_query() and kill_query() to send kill signal to all storage engines
Added reset_killed() to ensure we don't reset killed state while awake() is getting called
include/mysql/plugin.h:
Added thd_mark_as_hard_kill()
include/mysql/plugin_audit.h.pp:
Added thd_mark_as_hard_kill()
include/mysql/plugin_auth.h.pp:
Added thd_mark_as_hard_kill()
include/mysql/plugin_ftparser.h.pp:
Added thd_mark_as_hard_kill()
sql/handler.cc:
Added ha_kill_query() to send kill signal to all storage engines
sql/handler.h:
Added ha_kill_query() and kill_query() to send kill signal to all storage engines
sql/log_event.cc:
Use reset_killed()
sql/mdl.cc:
use thd->killed instead of thd_killed() to abort on soft kill
sql/sp_rcontext.cc:
Use reset_killed()
sql/sql_class.cc:
Fixed possible deadlock in INNODB STATUS by not getting thd->LOCK_thd_data if it's locked.
Use reset_killed()
Tell storge engines that KILL has been sent
sql/sql_class.h:
Added reset_killed() to ensure we don't reset killed state while awake() is getting called.
Added mark_as_hard_kill()
sql/sql_insert.cc:
Use reset_killed()
sql/sql_parse.cc:
Simplify detection of killed queries.
Use reset_killed()
sql/sql_select.cc:
Use reset_killed()
sql/sql_union.cc:
Use reset_killed()
storage/innobase/handler/ha_innodb.cc:
Added innobase_kill_query()
Fixed error reporting for interrupted queries.
storage/xtradb/handler/ha_innodb.cc:
Added innobase_kill_query()
Fixed error reporting for interrupted queries.
The problem is a shift operation that is not 64-bit safe.
The consequence is that used tables information for a join with 32 tables
or more will be incorrect.
Fixed by adding a type cast in Item_sum::update_used_tables().
Also used the opportunity to fix some other potential bugs by adding an
explicit type-cast to an integer in a left-shift operation.
Some of them were quite harmless, but was fixed in order to get the same
signed-ness as the other operand of the operation it was used in.
sql/item_cmpfunc.cc
Adjusted signed-ness for some integers in left-shift.
sql/item_subselect.cc
Added type-cast to nesting_map (which is a 32/64 bit type, so
potential bug for deeply nested queries).
sql/item_sum.cc
Added type-cast to nesting_map (32/64-bit type) and table_map
(64-bit type).
sql/opt_range.cc
Added type-cast to ulonglong (which is a 64-bit type).
sql/sql_base.cc
Added type-cast to nesting_map (which is a 32/64-bit type).
sql/sql_select.cc
Added type-cast to nesting_map (32/64-bit type) and key_part_map
(64-bit type).
sql/strfunc.cc
Changed type-cast from longlong to ulonglong, to preserve signed-ness.
Analysys:
In the beginning of JOIN::cleanup there is code that is supposed to
free all filesort buffers. The code assumes that the table being sorted
is the first non-constant table. To get this table it calls:
first_top_level_tab(this, WITHOUT_CONST_TABLES)
However, first_top_level_tab() instead returned the wrong table - the first
one in the plan, instead of the first non-constant table. There is no other
place outside filesort() where sort buffers may be freed. As a result, the
sort buffer was not freed, and there was a memory leak.
Solution:
Change first_top_level_tab(), to test for WITH_CONST_TABLES instead of
WITHOUT_CONST_TABLES.
The problem was that in debugging binaries it try to print item to assign human readable name to the item.
But subquery item was already freed (join_free/cleanup with full cleanup) so Item_field refers to temporary
table which memory had been already freed.
If the setting of system variables does not allow to use join buffer
for a join query with GROUP BY <f1,...> / ORDER BY <f1,...> then
filesort is not needed if the first joined table is scanned in
the order compatible with order specified by the list <f1,...>.
MDEV-567: Wrong result from a query with correlated subquery if ICP is allowed:
backport the fix developed for SHOW EXPLAIN:
revision-id: psergey@askmonty.org-20120719115219-212cxmm6qvf0wlrb
branch nick: 5.5-show-explain-r21
timestamp: Thu 2012-07-19 15:52:19 +0400
BUG#992942 & MDEV-325: Pre-liminary commit for testing
and adjust it so that it handles DS-MRR scans correctly.
If, when executing a query with ORDER BY col LIMIT n, the optimizer chose
an index-merge scan to access the table containing col while there existed
an index defined over col then optimizer did not consider the possibility
of using an alternative range scan by this index to avoid filesort. This
could cause a performance degradation if the optimizer flag index_merge was
set up to 'on'.
Problem:-
When we execute a query which has subquery with GROUP BY, ORDER BY and have a
BLOB column,results a memory leak.
Analysis:-
In case of subquery, which have GROUP BY on BLOB and a ORDER BY on other field
and BLOB is not a key. We allocate a tmp buffer to copy_field to take care of
BLOB value.This copy_field value can have copies of its in two join(objects),
so while freeing this copy_field we have to take care that it is
not deleted twice.
The double deletion of tmp_table_param.copy_field is handled by two patches.
One by Kostja :
revid:sp1r-konstantin@mysql.com-20050627101056-55153
Fix the broken test suite in -debug build.
and other by Oleksandr
revid:sp1r-bell@sanja.is.com.ua-20060118114857-19905
Excluded posibility of tmp_table_param.copy_field double deletion (BUG#14851).
both of this patches are commited in different branch and while
merging they both get placed,but there is no need for Kostja patch as Oleksandr
patch handle this.
sql/sql_select.cc:
Bug13726751, tmp_join clean up is not necessary as later in the code we are taking care of cleaning up of tmp_join copy_field.
Problem:-
When we execute a query which has subquery with GROUP BY, ORDER BY and have a
BLOB column,results a memory leak.
Analysis:-
In case of subquery, which have GROUP BY on BLOB and a ORDER BY on other field
and BLOB is not a key. We allocate a tmp buffer to copy_field to take care of
BLOB value.This copy_field value can have copies of its in two join(objects),
so while freeing this copy_field we have to take care that it is
not deleted twice.
The double deletion of tmp_table_param.copy_field is handled by two patches.
One by Kostja :
revid:sp1r-konstantin@mysql.com-20050627101056-55153
Fix the broken test suite in -debug build.
and other by Oleksandr
revid:sp1r-bell@sanja.is.com.ua-20060118114857-19905
Excluded posibility of tmp_table_param.copy_field double deletion (BUG#14851).
both of this patches are commited in different branch and while
merging they both get placed,but there is no need for Kostja patch as Oleksandr
patch handle this.
sql/sql_select.cc:
Bug13726751, tmp_join clean up is not necessary as later in the code we are taking care of cleaning up of tmp_join copy_field.
Fix by Sergey Petrunia.
This patch only prevents the evaluation of expensive subqueries during optimization.
The crash reported in this bug has been fixed by some other patch.
The fix is to call value->is_null() only when !value->is_expensive(), because is_null()
may trigger evaluation of the Item, which in turn triggers subquery evaluation if the
Item is a subquery.
.. into MariaDB 5.3
Fix for Bug#12667154 SAME QUERY EXEC AS WHERE SUBQ GIVES DIFFERENT
RESULTS ON IN() & NOT IN() COMP #3
This bug causes a wrong result in mysql-trunk when ICP is used
and bad performance in mysql-5.5 and mysql-trunk.
Using the query from bug report to explain what happens and causes
the wrong result from the query when ICP is enabled:
1. The t3 table contains four records. The outer query will read
these and for each of these it will execute the subquery.
2. Before the first execution of the subquery it will be optimized. In
this case the important is what happens to the first table t1:
-make_join_select() will call the range optimizer which decides
that t1 should be accessed using a range scan on the k1 index
It creates a QUICK_RANGE_SELECT object for this.
-As the last part of optimization the ICP code pushes the
condition down to the storage engine for table t1 on the k1 index.
This produces the following information in the explain for this table:
2 DEPENDENT SUBQUERY t1 range k1 k1 5 NULL 3 Using index condition; Using filesort
Note the use of filesort.
3. The first execution of the subquery does (among other things) due
to the need for sorting:
a. Call create_sort_index() which again will call find_all_keys():
b. find_all_keys() will read the required keys for all qualifying
rows from the storage engine. To do this it checks if it has a
quick-select for the table. It will use the quick-select for
reading records. In this case it will read four records from the
storage engine (based on the range criteria). The storage engine
will evaluate the pushed index condition for each record.
c. At the end of create_sort_index() there is code that cleans up a
lot of stuff on the join tab. One of the things that is cleaned
is the select object. The result of this is that the
quick-select object created in make_join_select is deleted.
4. The second execution of the subquery does the same as the first but
the result is different:
a. Call create_sort_index() which again will call find_all_keys()
(same as for the first execution)
b. find_all_keys() will read the keys from the storage engine. To
do this it checks if it has a quick-select for the table. Now
there is NO quick-select object(!) (since it was deleted in
step 3c). So find_all_keys defaults to read the table using a
table scan instead. So instead of reading the four relevant records
in the range it reads the entire table (6 records). It then
evaluates the table's condition (and here it goes wrong). Since
the entire condition has been pushed down to the storage engine
using ICP all 6 records qualify. (Note that the storage engine
will not evaluate the pushed index condition in this case since
it was pushed for the k1 index and now we do a table scan
without any index being used).
The result is that here we return six qualifying key values
instead of four due to not evaluating the table's condition.
c. As above.
5. The two last execution of the subquery will also produce wrong results
for the same reason.
Summary: The problem occurs due to all but the first executions of the
subquery is done as a table scan without evaluating the table's
condition (which is pushed to the storage engine on a different
index). This is caused by the create_sort_index() function deleting
the quick-select object that should have been used for executing the
subquery as a range scan.
Note that this bug in addition to causing wrong results also can
result in bad performance due to executing the subquery using a table
scan instead of a range scan. This is an issue in MySQL 5.5.
The fix for this problem is to avoid that the Quick-select-object that
the optimizer created is deleted when create_sort_index() is doing
clean-up of the join-tab. This will ensure that the quick-select
object and the corresponding pushed index condition will be available
and used by all following executions of the subquery.
CONSISTENT SNAPSHOT OPTION
A transaction is started with a consistent snapshot. After
the transaction is started new indexes are added to the
table. Now when we issue an update statement, the optimizer
chooses an index. When the index scan is being initialized
via ha_innobase::change_active_index(), InnoDB reports
the error code HA_ERR_TABLE_DEF_CHANGED, with message
stating that "insufficient history for index".
This error message is propagated up to the SQL layer. But
the my_error() api is never called. The statement level
diagnostics area is not updated with the correct error
status (it remains in Diagnostics_area::DA_EMPTY).
Hence the following check in the Protocol::end_statement()
fails.
516 case Diagnostics_area::DA_EMPTY:
517 default:
518 DBUG_ASSERT(0);
519 error= send_ok(thd->server_status, 0, 0, 0, NULL);
520 break;
The fix is to backport the fix of bugs 14365043, 11761652
and 11746399.
14365043 PROTOCOL::END_STATEMENT(): ASSERTION `0' FAILED
11761652 HA_RND_INIT() RESULT CODE NOT CHECKED
11746399 RETURN VALUES OF HA_INDEX_INIT() AND INDEX_INIT() IGNORED
rb://1227 approved by guilhem and mattiasj.
