The problem was caused by a merged semi-join, which contained a non-merged
semi-join, which used references to the top-level query in the left_expr.
When moving non-merged semi-join from the subquery to its parent, do not
forget to call fix_after_pullout for its Item_subselect. We need to do
that specifically, because non-merged semi-joins do not have their
IN-equality in the WHERE clause at this stage.
Alternative fix that doesn't cause view.test crash in --ps:
Remember when Item_ref was fixed right in the constructor
and did not have a full Item_ref::fix_fields() call. Later
in PS/SP, after Item_ref::cleanup, we use this knowledge
to avoid doing full fix_fields() for items that were never
supposed to be fix_field'ed.
Simplify the test case.
[Attempt #] Make the code that handles "Prepare" phase for multi-table
UPDATE statements handle non-merged semijoins. It can encounter them when
a prepared statement is executed for the second time.
SELECT ... WHERE XX IN (SELECT YY)
this was transformed to something like:
SELECT ... WHERE IF_EXISTS(SELECT ... HAVING XX=YY)
The bug was that for normal execution XX was fixed in the original outer SELECT context while in PS it was fixed in the sub query context and this confused the optimizer.
Fixed by ensuring that XX is always fixed in the outer context.
The problem was caused by the following scenario:
- range optimizer picks an index IDX1 which doesn't match the ORDER BY ...
LIMIT clause.
- test_if_skip_sort_order() decides to switch to index IDX2 which matches
the ORDER BY ... LIMIT.
- it runs SQL_SELECT::test_quick_select() for the second time to produce
an quick select for IDX2.
- However, test_quick_select() would figure that full index scan on IDX1
is still cheaper (its calculations ignore the LIMIT n).
Fixed this by
- passing force_quick_range=true to test_quick_select()
- in test_quick_select, don't consider full index scans if the mentioned
parameter is true.
Numerous changes in .result files are caused by test_quick_select() being
run after "early/late NULLs filtering" feature has injected NOT NULL
condition.
- When traversing JOIN_TABs with first_linear_tab/next_linear_tab(), don't forget
to enter the semi-join nest when it is the first table in the join order.
Failure to do so could cause e.g. I_S tables not to be filled.
- With big_tables=ON, materialized table will use Aria (or MyISAM) SE, which
allows prefix key reads. However, the temp.table has rec_per_key=NULL which
causes the optimizer to crash when attempting to read index statistics for a
prefix index read.
- Fixed by providing a rec_per_key array with zeros (i.e. "no statistics data")
- convert_subq_to_sj() must connect child select's tables into
parent select's TABLE_LIST::next_local chain.
- The problem was that it took child's leaf_tables.head() which
is different. This could cause certain tables (in this bug's case,
child select's non-merged semi-join) not to be present in
TABLE_LIST::next_local chain. Which would cause non-merged semi-join
not to be initialized in setup_tables(), which would lead to
NULL pointer dereference.
- convert_subq_to_sj() must connect child select's tables into
parent select's TABLE_LIST::next_local chain.
- The problem was that it took child's leaf_tables.head() which
is different. This could cause certain tables (in this bug's case,
child select's non-merged semi-join) not to be present in
TABLE_LIST::next_local chain. Which would cause non-merged semi-join
not to be initialized in setup_tables(), which would lead to
NULL pointer dereference.
Apply fix suggested by Igor:
- When eliminate_item_equal() generates pair-wise equalities from a
multi-equality, do generate a "bridge" equality between the first
field inside SJM nest and the field that's first in the overall multi-equality.
BNL and BNLH joins pre-filter the records from a joined table via JOIN_TAB::cache_select->cond.
There is no need to re-evaluate the same conditions via JOIN_TAB::select_cond. This patch removes
the duplicated conditions from the top-level conjuncts of each pushed condition.
The added "Using where" in few EXPLAINs is due to taking into account tab->cache_select->cond
in addition to tab->select_cond in JOIN::save_explain_data_intern.
- make multi_delete::initialize_tables() take into account that the JOIN structure may have
semi-join nests (which are not fully initialized when this function is called, they have
tab->table=NULL which caused the crash)
- Also checked multi_update::initialize_tables(): it has a different logic and needed no fixing.
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.
Analysis:
When the method JOIN::choose_subquery_plan() decided to apply
the IN-TO-EXISTS strategy, it set the unit and select_lex
uncacheable flag to UNCACHEABLE_DEPENDENT_INJECTED unconditionally.
As result, even if IN-TO-EXISTS injected non-correlated predicates,
the subquery was still treated as correlated.
Solution:
Set the subquery as correlated only if the injected predicate(s) depend
on the outer query.
Analysis:
The fix for lp:944706 introduces early subquery optimization.
While a subquery is being optimized some of its predicates may be
removed. In the test case, the EXISTS subquery is constant, and is
evaluated to TRUE. As a result the whole OR is TRUE, and thus the
correlated condition "b = alias1.b" is optimized away. The subquery
becomes non-correlated.
The subquery cache is designed to work only for correlated subqueries.
If constant subquery optimization is disallowed, then the constant
subquery is not evaluated, the subquery remains correlated, and its
execution is cached. As a result execution is fast.
However, when the constant subquery was optimized away, it was neither
cached by the subquery cache, nor it was cached by the internal subquery
caching. The latter was due to the fact that the subquery still appeared
as correlated to the subselect_XYZ_engine::exec methods, and they
re-executed the subquery on each call to Item_subselect::exec.
Solution:
The solution is to update the correlated status of the subquery after it has
been optimized. This status consists of:
- st_select_lex::is_correlated
- Item_subselect::is_correlated
- SELECT_LEX::uncacheable
- SELECT_LEX_UNIT::uncacheable
The status is updated by st_select_lex::update_correlated_cache(), and its
caller st_select_lex::optimize_unflattened_subqueries. The solution relies
on the fact that the optimizer already called
st_select_lex::update_used_tables() for each subquery. This allows to
efficiently update the correlated status of each subquery without walking
the whole subquery tree.
Notice that his patch is an improvement over MySQL 5.6 and older, where
subqueries are not pre-optimized, and the above analysis is not possible.
- make make_cond_after_sjm() correctly handle OR clauses where one branch refers to the semi-join table
while the other branch refers to the non-semijoin table.
The patch enables back constant subquery execution during
query optimization after it was disabled during the development
of MWL#89 (cost-based choice of IN-TO-EXISTS vs MATERIALIZATION).
The main idea is that constant subqueries are allowed to be executed
during optimization if their execution is not expensive.
The approach is as follows:
- Constant subqueries are recursively optimized in the beginning of
JOIN::optimize of the outer query. This is done by the new method
JOIN::optimize_constant_subqueries(). This is done so that the cost
of executing these queries can be estimated.
- Optimization of the outer query proceeds normally. During this phase
the optimizer may request execution of non-expensive constant subqueries.
Each place where the optimizer may potentially execute an expensive
expression is guarded with the predicate Item::is_expensive().
- The implementation of Item_subselect::is_expensive has been extended
to use the number of examined rows (estimated by the optimizer) as a
way to determine whether the subquery is expensive or not.
- The new system variable "expensive_subquery_limit" controls how many
examined rows are considered to be not expensive. The default is 100.
In addition, multiple changes were needed to make this solution work
in the light of the changes made by MWL#89. These changes were needed
to fix various crashes and wrong results, and legacy bugs discovered
during development.
- When doing join optimization, pre-sort the tables so that they mimic the execution
order we've had with 'semijoin=off'.
- That way, we will not get regressions when there are two query plans (the old and the
new) that have indentical costs but different execution times (because of factors that
the optimizer was not able to take into account).
