It was:
implicit conversion from 'ha_rows' (aka 'unsigned long long') to 'double'
changes value from 18446744073709551615 to 18446744073709551616
Follow what JOIN::get_examined_rows() does for similar code.
The issue here is the wrong estimate of the cardinality of a partial join,
the cardinality is too high because the function table_cond_selectivity()
returns an absurd number 100 while selectivity cannot be greater than 1.
When accessing table t by outer reference t1.a via index we do not perform any
range analysis for t. Yet we see TABLE::quick_key_parts[key] and
TABLE->quick_rows[key] contain a non-zero value though these should have been
remained untouched and equal to 0.
Thus real cause of the problem is that TABLE::init does not clean the arrays
TABLE::quick_key_parts[] and TABLE::>quick_rows[].
It should have done it because the TABLE structure created for any
instance of a table can be reused for many queries.
In the function prev_record_reads where one finds the different row combinations for a
subset of partial join, it did not take into account the selectivity of tables
involved in the subset of partial join.
selectivity values fails
After having set the assertion that checks validity of selectivity values
returned by the function table_cond_selectivity() a test case from
order_by.tesst failed. The failure occurred because range optimizer could
return as an estimate of the cardinality of the ranges built for an index
a number exceeding the total number of records in the table.
The second bug is more subtle. It may happen when there are several
indexes with same prefix defined on the first joined table t accessed by
a constant ref access. In this case the range optimizer estimates the
number of accessed records of t for each usable index and these
estimates can be different. Only the first of these estimates is taken
into account when the selectivity of the ref access is calculated.
However the optimizer later can choose a different index that provides
a different estimate. The function table_condition_selectivity() could use
this estimate to discount the selectivity of the ref access. This could
lead to an selectivity value returned by this function that was greater
that 1.
When discounting selectivity of ref access, don't discount the
selectivity we've already discounted for range access.
The 10.1 version of the fix. Will need to adjust condition filtering
test results in 10.4
cmake -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DCMAKE_BUILD_TYPE=Debug
Maintainer mode makes all warnings errors. This patch fix warnings. Mostly about
deprecated `register` keyword.
Too much warnings came from Mroonga and I gave up on it.
and WHERE filter afterwards
This patch complements the patch fixing the bug MDEV-6892. The latter
properly handled queries that used mergeable views returning constant
columns as inner tables of outer joins and whose where clause contained
predicates referring to these columns if the predicates of happened not
to be equality predicates. Otherwise the server still could return wrong
result sets for such queries. Besides the fix for MDEV-6892 prevented
some possible conversions of outer joins to inner joins for such queries.
This patch corrected the function check_simple_equality() to handle
properly conjunctive equalities of the where clause that refer to the
constant columns of mergeable views used as inner tables of an outer join.
The patch also changed the code of Item_direct_view_ref::not_null_tables().
This change allowed to take into account predicates containing references
to constant columns of mergeable views when converting outer joins into
inner joins.
Handling of top level conjuncts in WHERE whose used_tables() contained
RAND_TABLE_BIT in the function make_join_select() was incorrect.
As a result if such a conjunct referred to fields non of which belonged
to the last joined table it was pushed twice. (This could be seen
for a test case from subselect.test whose output was changed after this
patch had been applied. In 10.1 when running EXPLAIN FORMAT=JSON for
the query from this test case we clearly see that one of the conjuncts
is pushed twice.) This fact by itself was not good. Besides, if such a
conjunct was pushed to a table that was the result of materialization
of a semi-join the query could return a wrong result set. In particular
we could watch it for queries with semi-join subqueries whose left parts
used stored functions without "deterministic' specifier.
This patch complements the patch that fixes bug MDEV-18479.
This patch takes care of possible overflow when calculating the
estimated number of rows in a materialized derived table / view.
or server crashes in JOIN::fix_all_splittings_in_plan after EXPLAIN
This patch resolves the problem of overflowing when performing
calculations to estimate the cost of an evaluated query execution plan.
The overflowing in a non-debug build could cause different kind of
problems uncluding crashes of the server.
This patch corrects the patch for the bug 10006. The latter incorrectly
calculates the attribute TABLE_LIST::dep_tables for inner tables
of outer joins that are to be converted into inner joins.
As a result after the patch some valid join orders were not evaluated
and the optimizer could choose an execution plan that was far from
being optimal.
The code in best_access_path function, when it does not find a key suitable for ref access
and join_cache_level is set to a value so that hash_join is possible we build a hash key.
Later in the function we compare the cost of ref access with table scan (or index scan
or quick selects). No need to do this when we have got the hash key.
There were two newly enabled warnings:
1. cast for a function pointers. Affected sql_analyse.h, mi_write.c
and ma_write.cc, mf_iocache-t.cc, mysqlbinlog.cc, encryption.cc, etc
2. memcpy/memset of nontrivial structures. Fixed as:
* the warning disabled for InnoDB
* TABLE, TABLE_SHARE, and TABLE_LIST got a new method reset() which
does the bzero(), which is safe for these classes, but any other
bzero() will still cause a warning
* Table_scope_and_contents_source_st uses `TABLE_LIST *` (trivial)
instead of `SQL_I_List<TABLE_LIST>` (not trivial) so it's safe to
bzero now.
* added casts in debug_sync.cc and sql_select.cc (for JOIN)
* move assignment method for MDL_request instead of memcpy()
* PARTIAL_INDEX_INTERSECT_INFO::init() instead of bzero()
* remove constructor from READ_RECORD() to make it trivial
* replace some memcpy() with c++ copy assignments
derived table / view by equality
Now rows of a materialized derived table are always put into a
temporary table before join operation. If BNLH is used to join this
table with the result of a partial join then both operands of the
join are actually put into main memory. In most cases this is not
efficient.
We could avoid this by sending the rows of the derived table directly
to the join operation. However this kind of data flow is not supported
yet.
Fixed by not allowing usage of hash join algorithm to join a materialized
derived table if it's joined by an equality predicate of the form
f=e where f is a field of the derived table.
derived table / view by equality
Now rows of a materialized derived table are always put into a
temporary table before join operation. If BNLH is used to join this
table with the result of a partial join then both operands of the
join are actually put into main memory. In most cases this is not
efficient.
We could avoid this by sending the rows of the derived table directly
to the join operation. However this kind of data flow is not supported
yet.
Fixed by not allowing usage of hash join algorithm to join a materialized
derived table if it's joined by an equality predicate of the form
f=e where f is a field of the derived table.
derived table / view by equality
Now rows of a materialized derived table are always put into a
temporary table before join operation. If BNLH is used to join this
table with the result of a partial join then both operands of the
join are actually put into main memory. In most cases this is not
efficient.
We could avoid this by sending the rows of the derived table directly
to the join operation. However this kind of data flow is not supported
yet.
Fixed by not allowing usage of hash join algorithm to join a materialized
derived table if it's joined by an equality predicate of the form
f=e where f is a field of the derived table.
The bug was this scenario:
1. Join optimizer picks a range plan on index IDX1
(This index doesn't match the ORDER BY clause, so sorting will be needed)
2. Index Condition Pushdown pushes a part of WHERE down. The pushed
condition is removed from SQL_SELECT::cond
3. test_if_skip_sort_order() figures that it's better to use IDX2
(as it will match ORDER BY ... LIMIT and so will execute faster)
3.1 It sees that there was a possible range access on IDX2. It tries to
construct it by calling SQL_SELECT::test_quick_select(), but alas,
SQL_SELECT::cond doesn't have all parts of WHERE anymore.
So it uses full index scan which is slow.
(The execution works fine because there's code further in test_if_skip_sort_order()
which "Unpushes" the index condition and restores the original WHERE clause.
It was just the test_quick_select call that suffered).
After the commit b76b69cd5fe634d8ddb9406aa2c82ef2a375b4d8
loose index scan for queries with DISTINCT stopped working.
That is why that commit has to be reverted.
Additionally this patch fixes the problem of MDEV-10880.
with join_cache_level>2
During muliple equality propagation for a query in which we have an IN subquery, the items in the select list of the
subquery may not be part of the multiple equality because there might be another occurence of the same field in the
where clause of the subquery.
So we keyuse_is_valid_for_access_in_chosen_plan function which expects the items in the select list of the subquery to
be same to the ones in the multiple equality (through these multiple equalities we create keyuse array).
The solution would be that we expect the same field not the same Item because when we have SEMI JOIN MATERIALIZATION SCAN,
we use copy back technique to copies back the materialised table fields to the original fields of the base tables.
This patch fixes another problem introduced by the patch for mdev-4817.
The latter changed Item_cond::fix_fields() in such a way that it could
call the virtual method is_expensive(). With the first its call
the method saves the result in Item::is_expensive_cache. For all next
calls the method returns the result from this cache. So if the item
once was determined as expensive the method always returns true.
For subqueries it's not good, because non-optimized subqueries always
is considered as expensive.
It means that the cache should be invalidated after the call of
optimize_constant_subqueries().
This patch fixes another problem introduced by the patch for mdev-4817.
The latter changed Item_cond::fix_fields() in such a way that it could
call the virtual method is_expensive(). With the first its call
the method saves the result in Item::is_expensive_cache. For all next
calls the method returns the result from this cache. So if the item
once was determined as expensive the method always returns true.
For subqueries it's not good, because non-optimized subqueries always
is considered as expensive.
It means that the cache should be invalidated after the call of
optimize_constant_subqueries().
This patch fixes another problem introduced by the patch for mdev-4817.
The latter changed Item_cond::fix_fields() in such a way that it could
call the virtual method is_expensive(). With the first its call
the method saves the result in Item::is_expensive_cache. For all next
calls the method returns the result from this cache. So if the item
once was determined as expensive the method always returns true.
For subqueries it's not good, because non-optimized subqueries always
is considered as expensive.
It means that the cache should be invalidated after the call of
optimize_constant_subqueries().
Due to a legacy bug in the code of make_join_statistics() detecting
so-called constant tables could miss some of them in rare queries
that used RIGHT JOIN. As a result these queries had execution plans
different from the execution plans of the equivalent queries with
LEFT JOIN.
Besides starting from 10.2 this could trigger an assertion failure.
