MySQL 8.0.32
Source Code Documentation
sql_optimizer.h
Go to the documentation of this file.
1#ifndef SQL_OPTIMIZER_INCLUDED
2#define SQL_OPTIMIZER_INCLUDED
3
4/* Copyright (c) 2000, 2022, Oracle and/or its affiliates.
5
6 This program is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License, version 2.0,
8 as published by the Free Software Foundation.
9
10 This program is also distributed with certain software (including
11 but not limited to OpenSSL) that is licensed under separate terms,
12 as designated in a particular file or component or in included license
13 documentation. The authors of MySQL hereby grant you an additional
14 permission to link the program and your derivative works with the
15 separately licensed software that they have included with MySQL.
16
17 This program is distributed in the hope that it will be useful,
18 but WITHOUT ANY WARRANTY; without even the implied warranty of
19 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
20 GNU General Public License, version 2.0, for more details.
21
22 You should have received a copy of the GNU General Public License
23 along with this program; if not, write to the Free Software
24 Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA */
25
26/**
27 @file sql/sql_optimizer.h
28 Classes used for query optimizations.
29*/
30
31#include <sys/types.h>
32
33#include <cstring>
34#include <memory>
35#include <utility>
36
37#include "field_types.h"
38#include "my_alloc.h"
39#include "my_base.h"
40#include "my_dbug.h"
41#include "my_table_map.h"
42#include "sql/field.h"
43#include "sql/item.h"
45#include "sql/mem_root_array.h"
46#include "sql/opt_explain_format.h" // Explain_sort_clause
47#include "sql/sql_executor.h"
48#include "sql/sql_lex.h"
49#include "sql/sql_list.h"
51#include "sql/sql_select.h" // Key_use
52#include "sql/table.h"
54
55enum class Subquery_strategy : int;
56class COND_EQUAL;
57class Item_subselect;
58class Item_sum;
60class THD;
61class Window;
62struct AccessPath;
63struct MYSQL_LOCK;
64
65class Item_equal;
66template <class T>
67class mem_root_deque;
68
69// Key_use has a trivial destructor, no need to run it from Mem_root_array.
71
73
74/*
75 This structure is used to collect info on potentially sargable
76 predicates in order to check whether they become sargable after
77 reading const tables.
78 We form a bitmap of indexes that can be used for sargable predicates.
79 Only such indexes are involved in range analysis.
80*/
81
83 Field *field; /* field against which to check sargability */
84 Item **arg_value; /* values of potential keys for lookups */
85 uint num_values; /* number of values in the above array */
86};
87
88/**
89 Wrapper for ORDER* pointer to trace origins of ORDER list
90
91 As far as ORDER is just a head object of ORDER expression
92 chain, we need some wrapper object to associate flags with
93 the whole ORDER list.
94*/
96 public:
97 ORDER *order; ///< ORDER expression that we are wrapping with this class
98 Explain_sort_clause src; ///< origin of order list
99
100 private:
101 int flags; ///< bitmap of Explain_sort_property
102
103 public:
105
107 : order(order_arg),
108 src(src_arg),
109 flags(order_arg ? ESP_EXISTS : ESP_none) {}
110
111 bool empty() const { return order == nullptr; }
112
113 void clean() {
114 order = nullptr;
115 src = ESC_none;
116 flags = ESP_none;
117 }
118
119 int get_flags() const {
120 assert(order);
121 return flags;
122 }
123};
124
125class JOIN {
126 public:
127 JOIN(THD *thd_arg, Query_block *select);
128 JOIN(const JOIN &rhs) = delete;
129 JOIN &operator=(const JOIN &rhs) = delete;
130
131 /// Query expression referring this query block
134 }
135
136 /// Query block that is optimized and executed using this JOIN
138 /// Thread handler
139 THD *const thd;
140
141 /**
142 Optimal query execution plan. Initialized with a tentative plan in
143 JOIN::make_join_plan() and later replaced with the optimal plan in
144 get_best_combination().
145 */
147 /// Array of QEP_TABs
148 QEP_TAB *qep_tab{nullptr};
149
150 /**
151 Array of plan operators representing the current (partial) best
152 plan. The array is allocated in JOIN::make_join_plan() and is valid only
153 inside this function. Initially (*best_ref[i]) == join_tab[i].
154 The optimizer reorders best_ref.
155 */
156 JOIN_TAB **best_ref{nullptr};
157 /// mapping between table indexes and JOIN_TABs
158 JOIN_TAB **map2table{nullptr};
159 /*
160 The table which has an index that allows to produce the required ordering.
161 A special value of 0x1 means that the ordering will be produced by
162 passing 1st non-const table to filesort(). NULL means no such table exists.
163 */
165
166 // Temporary tables that need to be cleaned up after the query.
167 // Only used for the hypergraph optimizer; the non-hypergraph optimizer
168 // uses QEP_TABs to hold the list of tables (including temporary tables).
171
172 // Allocated on the MEM_ROOT, but can hold some objects
173 // that allocate on the heap and thus need destruction.
175 };
178
179 // Similarly, filesorts that need to be cleaned up after the query.
180 // Only used for the hypergraph optimizer, for the same reason as above.
182
183 /**
184 Before plan has been created, "tables" denote number of input tables in the
185 query block and "primary_tables" is equal to "tables".
186 After plan has been created (after JOIN::get_best_combination()),
187 the JOIN_TAB objects are enumerated as follows:
188 - "tables" gives the total number of allocated JOIN_TAB objects
189 - "primary_tables" gives the number of input tables, including
190 materialized temporary tables from semi-join operation.
191 - "const_tables" are those tables among primary_tables that are detected
192 to be constant.
193 - "tmp_tables" is 0, 1 or 2 (more if windows) and counts the maximum
194 possible number of intermediate tables in post-processing (ie sorting and
195 duplicate removal).
196 Later, tmp_tables will be adjusted to the correct number of
197 intermediate tables, @see JOIN::make_tmp_tables_info.
198 - The remaining tables (ie. tables - primary_tables - tmp_tables) are
199 input tables to materialized semi-join operations.
200 The tables are ordered as follows in the join_tab array:
201 1. const primary table
202 2. non-const primary tables
203 3. intermediate sort/group tables
204 4. possible holes in array
205 5. semi-joined tables used with materialization strategy
206 */
207 uint tables{0}; ///< Total number of tables in query block
208 uint primary_tables{0}; ///< Number of primary input tables in query block
209 uint const_tables{0}; ///< Number of primary tables deemed constant
210 uint tmp_tables{0}; ///< Number of temporary tables used by query
212 /**
213 Indicates that the data will be aggregated (typically GROUP BY),
214 _and_ that it is already processed in an order that is compatible with
215 the grouping in use (e.g. because we are scanning along an index,
216 or because an earlier step sorted the data in a group-compatible order).
217
218 Note that this flag changes value at multiple points during optimization;
219 if it's set when a temporary table is created, this means we aggregate
220 into said temporary table (end_write_group is chosen instead of end_write),
221 but if it's set later, it means that we can aggregate as we go,
222 just before sending the data to the client (end_send_group is chosen
223 instead of end_send).
224
225 @see make_group_fields, alloc_group_fields, JOIN::exec
226 */
228 /// If query contains GROUP BY clause
230 /// If true, send produced rows using query_result
231 bool do_send_rows{true};
232 /// Set of tables contained in query
234 table_map const_table_map; ///< Set of tables found to be const
235 /**
236 Const tables which are either:
237 - not empty
238 - empty but inner to a LEFT JOIN, thus "considered" not empty for the
239 rest of execution (a NULL-complemented row will be used).
240 */
242 /**
243 This is the bitmap of all tables which are dependencies of
244 lateral derived tables which are not (yet) part of the partial
245 plan. (The value is a logical 'or' of zero or more
246 Table_ref.map() values.)
247
248 When we are building the join order, there is a partial plan (an
249 ordered sequence of JOIN_TABs), and an unordered set of JOIN_TABs
250 not yet added to the plan. Due to backtracking, the partial plan
251 may both grow and shrink. When we add a new table to the plan, we
252 may wish to set up join buffering, so that rows from the preceding
253 table are buffered. If any of the remaining tables are derived
254 tables that depends on any of the predecessors of the table we
255 are adding (i.e. a lateral dependency), join buffering would be
256 inefficient. (@see setup_join_buffering() for a detailed
257 explanation of why this is so.)
258
259 For this reason we need to maintain this table_map of lateral
260 dependencies of tables not yet in the plan. Whenever we add a new
261 table to the plan, we update the map by calling
262 Optimize_table_order::recalculate_lateral_deps_incrementally().
263 And when we remove a table, we restore the previous map value
264 using a Tabel_map_restorer object.
265
266 As an example, assume that we join four tables, t1, t2, t3 and
267 d1, where d1 is a derived table that depends on t1:
268
269 SELECT * FROM t1 JOIN t2 ON t1.a=t2.b JOIN t3 ON t2.c=t3.d
270 JOIN LATERAL (SELECT DISTINCT e AS x FROM t4 WHERE t4.f=t1.c)
271 AS d1 ON t3.e=d1.x;
272
273 Now, if our partial plan is t1->t2, the map (of lateral
274 dependencies of the remaining tables) will contain t1.
275 This tells us that we should not use join buffering when joining t1
276 with t2. But if the partial plan is t1->d2->t2, the map will be
277 empty. We may thus use join buffering when joining d2 with t2.
278 */
280
281 /* Number of records produced after join + group operation */
286 // m_select_limit is used to decide if we are likely to scan the whole table.
288 /**
289 Used to fetch no more than given amount of rows per one
290 fetch operation of server side cursor.
291 The value is checked in end_send and end_send_group in fashion, similar
292 to offset_limit_cnt:
293 - fetch_limit= HA_POS_ERROR if there is no cursor.
294 - when we open a cursor, we set fetch_limit to 0,
295 - on each fetch iteration we add num_rows to fetch to fetch_limit
296 */
298
299 /**
300 This is the result of join optimization.
301
302 @note This is a scratch array, not used after get_best_combination().
303 */
305
306 /******* Join optimization state members start *******/
307
308 /* Current join optimization state */
310
311 /* We also maintain a stack of join optimization states in * join->positions[]
312 */
313 /******* Join optimization state members end *******/
314
315 /// A hook that secondary storage engines can use to override the executor
316 /// completely.
319
320 /**
321 The cost of best complete join plan found so far during optimization,
322 after optimization phase - cost of picked join order (not taking into
323 account the changes made by test_if_skip_sort_order()).
324 */
325 double best_read{0.0};
326 /**
327 The estimated row count of the plan with best read time (see above).
328 */
330 /// Expected cost of filesort.
331 double sort_cost{0.0};
332 /// Expected cost of windowing;
333 double windowing_cost{0.0};
337
338 // For destroying fields otherwise owned by RemoveDuplicatesIterator.
340
341 Item_sum **sum_funcs{nullptr};
342 /**
343 Describes a temporary table.
344 Each tmp table has its own tmp_table_param.
345 The one here is transiently used as a model by create_intermediate_table(),
346 to build the tmp table's own tmp_table_param.
347 */
350
351 enum class RollupState { NONE, INITED, READY };
353 bool implicit_grouping; ///< True if aggregated but no GROUP BY
354
355 /**
356 At construction time, set if SELECT DISTINCT. May be reset to false
357 later, when we set up a temporary table operation that deduplicates for us.
358 */
360
361 /**
362 If we have the GROUP BY statement in the query,
363 but the group_list was emptied by optimizer, this
364 flag is true.
365 It happens when fields in the GROUP BY are from
366 constant table
367 */
369
370 /*
371 simple_xxxxx is set if ORDER/GROUP BY doesn't include any references
372 to other tables than the first non-constant table in the JOIN.
373 It's also set if ORDER/GROUP BY is empty.
374 Used for deciding for or against using a temporary table to compute
375 GROUP/ORDER BY.
376 */
377 bool simple_order{false};
378 bool simple_group{false};
379
380 /*
381 m_ordered_index_usage is set if an ordered index access
382 should be used instead of a filesort when computing
383 ORDER/GROUP BY.
384 */
385 enum {
386 ORDERED_INDEX_VOID, // No ordered index avail.
387 ORDERED_INDEX_GROUP_BY, // Use index for GROUP BY
388 ORDERED_INDEX_ORDER_BY // Use index for ORDER BY
389 } m_ordered_index_usage{ORDERED_INDEX_VOID};
390
391 /**
392 Is set if we have a GROUP BY and we have ORDER BY on a constant or when
393 sorting isn't required.
394 */
395 bool skip_sort_order{false};
396
397 /**
398 If true we need a temporary table on the result set before any
399 windowing steps, e.g. for DISTINCT or we have a query ORDER BY.
400 See details in JOIN::optimize
401 */
403
404 /// If JOIN has lateral derived tables (is set at start of planning)
405 bool has_lateral{false};
406
407 /// Used and updated by JOIN::make_join_plan() and optimize_keyuse()
409
410 /**
411 Array of pointers to lists of expressions.
412 Each list represents the SELECT list at a certain stage of execution,
413 and also contains necessary extras: expressions added for ORDER BY,
414 GROUP BY, window clauses, underlying items of split items.
415 This array is only used when the query makes use of tmp tables: after
416 writing to tmp table (e.g. for GROUP BY), if this write also does a
417 function's calculation (e.g. of SUM), after the write the function's value
418 is in a column of the tmp table. If a SELECT list expression is the SUM,
419 and we now want to read that materialized SUM and send it forward, a new
420 expression (Item_field type instead of Item_sum), is needed. The new
421 expressions are listed in JOIN::tmp_fields_list[x]; 'x' is a number
422 (REF_SLICE_).
423 @see JOIN::make_tmp_tables_info()
424 */
426
427 int error{0}; ///< set in optimize(), exec(), prepare_result()
428
429 /**
430 Incremented each time clear_hash_tables() is run, signaling to
431 HashJoinIterators that they cannot keep their hash tables anymore
432 (since outer references may have changed).
433 */
435
436 /**
437 ORDER BY and GROUP BY lists, to transform with prepare,optimize and exec
438 */
440
441 // Used so that AggregateIterator knows which items to signal when the rollup
442 // level changes. Obviously only used in the presence of rollup.
447
448 /**
449 Any window definitions
450 */
452
453 /**
454 True if a window requires a certain order of rows, which implies that any
455 order of rows coming out of the pre-window join will be disturbed.
456 */
457 bool m_windows_sort{false};
458
459 /// If we have set up tmp tables for windowing, @see make_tmp_tables_info
460 bool m_windowing_steps{false};
461
462 /**
463 Buffer to gather GROUP BY, ORDER BY and DISTINCT QEP details for EXPLAIN
464 */
466
467 /**
468 JOIN::having_cond is initially equal to query_block->having_cond, but may
469 later be changed by optimizations performed by JOIN.
470 The relationship between the JOIN::having_cond condition and the
471 associated variable query_block->having_value is so that
472 having_value can be:
473 - COND_UNDEF if a having clause was not specified in the query or
474 if it has not been optimized yet
475 - COND_TRUE if the having clause is always true, in which case
476 JOIN::having_cond is set to NULL.
477 - COND_FALSE if the having clause is impossible, in which case
478 JOIN::having_cond is set to NULL
479 - COND_OK otherwise, meaning that the having clause needs to be
480 further evaluated
481 All of the above also applies to the where_cond/query_block->cond_value
482 pair.
483 */
484 /**
485 Optimized WHERE clause item tree (valid for one single execution).
486 Used in JOIN execution if no tables. Otherwise, attached in pieces to
487 JOIN_TABs and then not used in JOIN execution.
488 Printed by EXPLAIN EXTENDED.
489 Initialized by Query_block::get_optimizable_conditions().
490 */
492 /**
493 Optimized HAVING clause item tree (valid for one single execution).
494 Used in JOIN execution, as last "row filtering" step. With one exception:
495 may be pushed to the JOIN_TABs of temporary tables used in DISTINCT /
496 GROUP BY (see JOIN::make_tmp_tables_info()); in that case having_cond is
497 set to NULL, but is first saved to having_for_explain so that EXPLAIN
498 EXTENDED can still print it.
499 Initialized by Query_block::get_optimizable_conditions().
500 */
502 Item *having_for_explain; ///< Saved optimized HAVING for EXPLAIN
503 /**
504 Pointer set to query_block->get_table_list() at the start of
505 optimization. May be changed (to NULL) only if optimize_aggregated_query()
506 optimizes tables away.
507 */
510 /*
511 Join tab to return to. Points to an element of join->join_tab array, or to
512 join->join_tab[-1].
513 This is used at execution stage to shortcut join enumeration. Currently
514 shortcutting is done to handle outer joins or handle semi-joins with
515 FirstMatch strategy.
516 */
518
519 /**
520 ref_items is an array of 4+ slices, each containing an array of Item
521 pointers. ref_items is used in different phases of query execution.
522 - slice 0 is initially the same as Query_block::base_ref_items, ie it is
523 the set of items referencing fields from base tables. During optimization
524 and execution it may be temporarily overwritten by slice 1-3.
525 - slice 1 is a representation of the used items when being read from
526 the first temporary table.
527 - slice 2 is a representation of the used items when being read from
528 the second temporary table.
529 - slice 3 is a copy of the original slice 0. It is created if
530 slice overwriting is necessary, and it is used to restore
531 original values in slice 0 after having been overwritten.
532 - slices 4 -> N are used by windowing: all the window's out tmp tables,
533
534 Two windows: 4: window 1's out table
535 5: window 2's out table
536
537 and so on.
538
539 Slice 0 is allocated for the lifetime of a statement, whereas slices 1-3
540 are associated with a single optimization. The size of slice 0 determines
541 the slice size used when allocating the other slices.
542 */
544 nullptr}; // cardinality: REF_SLICE_SAVED_BASE + 1 + #windows*2
545
546 /**
547 The slice currently stored in ref_items[0].
548 Used to restore the base ref_items slice from the "save" slice after it
549 has been overwritten by another slice (1-3).
550 */
552
553 /**
554 Used only if this query block is recursive. Contains count of
555 all executions of this recursive query block, since the last
556 this->reset().
557 */
559
560 /**
561 <> NULL if optimization has determined that execution will produce an
562 empty result before aggregation, contains a textual explanation on why
563 result is empty. Implicitly grouped queries may still produce an
564 aggregation row.
565 @todo - suggest to set to "Preparation determined that query is empty"
566 when Query_block::is_empty_query() is true.
567 */
568 const char *zero_result_cause{nullptr};
569
570 /**
571 True if, at this stage of processing, subquery materialization is allowed
572 for children subqueries of this JOIN (those in the SELECT list, in WHERE,
573 etc). If false, and we have to evaluate a subquery at this stage, then we
574 must choose EXISTS.
575 */
577 /**
578 True if plan search is allowed to use references to expressions outer to
579 this JOIN (for example may set up a 'ref' access looking up an outer
580 expression in the index, etc).
581 */
582 bool allow_outer_refs{false};
583
584 /* Temporary tables used to weed-out semi-join duplicates */
587 /* end of allocation caching storage */
588
589 /** Exec time only: true <=> current group has been sent */
590 bool group_sent{false};
591 /// If true, calculate found rows for this query block
592 bool calc_found_rows{false};
593
594 /**
595 This will force tmp table to NOT use index + update for group
596 operation as it'll cause [de]serialization for each json aggregated
597 value and is very ineffective (times worse).
598 Server should use filesort, or tmp table + filesort to resolve GROUP BY
599 with JSON aggregate functions.
600 */
602
603 /// True if plan is const, ie it will return zero or one rows.
604 bool plan_is_const() const { return const_tables == primary_tables; }
605
606 /**
607 True if plan contains one non-const primary table (ie not including
608 tables taking part in semi-join materialization).
609 */
611
612 /**
613 Returns true if any of the items in JOIN::fields contains a call to the
614 full-text search function MATCH, which is not wrapped in an aggregation
615 function.
616 */
617 bool contains_non_aggregated_fts() const;
618
619 bool optimize(bool finalize_access_paths);
620 void reset();
621 bool prepare_result();
622 void destroy();
623 bool alloc_func_list();
625 bool before_group_by, bool recompute = false);
626
627 /**
628 Overwrites one slice of ref_items with the contents of another slice.
629 In the normal case, dst and src have the same size().
630 However: the rollup slices may have smaller size than slice_sz.
631 */
632 void copy_ref_item_slice(uint dst_slice, uint src_slice) {
633 copy_ref_item_slice(ref_items[dst_slice], ref_items[src_slice]);
634 }
636 assert(dst_arr.size() >= src_arr.size());
637 void *dest = dst_arr.array();
638 const void *src = src_arr.array();
639 if (!src_arr.is_null())
640 memcpy(dest, src, src_arr.size() * src_arr.element_size());
641 }
642
643 /**
644 Allocate a ref_item slice, assume that slice size is in ref_items[0]
645
646 @param thd_arg thread handler
647 @param sliceno The slice number to allocate in JOIN::ref_items
648
649 @returns false if success, true if error
650 */
651 bool alloc_ref_item_slice(THD *thd_arg, int sliceno);
652
653 /**
654 Overwrite the base slice of ref_items with the slice supplied as argument.
655
656 @param sliceno number to overwrite the base slice with, must be 1-4 or
657 4 + windowno.
658 */
659 void set_ref_item_slice(uint sliceno) {
660 assert((int)sliceno >= 1);
661 if (current_ref_item_slice != sliceno) {
663 DBUG_PRINT("info", ("JOIN %p ref slice %u -> %u", this,
664 current_ref_item_slice, sliceno));
665 current_ref_item_slice = sliceno;
666 }
667 }
668
669 /// @note do also consider Switch_ref_item_slice
671
672 /**
673 Returns the clone of fields_list which is appropriate for evaluating
674 expressions at the current stage of execution; which stage is denoted by
675 the value of current_ref_item_slice.
676 */
678
679 bool optimize_rollup();
681 /**
682 Release memory and, if possible, the open tables held by this execution
683 plan (and nested plans). It's used to release some tables before
684 the end of execution in order to increase concurrency and reduce
685 memory consumption.
686 */
687 void join_free();
688 /** Cleanup this JOIN. Not a full cleanup. reusable? */
689 void cleanup();
690
691 bool clear_fields(table_map *save_nullinfo);
692 void restore_fields(table_map save_nullinfo);
693
694 private:
695 /**
696 Return whether the caller should send a row even if the join
697 produced no rows if:
698 - there is an aggregate function (sum_func_count!=0), and
699 - the query is not grouped, and
700 - a possible HAVING clause evaluates to TRUE.
701
702 @note: if there is a having clause, it must be evaluated before
703 returning the row.
704 */
709 }
710
711 public:
715 bool attach_join_conditions(plan_idx last_tab);
716
717 private:
718 bool attach_join_condition_to_nest(plan_idx first_inner, plan_idx last_tab,
719 Item *join_cond, bool is_sj_mat_cond);
720
721 public:
724 bool sort_before_group);
728 table_map plan_tables, uint idx) const;
729 bool clear_sj_tmp_tables();
732
733 void mark_const_table(JOIN_TAB *table, Key_use *key);
734 /// State of execution plan. Currently used only for EXPLAIN
736 NO_PLAN, ///< No plan is ready yet
737 ZERO_RESULT, ///< Zero result cause is set
738 NO_TABLES, ///< Plan has no tables
739 PLAN_READY ///< Plan is ready
740 };
741 /// See enum_plan_state
743 bool is_optimized() const { return optimized; }
744 void set_optimized() { optimized = true; }
745 bool is_executed() const { return executed; }
746 void set_executed() { executed = true; }
747
748 /**
749 Retrieve the cost model object to be used for this join.
750
751 @return Cost model object for the join
752 */
753
754 const Cost_model_server *cost_model() const;
755
756 /**
757 Check if FTS index only access is possible
758 */
759 bool fts_index_access(JOIN_TAB *tab);
760
762 /**
763 Propagate dependencies between tables due to outer join relations.
764
765 @returns false if success, true if error
766 */
768
769 /**
770 Handle offloading of query parts to the underlying engines, when
771 such is supported by their implementation.
772
773 @returns false if success, true if error
774 */
775 bool push_to_engines();
776
779
780 /**
781 If this query block was planned twice, once with and once without conditions
782 added by in2exists, changes the root access path to the one without
783 in2exists. If not (ie., there were never any such conditions in the first
784 place), does nothing.
785 */
787
788 /**
789 In the case of rollup (only): After the base slice list was made, we may
790 have modified the field list to add rollup group items and sum switchers,
791 but there may be Items with refs that refer to the base slice. This function
792 refreshes the base slice (and its copy, REF_SLICE_SAVED_BASE) with a fresh
793 copy of the list from “fields”.
794
795 When we get rid of slices entirely, we can get rid of this, too.
796 */
797 void refresh_base_slice();
798
799 /**
800 Whether this query block needs finalization (see
801 FinalizePlanForQueryBlock()) before it can be actually used.
802 This only happens when using the hypergraph join optimizer.
803 */
804 bool needs_finalize{false};
805
806 private:
807 bool optimized{false}; ///< flag to avoid double optimization in EXPLAIN
808
809 /**
810 Set by exec(), reset by reset(). Note that this needs to be set
811 _during_ the query (not only when it's done executing), or the
812 dynamic range optimizer will not understand which tables have been
813 read.
814 */
815 bool executed{false};
816
817 /// Final execution plan state. Currently used only for EXPLAIN
819
820 public:
821 /*
822 When join->select_count is set, tables will not be optimized away.
823 The call to records() will be delayed until the execution phase and
824 the counting will be done on an index of Optimizer's choice.
825 The index will be decided in find_shortest_key(), called from
826 optimize_aggregated_query().
827 */
828 bool select_count{false};
829
830 private:
831 /**
832 Create a temporary table to be used for processing DISTINCT/ORDER
833 BY/GROUP BY.
834
835 @note Will modify JOIN object wrt sort/group attributes
836
837 @param tab the JOIN_TAB object to attach created table to
838 @param tmp_table_fields List of items that will be used to define
839 column types of the table.
840 @param tmp_table_group Group key to use for temporary table, empty if none.
841 @param save_sum_fields If true, do not replace Item_sum items in
842 @c tmp_fields list with Item_field items referring
843 to fields in temporary table.
844
845 @returns false on success, true on failure
846 */
848 const mem_root_deque<Item *> &tmp_table_fields,
849 ORDER_with_src &tmp_table_group,
850 bool save_sum_fields);
851
852 /**
853 Optimize distinct when used on a subset of the tables.
854
855 E.g.,: SELECT DISTINCT t1.a FROM t1,t2 WHERE t1.b=t2.b
856 In this case we can stop scanning t2 when we have found one t1.a
857 */
858 void optimize_distinct();
859
860 /**
861 Function sets FT hints, initializes FT handlers and
862 checks if FT index can be used as covered.
863 */
864 bool optimize_fts_query();
865
866 /**
867 Checks if the chosen plan suffers from a problem related to full-text search
868 and streaming aggregation, which is likely to cause wrong results or make
869 the query misbehave in other ways, and raises an error if so. Only to be
870 called for queries with full-text search and GROUP BY WITH ROLLUP.
871
872 If there are calls to MATCH in the SELECT list (including the hidden
873 elements lifted there from other clauses), and they are not inside an
874 aggregate function, the results of the MATCH clause need to be materialized
875 before streaming aggregation is performed. The hypergraph optimizer adds a
876 materialization step before aggregation if needed (see
877 CreateStreamingAggregationPath()), but the old optimizer only does that for
878 implicitly grouped queries. For explicitly grouped queries, it instead
879 disables streaming aggregation for the queries that would need a
880 materialization step to work correctly (see JOIN::test_skip_sort()).
881
882 For explicitly grouped queries WITH ROLLUP, however, streaming aggregation
883 is currently the only alternative. In many cases it still works correctly
884 because an intermediate materialization step has been added for some other
885 reason, typically for a sort. For now, in those cases where a
886 materialization step has not been added, we raise an error instead of going
887 ahead with an invalid execution plan.
888
889 @return true if an error was raised.
890 */
891 bool check_access_path_with_fts() const;
892
894 /**
895 Initialize key dependencies for join tables.
896
897 TODO figure out necessity of this method. Current test
898 suite passed without this initialization.
899 */
901 JOIN_TAB *const tab_end = join_tab + tables;
902 for (JOIN_TAB *tab = join_tab; tab < tab_end; tab++)
903 tab->key_dependent = tab->dependent;
904 }
905
906 private:
907 void set_prefix_tables();
908 void cleanup_item_list(const mem_root_deque<Item *> &items) const;
910 bool make_join_plan();
911 bool init_planner_arrays();
915 bool estimate_rowcount();
916 void optimize_keyuse();
917 void set_semijoin_info();
918 /**
919 An utility function - apply heuristics and optimize access methods to tables.
920 @note Side effect - this function could set 'Impossible WHERE' zero
921 result.
922 */
924 void update_depend_map();
926 /**
927 Fill in outer join related info for the execution plan structure.
928
929 For each outer join operation left after simplification of the
930 original query the function set up the following pointers in the linear
931 structure join->join_tab representing the selected execution plan.
932 The first inner table t0 for the operation is set to refer to the last
933 inner table tk through the field t0->last_inner.
934 Any inner table ti for the operation are set to refer to the first
935 inner table ti->first_inner.
936 The first inner table t0 for the operation is set to refer to the
937 first inner table of the embedding outer join operation, if there is any,
938 through the field t0->first_upper.
939 The on expression for the outer join operation is attached to the
940 corresponding first inner table through the field t0->on_expr_ref.
941 Here ti are structures of the JOIN_TAB type.
942
943 EXAMPLE. For the query:
944 @code
945 SELECT * FROM t1
946 LEFT JOIN
947 (t2, t3 LEFT JOIN t4 ON t3.a=t4.a)
948 ON (t1.a=t2.a AND t1.b=t3.b)
949 WHERE t1.c > 5,
950 @endcode
951
952 given the execution plan with the table order t1,t2,t3,t4
953 is selected, the following references will be set;
954 t4->last_inner=[t4], t4->first_inner=[t4], t4->first_upper=[t2]
955 t2->last_inner=[t4], t2->first_inner=t3->first_inner=[t2],
956 on expression (t1.a=t2.a AND t1.b=t3.b) will be attached to
957 *t2->on_expr_ref, while t3.a=t4.a will be attached to *t4->on_expr_ref.
958
959 @note
960 The function assumes that the simplification procedure has been
961 already applied to the join query (see simplify_joins).
962 This function can be called only after the execution plan
963 has been chosen.
964 */
965 void make_outerjoin_info();
966
967 /**
968 Initialize ref access for all tables that use it.
969
970 @return False if success, True if error
971
972 @note We cannot setup fields used for ref access before we have sorted
973 the items within multiple equalities according to the final order of
974 the tables involved in the join operation. Currently, this occurs in
975 @see substitute_for_best_equal_field().
976 */
977 bool init_ref_access();
978 bool alloc_qep(uint n);
979 void unplug_join_tabs();
981 POSITION *inner_pos,
982 POSITION *sjm_pos);
983
984 bool add_having_as_tmp_table_cond(uint curr_tmp_table);
986 void set_plan_state(enum_plan_state plan_state_arg);
988 ORDER *remove_const(ORDER *first_order, Item *cond, bool change_list,
989 bool *simple_order, bool group_by);
990
991 /**
992 Check whether this is a subquery that can be evaluated by index look-ups.
993 If so, change subquery engine to subselect_indexsubquery_engine.
994
995 @retval 1 engine was changed
996 @retval 0 engine wasn't changed
997 @retval -1 OOM or other error
998 */
1000
1001 /**
1002 Optimize DISTINCT, GROUP BY, ORDER BY clauses
1003
1004 @retval false ok
1005 @retval true an error occurred
1006 */
1008
1009 /**
1010 Test if an index could be used to replace filesort for ORDER BY/GROUP BY
1011
1012 @details
1013 Investigate whether we may use an ordered index as part of either
1014 DISTINCT, GROUP BY or ORDER BY execution. An ordered index may be
1015 used for only the first of any of these terms to be executed. This
1016 is reflected in the order which we check for test_if_skip_sort_order()
1017 below. However we do not check for DISTINCT here, as it would have
1018 been transformed to a GROUP BY at this stage if it is a candidate for
1019 ordered index optimization.
1020 If a decision was made to use an ordered index, the availability
1021 if such an access path is stored in 'm_ordered_index_usage' for later
1022 use by 'execute' or 'explain'
1023 */
1024 void test_skip_sort();
1025
1027
1028 /**
1029 Convert the executor structures to a set of access paths, storing
1030 the result in m_root_access_path.
1031 */
1032 void create_access_paths();
1033
1034 public:
1035 /**
1036 Create access paths with the knowledge that there are going to be zero rows
1037 coming from tables (before aggregation); typically because we know that
1038 all of them would be filtered away by WHERE (e.g. SELECT * FROM t1
1039 WHERE 1=2). This will normally yield no output rows, but if we have implicit
1040 aggregation, it might yield a single one.
1041 */
1043
1044 private:
1046
1047 /** @{ Helpers for create_access_paths. */
1051 /** @} */
1052
1053 /**
1054 An access path you can read from to get all records for this query
1055 (after you create an iterator from it).
1056 */
1058
1059 /**
1060 If this query block contains conditions synthesized during IN-to-EXISTS
1061 conversion: A second query plan with all such conditions removed.
1062 See comments in JOIN::optimize().
1063 */
1065};
1066
1067/**
1068 Use this in a function which depends on best_ref listing tables in the
1069 final join order. If 'tables==0', one is not expected to consult best_ref
1070 cells, and best_ref may not even have been allocated.
1071*/
1072#define ASSERT_BEST_REF_IN_JOIN_ORDER(join) \
1073 do { \
1074 assert((join)->tables == 0 || ((join)->best_ref && !(join)->join_tab)); \
1075 } while (0)
1076
1077/**
1078 RAII class to ease the temporary switching to a different slice of
1079 the ref item array.
1080*/
1084
1085 public:
1087 : join(join_arg), saved(join->get_ref_item_slice()) {
1088 if (!join->ref_items[new_v].is_null()) join->set_ref_item_slice(new_v);
1089 }
1091};
1092
1093bool uses_index_fields_only(Item *item, TABLE *tbl, uint keyno,
1094 bool other_tbls_ok);
1095bool remove_eq_conds(THD *thd, Item *cond, Item **retcond,
1096 Item::cond_result *cond_value);
1097bool optimize_cond(THD *thd, Item **conds, COND_EQUAL **cond_equal,
1098 mem_root_deque<Table_ref *> *join_list,
1099 Item::cond_result *cond_value);
1101 COND_EQUAL *cond_equal,
1102 JOIN_TAB **table_join_idx);
1103bool build_equal_items(THD *thd, Item *cond, Item **retcond,
1104 COND_EQUAL *inherited, bool do_inherit,
1105 mem_root_deque<Table_ref *> *join_list,
1106 COND_EQUAL **cond_equal_ref);
1110 THD *thd, uint keyparts, Item_field **fields,
1111 const mem_root_deque<Item *> &outer_exprs);
1112Item_field *get_best_field(Item_field *item_field, COND_EQUAL *cond_equal);
1113Item *make_cond_for_table(THD *thd, Item *cond, table_map tables,
1114 table_map used_table, bool exclude_expensive_cond);
1116 uint first_unused);
1117
1118/**
1119 Create an order list that consists of all non-const fields and items.
1120 This is usable for e.g. converting DISTINCT into GROUP or ORDER BY.
1121 Is ref_item_array is non-null (is_null() returns false), the items
1122 will point into the slice given by it. Otherwise, it points directly
1123 into *fields (this is the only reason why fields is not const).
1124
1125 Try to put the items in "order_list" first, to allow one to optimize away
1126 a later ORDER BY.
1127 */
1129 ORDER *order_list,
1130 mem_root_deque<Item *> *fields,
1131 bool skip_aggregates,
1132 bool convert_bit_fields_to_long,
1133 bool *all_order_by_fields_used);
1134
1135/**
1136 Returns true if arguments are a temporal Field having no date,
1137 part and a temporal expression having a date part.
1138 @param f Field
1139 @param v Expression
1140 */
1141inline bool field_time_cmp_date(const Field *f, const Item *v) {
1142 const enum_field_types ft = f->type();
1143 return is_temporal_type(ft) && !is_temporal_type_with_date(ft) &&
1145}
1146
1147bool substitute_gc(THD *thd, Query_block *query_block, Item *where_cond,
1148 ORDER *group_list, ORDER *order);
1149
1150/**
1151 This class restores a table_map object to its original value
1152 when '*this' is destroyed.
1153 */
1155 /** The location to be restored.*/
1157 /** The original value to restore.*/
1159
1160 public:
1161 /**
1162 Constructor.
1163 @param map The table map that we wish to restore.
1164 */
1167
1168 // This class is not intended to be copied.
1171
1174 void assert_unchanged() const { assert(*m_location == m_saved_value); }
1175};
1176
1177/**
1178 Estimates how many times a subquery will be executed as part of a
1179 query execution. If it is a cacheable subquery, the estimate tells
1180 how many times the subquery will be executed if it is not cached.
1181
1182 @param[in] subquery the Item that represents the subquery
1183 @param[in,out] trace optimizer trace context
1184
1185 @return the number of times the subquery is expected to be executed
1186*/
1187double calculate_subquery_executions(const Item_subselect *subquery,
1188 Opt_trace_context *trace);
1189
1190extern const char *antijoin_null_cond;
1191
1192/**
1193 Checks if an Item, which is constant for execution, can be evaluated during
1194 optimization. It cannot be evaluated if it contains a subquery and the
1195 OPTION_NO_SUBQUERY_DURING_OPTIMIZATION query option is active.
1196
1197 @param item the Item to check
1198 @param select the query block that contains the Item
1199 @return false if this Item contains a subquery and subqueries cannot be
1200 evaluated during optimization, or true otherwise
1201*/
1202bool evaluate_during_optimization(const Item *item, const Query_block *select);
1203
1204/**
1205 Find the multiple equality predicate containing a field.
1206
1207 The function retrieves the multiple equalities accessed through
1208 the cond_equal structure from current level and up looking for
1209 an equality containing a field. It stops retrieval as soon as the equality
1210 is found and set up inherited_fl to true if it's found on upper levels.
1211
1212 @param cond_equal multiple equalities to search in
1213 @param item_field field to look for
1214 @param[out] inherited_fl set up to true if multiple equality is found
1215 on upper levels (not on current level of
1216 cond_equal)
1217
1218 @return
1219 - Item_equal for the found multiple equality predicate if a success;
1220 - nullptr otherwise.
1221*/
1223 const Item_field *item_field, bool *inherited_fl);
1224
1225/**
1226 Find an artificial cap for ref access. This is mostly a crutch to mitigate
1227 that we don't estimate the cache effects of ref accesses properly
1228 (ie., normally, if we do many, they will hit cache instead of being
1229 separate seeks). Given to find_cost_for_ref().
1230 */
1231double find_worst_seeks(const TABLE *table, double num_rows,
1232 double table_scan_cost);
1233
1234/**
1235 Whether a ref lookup of “right_item” on “field” will give an exact
1236 comparison in all cases, ie., one can remove any further checks on
1237 field = right_item. If not, there may be false positives, and one
1238 needs to keep the comparison after the ref lookup.
1239
1240 @param thd thread handler
1241 @param field field that is looked up through an index
1242 @param right_item value used to perform look up
1243 @param[out] subsumes true if an exact comparison can be done, false otherwise
1244
1245 @returns false if success, true if error
1246 */
1247bool ref_lookup_subsumes_comparison(THD *thd, Field *field, Item *right_item,
1248 bool *subsumes);
1249
1250/**
1251 Checks if we need to create iterators for this query. We usually have to. The
1252 exception is if a secondary engine is used, and that engine will offload the
1253 query execution to an external executor using #JOIN::override_executor_func.
1254 In this case, the external executor will use its own execution structures and
1255 we don't need to bother with creating the iterators needed by the MySQL
1256 executor.
1257 */
1258bool IteratorsAreNeeded(const THD *thd, AccessPath *root_path);
1259
1260/**
1261 Estimates the number of base table row accesses that will be performed when
1262 executing a query using the given plan.
1263
1264 @param path The access path representing the plan.
1265 @param num_evaluations The number of times this path is expected to be
1266 evaluated during a single execution of the query.
1267 @param limit The maximum number of rows expected to be read from this path.
1268 @return An estimate of the number of row accesses.
1269 */
1270double EstimateRowAccesses(const AccessPath *path, double num_evaluations,
1271 double limit);
1272
1273#endif /* SQL_OPTIMIZER_INCLUDED */
bool is_null() const
Definition: sql_array.h:138
Element_type * array() const
Definition: sql_array.h:146
size_t size() const
Definition: sql_array.h:135
size_t element_size() const
Definition: sql_array.h:134
Definition: item_cmpfunc.h:2700
API for getting cost estimates for server operations that are not directly related to a table object.
Definition: opt_costmodel.h:51
Definition: opt_explain_format.h:448
Definition: field.h:574
virtual enum_field_types type() const =0
Definition: item_cmpfunc.h:2562
Definition: item.h:4103
Definition: item_subselect.h:79
Class Item_sum is the base class used for special expressions that SQL calls 'set functions'.
Definition: item_sum.h:398
Base class that is used to represent any kind of expression in a relational query.
Definition: item.h:850
bool is_temporal_with_date() const
Definition: item.h:3057
cond_result
Definition: item.h:919
@ COND_FALSE
Definition: item.h:919
Query optimization plan node.
Definition: sql_select.h:598
Definition: sql_optimizer.h:125
const Cost_model_server * cost_model() const
Retrieve the cost model object to be used for this join.
Definition: sql_optimizer.cc:11313
@ ORDERED_INDEX_GROUP_BY
Definition: sql_optimizer.h:387
@ ORDERED_INDEX_VOID
Definition: sql_optimizer.h:386
@ ORDERED_INDEX_ORDER_BY
Definition: sql_optimizer.h:388
bool skip_sort_order
Is set if we have a GROUP BY and we have ORDER BY on a constant or when sorting isn't required.
Definition: sql_optimizer.h:395
Table_ref * tables_list
Pointer set to query_block->get_table_list() at the start of optimization.
Definition: sql_optimizer.h:508
bool attach_join_condition_to_nest(plan_idx first_inner, plan_idx last_tab, Item *join_cond, bool is_sj_mat_cond)
Helper for JOIN::attach_join_conditions().
Definition: sql_optimizer.cc:8595
void set_root_access_path(AccessPath *path)
Definition: sql_optimizer.h:778
bool calc_found_rows
If true, calculate found rows for this query block.
Definition: sql_optimizer.h:592
bool plan_is_single_table()
True if plan contains one non-const primary table (ie not including tables taking part in semi-join m...
Definition: sql_optimizer.h:610
void mark_const_table(JOIN_TAB *table, Key_use *key)
Move const tables first in the position array.
Definition: sql_optimizer.cc:8426
JOIN_TAB * join_tab
Optimal query execution plan.
Definition: sql_optimizer.h:146
ha_rows fetch_limit
Used to fetch no more than given amount of rows per one fetch operation of server side cursor.
Definition: sql_optimizer.h:297
Item_sum ** sum_funcs
Definition: sql_optimizer.h:341
List< Cached_item > group_fields
Definition: sql_optimizer.h:335
bool m_windows_sort
True if a window requires a certain order of rows, which implies that any order of rows coming out of...
Definition: sql_optimizer.h:457
MYSQL_LOCK * lock
Definition: sql_optimizer.h:349
bool executed
Set by exec(), reset by reset().
Definition: sql_optimizer.h:815
enum JOIN::@175 ORDERED_INDEX_VOID
QEP_TAB * qep_tab
Array of QEP_TABs.
Definition: sql_optimizer.h:148
bool send_row_on_empty_set() const
Return whether the caller should send a row even if the join produced no rows if:
Definition: sql_optimizer.h:705
ha_rows found_records
Definition: sql_optimizer.h:283
uint recursive_iteration_count
Used only if this query block is recursive.
Definition: sql_optimizer.h:558
void copy_ref_item_slice(Ref_item_array dst_arr, Ref_item_array src_arr)
Definition: sql_optimizer.h:635
bool child_subquery_can_materialize
True if, at this stage of processing, subquery materialization is allowed for children subqueries of ...
Definition: sql_optimizer.h:576
Prealloced_array< Item_rollup_group_item *, 4 > rollup_group_items
Definition: sql_optimizer.h:443
COND_EQUAL * cond_equal
Definition: sql_optimizer.h:509
JOIN_TAB ** map2table
mapping between table indexes and JOIN_TABs
Definition: sql_optimizer.h:158
ha_rows m_select_limit
Definition: sql_optimizer.h:287
POSITION * positions
Definition: sql_optimizer.h:309
uint current_ref_item_slice
The slice currently stored in ref_items[0].
Definition: sql_optimizer.h:551
bool is_executed() const
Definition: sql_optimizer.h:745
uint tables
Before plan has been created, "tables" denote number of input tables in the query block and "primary_...
Definition: sql_optimizer.h:207
bool has_lateral
If JOIN has lateral derived tables (is set at start of planning)
Definition: sql_optimizer.h:405
bool need_tmp_before_win
If true we need a temporary table on the result set before any windowing steps, e....
Definition: sql_optimizer.h:402
Prealloced_array< Item_rollup_sum_switcher *, 4 > rollup_sums
Definition: sql_optimizer.h:445
uint tmp_tables
Number of temporary tables used by query.
Definition: sql_optimizer.h:210
int error
set in optimize(), exec(), prepare_result()
Definition: sql_optimizer.h:427
bool plan_is_const() const
True if plan is const, ie it will return zero or one rows.
Definition: sql_optimizer.h:604
table_map const_table_map
Set of tables found to be const.
Definition: sql_optimizer.h:234
Prealloced_array< TemporaryTableToCleanup, 1 > temp_tables
Definition: sql_optimizer.h:176
Query_block *const query_block
Query block that is optimized and executed using this JOIN.
Definition: sql_optimizer.h:137
bool select_distinct
At construction time, set if SELECT DISTINCT.
Definition: sql_optimizer.h:359
RollupState
Definition: sql_optimizer.h:351
List< TABLE > sj_tmp_tables
Definition: sql_optimizer.h:585
table_map found_const_table_map
Const tables which are either:
Definition: sql_optimizer.h:241
bool simple_order
Definition: sql_optimizer.h:377
void set_executed()
Definition: sql_optimizer.h:746
List< Window > m_windows
Any window definitions.
Definition: sql_optimizer.h:451
Explain_format_flags explain_flags
Buffer to gather GROUP BY, ORDER BY and DISTINCT QEP details for EXPLAIN.
Definition: sql_optimizer.h:465
mem_root_deque< Item * > * tmp_fields
Array of pointers to lists of expressions.
Definition: sql_optimizer.h:425
uint const_tables
Number of primary tables deemed constant.
Definition: sql_optimizer.h:209
Prealloced_array< Filesort *, 1 > filesorts_to_cleanup
Definition: sql_optimizer.h:181
ha_rows examined_rows
Definition: sql_optimizer.h:284
ha_rows row_limit
Definition: sql_optimizer.h:285
bool allow_outer_refs
True if plan search is allowed to use references to expressions outer to this JOIN (for example may s...
Definition: sql_optimizer.h:582
JOIN_TAB ** best_ref
Array of plan operators representing the current (partial) best plan.
Definition: sql_optimizer.h:156
Item * having_for_explain
Saved optimized HAVING for EXPLAIN.
Definition: sql_optimizer.h:502
RollupState rollup_state
Definition: sql_optimizer.h:352
AccessPath * root_access_path() const
Definition: sql_optimizer.h:777
ha_rows send_records
Definition: sql_optimizer.h:282
Override_executor_func override_executor_func
Definition: sql_optimizer.h:318
plan_idx return_tab
Definition: sql_optimizer.h:517
enum_plan_state plan_state
Final execution plan state. Currently used only for EXPLAIN.
Definition: sql_optimizer.h:818
Item * having_cond
Optimized HAVING clause item tree (valid for one single execution).
Definition: sql_optimizer.h:501
void make_outerjoin_info()
Fill in outer join related info for the execution plan structure.
Definition: sql_optimizer.cc:8455
mem_root_deque< Item * > * fields
Definition: sql_optimizer.h:334
Temp_table_param tmp_table_param
Describes a temporary table.
Definition: sql_optimizer.h:348
bool select_count
Definition: sql_optimizer.h:828
bool m_windowing_steps
If we have set up tmp tables for windowing,.
Definition: sql_optimizer.h:460
bool finalize_table_conditions(THD *thd)
Remove redundant predicates and cache constant expressions.
Definition: sql_optimizer.cc:9102
bool fts_index_access(JOIN_TAB *tab)
Check if FTS index only access is possible.
Definition: sql_optimizer.cc:10821
void optimize_keyuse()
Update some values in keyuse for faster choose_table_order() loop.
Definition: sql_optimizer.cc:10723
bool with_json_agg
This will force tmp table to NOT use index + update for group operation as it'll cause [de]serializat...
Definition: sql_optimizer.h:601
uint send_group_parts
Definition: sql_optimizer.h:211
AccessPath * m_root_access_path_no_in2exists
If this query block contains conditions synthesized during IN-to-EXISTS conversion: A second query pl...
Definition: sql_optimizer.h:1064
ORDER_with_src group_list
Definition: sql_optimizer.h:439
double sort_cost
Expected cost of filesort.
Definition: sql_optimizer.h:331
bool generate_derived_keys()
Add keys to derived tables'/views' result tables in a list.
Definition: sql_optimizer.cc:9181
bool optimize_fts_query()
Function sets FT hints, initializes FT handlers and checks if FT index can be used as covered.
Definition: sql_optimizer.cc:10762
enum_plan_state
State of execution plan. Currently used only for EXPLAIN.
Definition: sql_optimizer.h:735
@ NO_TABLES
Plan has no tables.
Definition: sql_optimizer.h:738
@ NO_PLAN
No plan is ready yet.
Definition: sql_optimizer.h:736
@ ZERO_RESULT
Zero result cause is set.
Definition: sql_optimizer.h:737
@ PLAN_READY
Plan is ready.
Definition: sql_optimizer.h:739
Key_use_array keyuse_array
Used and updated by JOIN::make_join_plan() and optimize_keyuse()
Definition: sql_optimizer.h:408
bool group_sent
Exec time only: true <=> current group has been sent.
Definition: sql_optimizer.h:590
bool needs_finalize
Whether this query block needs finalization (see FinalizePlanForQueryBlock()) before it can be actual...
Definition: sql_optimizer.h:804
void init_key_dependencies()
Initialize key dependencies for join tables.
Definition: sql_optimizer.h:900
void set_ref_item_slice(uint sliceno)
Overwrite the base slice of ref_items with the slice supplied as argument.
Definition: sql_optimizer.h:659
List< Cached_item > group_fields_cache
Definition: sql_optimizer.h:336
bool contains_non_aggregated_fts() const
Returns true if any of the items in JOIN::fields contains a call to the full-text search function MAT...
Definition: sql_optimizer.cc:10855
bool streaming_aggregation
Indicates that the data will be aggregated (typically GROUP BY), and that it is already processed in ...
Definition: sql_optimizer.h:227
void set_optimized()
Definition: sql_optimizer.h:744
uint primary_tables
Number of primary input tables in query block.
Definition: sql_optimizer.h:208
bool optimize_rollup()
Optimize rollup specification.
Definition: sql_optimizer.cc:11270
THD *const thd
Thread handler.
Definition: sql_optimizer.h:139
table_map all_table_map
Set of tables contained in query.
Definition: sql_optimizer.h:233
bool attach_join_conditions(plan_idx last_tab)
Attach outer join conditions to generated table conditions in an optimal way.
Definition: sql_optimizer.cc:8706
bool decide_subquery_strategy()
Decides between EXISTS and materialization; performs last steps to set up the chosen strategy.
Definition: sql_optimizer.cc:10984
List< Semijoin_mat_exec > sjm_exec_list
Definition: sql_optimizer.h:586
JOIN(const JOIN &rhs)=delete
TABLE * sort_by_table
Definition: sql_optimizer.h:164
Ref_item_array * ref_items
ref_items is an array of 4+ slices, each containing an array of Item pointers.
Definition: sql_optimizer.h:543
bool do_send_rows
If true, send produced rows using query_result.
Definition: sql_optimizer.h:231
double windowing_cost
Expected cost of windowing;.
Definition: sql_optimizer.h:333
enum_plan_state get_plan_state() const
See enum_plan_state.
Definition: sql_optimizer.h:742
JOIN & operator=(const JOIN &rhs)=delete
ORDER_with_src order
ORDER BY and GROUP BY lists, to transform with prepare,optimize and exec.
Definition: sql_optimizer.h:439
bool group_optimized_away
If we have the GROUP BY statement in the query, but the group_list was emptied by optimizer,...
Definition: sql_optimizer.h:368
double best_read
The cost of best complete join plan found so far during optimization, after optimization phase - cost...
Definition: sql_optimizer.h:325
ORDER * remove_const(ORDER *first_order, Item *cond, bool change_list, bool *simple_order, bool group_by)
Remove all constants and check if ORDER only contains simple expressions.
Definition: sql_optimizer.cc:10098
bool implicit_grouping
True if aggregated but no GROUP BY.
Definition: sql_optimizer.h:353
POSITION * best_positions
This is the result of join optimization.
Definition: sql_optimizer.h:304
void finalize_derived_keys()
For each materialized derived table/view, informs every TABLE of the key it will (not) use,...
Definition: sql_optimizer.cc:9201
bool optimized
flag to avoid double optimization in EXPLAIN
Definition: sql_optimizer.h:807
bool compare_costs_of_subquery_strategies(Subquery_strategy *method)
Tells what is the cheapest between IN->EXISTS and subquery materialization, in terms of cost,...
Definition: sql_optimizer.cc:11045
bool is_optimized() const
Definition: sql_optimizer.h:743
const char * zero_result_cause
<> NULL if optimization has determined that execution will produce an empty result before aggregation...
Definition: sql_optimizer.h:568
Query_expression * query_expression() const
Query expression referring this query block.
Definition: sql_optimizer.h:132
mem_root_deque< Item * > * get_current_fields()
Returns the clone of fields_list which is appropriate for evaluating expressions at the current stage...
Definition: sql_optimizer.cc:11307
bool(*)(JOIN *, Query_result *) Override_executor_func
A hook that secondary storage engines can use to override the executor completely.
Definition: sql_optimizer.h:317
void clear_hash_tables()
Definition: sql_optimizer.h:731
uint get_ref_item_slice() const
Definition: sql_optimizer.h:670
List< Cached_item > semijoin_deduplication_fields
Definition: sql_optimizer.h:339
bool grouped
If query contains GROUP BY clause.
Definition: sql_optimizer.h:229
void refine_best_rowcount()
Refine the best_rowcount estimation based on what happens after tables have been joined: LIMIT and ty...
Definition: sql_optimizer.cc:11281
AccessPath * m_root_access_path
An access path you can read from to get all records for this query (after you create an iterator from...
Definition: sql_optimizer.h:1057
ha_rows best_rowcount
The estimated row count of the plan with best read time (see above).
Definition: sql_optimizer.h:329
Item * where_cond
JOIN::having_cond is initially equal to query_block->having_cond, but may later be changed by optimiz...
Definition: sql_optimizer.h:491
bool simple_group
Definition: sql_optimizer.h:378
uint64_t hash_table_generation
Incremented each time clear_hash_tables() is run, signaling to HashJoinIterators that they cannot kee...
Definition: sql_optimizer.h:434
table_map deps_of_remaining_lateral_derived_tables
This is the bitmap of all tables which are dependencies of lateral derived tables which are not (yet)...
Definition: sql_optimizer.h:279
void copy_ref_item_slice(uint dst_slice, uint src_slice)
Overwrites one slice of ref_items with the contents of another slice.
Definition: sql_optimizer.h:632
A Key_use represents an equality predicate of the form (table.column = val), where the column is inde...
Definition: sql_select.h:173
Definition: sql_list.h:433
A typesafe replacement for DYNAMIC_ARRAY.
Definition: mem_root_array.h:425
Wrapper for ORDER* pointer to trace origins of ORDER list.
Definition: sql_optimizer.h:95
bool empty() const
Definition: sql_optimizer.h:111
int get_flags() const
Definition: sql_optimizer.h:119
void clean()
Definition: sql_optimizer.h:113
ORDER * order
ORDER expression that we are wrapping with this class.
Definition: sql_optimizer.h:97
ORDER_with_src()
Definition: sql_optimizer.h:104
Explain_sort_clause src
origin of order list
Definition: sql_optimizer.h:98
int flags
bitmap of Explain_sort_property
Definition: sql_optimizer.h:101
ORDER_with_src(ORDER *order_arg, Explain_sort_clause src_arg)
Definition: sql_optimizer.h:106
A per-session context which is always available at any point of execution, because in practice it's a...
Definition: opt_trace_context.h:89
A typesafe replacement for DYNAMIC_ARRAY.
Definition: prealloced_array.h:70
Definition: sql_executor.h:259
enum_op_type
Definition: sql_executor.h:406
This class represents a query block, aka a query specification, which is a query consisting of a SELE...
Definition: sql_lex.h:1153
Item::cond_result having_value
Definition: sql_lex.h:2008
Query_expression * master_query_expression() const
Definition: sql_lex.h:1243
This class represents a query expression (one query block or several query blocks combined with UNION...
Definition: sql_lex.h:623
Definition: query_result.h:53
RAII class to ease the temporary switching to a different slice of the ref item array.
Definition: sql_optimizer.h:1081
Switch_ref_item_slice(JOIN *join_arg, uint new_v)
Definition: sql_optimizer.h:1086
uint saved
Definition: sql_optimizer.h:1083
JOIN * join
Definition: sql_optimizer.h:1082
~Switch_ref_item_slice()
Definition: sql_optimizer.h:1090
For each client connection we create a separate thread with THD serving as a thread/connection descri...
Definition: sql_lexer_thd.h:33
This class restores a table_map object to its original value when '*this' is destroyed.
Definition: sql_optimizer.h:1154
Table_map_restorer(table_map *map)
Constructor.
Definition: sql_optimizer.h:1165
Table_map_restorer & operator=(const Table_map_restorer &)=delete
table_map *const m_location
The location to be restored.
Definition: sql_optimizer.h:1156
~Table_map_restorer()
Definition: sql_optimizer.h:1172
const table_map m_saved_value
The original value to restore.
Definition: sql_optimizer.h:1158
void restore()
Definition: sql_optimizer.h:1173
void assert_unchanged() const
Definition: sql_optimizer.h:1174
Table_map_restorer(const Table_map_restorer &)=delete
Definition: table.h:2755
Object containing parameters used when creating and using temporary tables.
Definition: temp_table_param.h:94
uint sum_func_count
Number of fields in the query that have aggregate functions.
Definition: temp_table_param.h:131
Represents the (explicit) window of a SQL 2003 section 7.11 <window clause>, or the implicit (inlined...
Definition: window.h:104
A (partial) implementation of std::deque allocating its blocks on a MEM_ROOT.
Definition: mem_root_deque.h:109
bool is_temporal_type(enum_field_types type)
Tests if field type is temporal, i.e.
Definition: field_common_properties.h:114
bool is_temporal_type_with_date(enum_field_types type)
Tests if field type is temporal and has date part, i.e.
Definition: field_common_properties.h:155
This file contains the field type.
enum_field_types
Column types for MySQL.
Definition: field_types.h:52
void optimize_distinct()
Optimize distinct when used on a subset of the tables.
Definition: sql_executor.cc:344
AccessPath * create_root_access_path_for_join()
Definition: sql_executor.cc:3004
void restore_fields(table_map save_nullinfo)
Restore all result fields for all tables specified in save_nullinfo.
Definition: sql_executor.cc:4637
bool create_intermediate_table(QEP_TAB *tab, const mem_root_deque< Item * > &tmp_table_fields, ORDER_with_src &tmp_table_group, bool save_sum_fields)
Create a temporary table to be used for processing DISTINCT/ORDER BY/GROUP BY.
Definition: sql_executor.cc:177
AccessPath * attach_access_paths_for_having_and_limit(AccessPath *path)
Definition: sql_executor.cc:3346
QEP_TAB::enum_op_type get_end_select_func()
Definition: sql_executor.cc:572
void create_access_paths()
Convert the executor structures to a set of access paths, storing the result in m_root_access_path.
Definition: sql_executor.cc:2974
void create_access_paths_for_index_subquery()
Definition: sql_executor.cc:3381
AccessPath * attach_access_path_for_update_or_delete(AccessPath *path)
Definition: sql_executor.cc:2938
bool clear_fields(table_map *save_nullinfo)
Set all column values from all input tables to NULL.
Definition: sql_executor.cc:4616
bool clear_corr_derived_tmp_tables()
Empties all correlated materialized derived tables.
Definition: sql_select.cc:1576
Item_equal * find_item_equal(COND_EQUAL *cond_equal, const Item_field *item_field, bool *inherited_fl)
Find the multiple equality predicate containing a field.
Definition: sql_optimizer.cc:3689
bool alloc_func_list()
Make an array of pointers to sum_functions to speed up sum_func calculation.
Definition: sql_select.cc:3882
Item_field * get_best_field(Item_field *item_field, COND_EQUAL *cond_equal)
Get the best field substitution for a given field.
Definition: sql_optimizer.cc:3722
void update_sargable_from_const(SARGABLE_PARAM *sargables)
Update info on indexes that can be used for search lookups as reading const tables may has added new ...
Definition: sql_optimizer.cc:5817
bool add_sorting_to_table(uint idx, ORDER_with_src *order, bool sort_before_group)
Add Filesort object to the given table to sort if with filesort.
Definition: sql_select.cc:4795
bool clear_sj_tmp_tables()
Remove all rows from all temp tables used by NL-semijoin runtime.
Definition: sql_select.cc:1566
bool alloc_qep(uint n)
Definition: sql_optimizer.cc:1327
void change_to_access_path_without_in2exists()
If this query block was planned twice, once with and once without conditions added by in2exists,...
Definition: sql_optimizer.cc:1126
void unplug_join_tabs()
Definition: sql_select.cc:4767
bool push_to_engines()
Handle offloading of query parts to the underlying engines, when such is supported by their implement...
Definition: sql_optimizer.cc:1162
bool make_tmp_tables_info()
Init tmp tables usage info.
Definition: sql_select.cc:4165
bool make_join_plan()
Calculate best possible join order and initialize the join structure.
Definition: sql_optimizer.cc:5258
void set_semijoin_embedding()
Set semi-join embedding join nest pointers.
Definition: sql_optimizer.cc:5991
bool optimize_distinct_group_order()
Optimize DISTINCT, GROUP BY, ORDER BY clauses.
Definition: sql_optimizer.cc:1475
JOIN(THD *thd_arg, Query_block *select)
Definition: sql_optimizer.cc:163
table_map calculate_deps_of_remaining_lateral_derived_tables(table_map plan_tables, uint idx) const
Finds the dependencies of the remaining lateral derived tables.
Definition: sql_optimizer.cc:3274
void set_prefix_tables()
Assign set of available (prefix) tables to all tables in query block.
Definition: sql_optimizer.cc:5151
void cleanup()
Cleanup this JOIN.
Definition: sql_select.cc:3510
bool uses_index_fields_only(Item *item, TABLE *tbl, uint keyno, bool other_tbls_ok)
Check if given expression only uses fields covered by index keyno in the table tbl.
Definition: sql_optimizer.cc:6445
void test_skip_sort()
Test if an index could be used to replace filesort for ORDER BY/GROUP BY.
Definition: sql_optimizer.cc:1668
void create_access_paths_for_zero_rows()
Create access paths with the knowledge that there are going to be zero rows coming from tables (befor...
Definition: sql_optimizer.cc:1132
bool propagate_dependencies()
Propagate dependencies between tables due to outer join relations.
Definition: sql_optimizer.cc:5515
void adjust_access_methods()
An utility function - apply heuristics and optimize access methods to tables.
Definition: sql_optimizer.cc:2932
bool estimate_rowcount()
Estimate the number of matched rows for each joined table.
Definition: sql_optimizer.cc:5853
bool prune_table_partitions()
Prune partitions for all tables of a join (query block).
Definition: sql_optimizer.cc:2781
uint build_bitmap_for_nested_joins(mem_root_deque< Table_ref * > *join_list, uint first_unused)
Assign each nested join structure a bit in nested_join_map.
Definition: sql_optimizer.cc:4976
void refresh_base_slice()
In the case of rollup (only): After the base slice list was made, we may have modified the field list...
Definition: sql_select.cc:4742
bool alloc_ref_item_slice(THD *thd_arg, int sliceno)
Allocate a ref_item slice, assume that slice size is in ref_items[0].
Definition: sql_optimizer.cc:203
bool setup_semijoin_materialized_table(JOIN_TAB *tab, uint tableno, POSITION *inner_pos, POSITION *sjm_pos)
Setup the materialized table for a semi-join nest.
Definition: sql_select.cc:2893
void join_free()
Release memory and, if possible, the open tables held by this execution plan (and nested plans).
Definition: sql_select.cc:3442
bool make_sum_func_list(const mem_root_deque< Item * > &fields, bool before_group_by, bool recompute=false)
Initialize 'sum_funcs' array with all Item_sum objects.
Definition: sql_select.cc:3930
bool extract_func_dependent_tables()
Extract const tables based on functional dependencies.
Definition: sql_optimizer.cc:5660
bool init_planner_arrays()
Initialize scratch arrays for the join order optimization.
Definition: sql_optimizer.cc:5388
bool substitute_gc(THD *thd, Query_block *query_block, Item *where_cond, ORDER *group_list, ORDER *order)
Substitute all expressions in the WHERE condition and ORDER/GROUP lists that match generated columns ...
Definition: sql_optimizer.cc:1208
void set_semijoin_info()
Set the first_sj_inner_tab and last_sj_inner_tab fields for all tables inside the semijoin nests of t...
Definition: sql_select.cc:2025
bool prepare_result()
Prepare join result.
Definition: sql_select.cc:1660
void destroy()
Clean up and destroy join object.
Definition: sql_select.cc:1678
bool add_having_as_tmp_table_cond(uint curr_tmp_table)
Add having condition as a filter condition, which is applied when reading from the temp table.
Definition: sql_select.cc:4009
Item * substitute_for_best_equal_field(THD *thd, Item *cond, COND_EQUAL *cond_equal, JOIN_TAB **table_join_idx)
Substitute every field reference in a condition by the best equal field and eliminate all multiple eq...
Definition: sql_optimizer.cc:4724
double find_worst_seeks(const TABLE *table, double num_rows, double table_scan_cost)
Find an artificial cap for ref access.
Definition: sql_optimizer.cc:5833
const char * antijoin_null_cond
Definition: sql_optimizer.cc:120
bool alloc_indirection_slices()
Definition: sql_optimizer.cc:213
void reset()
Reset the state of this join object so that it is ready for a new execution.
Definition: sql_select.cc:1597
bool optimize(bool finalize_access_paths)
Optimizes one query block into a query execution plan (QEP.)
Definition: sql_optimizer.cc:336
bool build_equal_items(THD *thd, Item *cond, Item **retcond, COND_EQUAL *inherited, bool do_inherit, mem_root_deque< Table_ref * > *join_list, COND_EQUAL **cond_equal_ref)
Build multiple equalities for a WHERE condition and all join conditions that inherit these multiple e...
Definition: sql_optimizer.cc:4418
bool init_ref_access()
Initialize ref access for all tables that use it.
Definition: sql_select.cc:2002
bool get_best_combination()
Set up JOIN_TAB structs according to the picked join order in best_positions.
Definition: sql_optimizer.cc:3050
bool extract_const_tables()
Extract const tables based on row counts.
Definition: sql_optimizer.cc:5566
bool update_equalities_for_sjm()
Update equalities and keyuse references after semi-join materialization strategy is chosen.
Definition: sql_optimizer.cc:5087
void set_plan_state(enum_plan_state plan_state_arg)
Sets the plan's state of the JOIN.
Definition: sql_optimizer.cc:1299
bool check_access_path_with_fts() const
Checks if the chosen plan suffers from a problem related to full-text search and streaming aggregatio...
Definition: sql_optimizer.cc:253
void cleanup_item_list(const mem_root_deque< Item * > &items) const
Definition: sql_select.cc:1783
int replace_index_subquery()
Check whether this is a subquery that can be evaluated by index look-ups.
Definition: sql_optimizer.cc:1407
void update_depend_map()
Update the dependency map for the tables.
Definition: sql_optimizer.cc:5018
Subquery_strategy
Strategy which will be used to handle this subquery: flattening to a semi-join, conversion to a deriv...
Definition: item_subselect.h:392
This file follows Google coding style, except for the name MEM_ROOT (which is kept for historical rea...
This file includes constants used by all storage engines.
my_off_t ha_rows
Definition: my_base.h:1139
#define HA_POS_ERROR
Definition: my_base.h:1141
#define DBUG_PRINT(keyword, arglist)
Definition: my_dbug.h:180
uint64_t table_map
Definition: my_table_map.h:29
static char * path
Definition: mysqldump.cc:133
std::string join(Container cont, const std::string &delim)
join elements of an container into a string separated by a delimiter.
Definition: string.h:150
std::map< Key, Value, Compare, ut::allocator< std::pair< const Key, Value > > > map
Specialization of map which uses ut_allocator.
Definition: ut0new.h:2890
EXPLAIN FORMAT=<format> <command>.
Explain_sort_clause
Enumeration of ORDER BY, GROUP BY and DISTINCT clauses for array indexing.
Definition: opt_explain_format.h:425
@ ESC_none
Definition: opt_explain_format.h:426
@ ESP_EXISTS
Original query has this clause.
Definition: opt_explain_format.h:441
@ ESP_none
Definition: opt_explain_format.h:440
required string key
Definition: replication_asynchronous_connection_failover.proto:59
Classes for query execution.
Common types of the Optimizer, used by optimization and execution.
@ REF_SLICE_ACTIVE
The slice which is used during evaluation of expressions; Item_ref::ref points there.
Definition: sql_opt_exec_shared.h:620
int plan_idx
This represents the index of a JOIN_TAB/QEP_TAB in an array.
Definition: sql_opt_exec_shared.h:53
bool evaluate_during_optimization(const Item *item, const Query_block *select)
Checks if an Item, which is constant for execution, can be evaluated during optimization.
Definition: sql_optimizer.cc:11352
Item * make_cond_for_table(THD *thd, Item *cond, table_map tables, table_map used_table, bool exclude_expensive_cond)
Extract a condition that can be checked after reading given table.
Definition: sql_optimizer.cc:9417
double calculate_subquery_executions(const Item_subselect *subquery, Opt_trace_context *trace)
Estimates how many times a subquery will be executed as part of a query execution.
Definition: sql_optimizer.cc:11155
bool ref_lookup_subsumes_comparison(THD *thd, Field *field, Item *right_item, bool *subsumes)
Whether a ref lookup of “right_item” on “field” will give an exact comparison in all cases,...
Definition: sql_optimizer.cc:8833
Key_use_array * create_keyuse_for_table(THD *thd, uint keyparts, Item_field **fields, const mem_root_deque< Item * > &outer_exprs)
Create a keyuse array for a table with a primary key.
Definition: sql_optimizer.cc:8389
double EstimateRowAccesses(const AccessPath *path, double num_evaluations, double limit)
Estimates the number of base table row accesses that will be performed when executing a query using t...
Definition: sql_optimizer.cc:11494
ORDER * create_order_from_distinct(THD *thd, Ref_item_array ref_item_array, ORDER *order_list, mem_root_deque< Item * > *fields, bool skip_aggregates, bool convert_bit_fields_to_long, bool *all_order_by_fields_used)
Create an order list that consists of all non-const fields and items.
Definition: sql_optimizer.cc:10610
bool field_time_cmp_date(const Field *f, const Item *v)
Returns true if arguments are a temporal Field having no date, part and a temporal expression having ...
Definition: sql_optimizer.h:1141
bool is_indexed_agg_distinct(JOIN *join, mem_root_deque< Item_field * > *out_args)
Check for the presence of AGGFN(DISTINCT a) queries that may be subject to loose index scan.
Definition: sql_optimizer.cc:7972
bool IteratorsAreNeeded(const THD *thd, AccessPath *root_path)
Checks if we need to create iterators for this query.
Definition: sql_optimizer.cc:11386
bool optimize_cond(THD *thd, Item **conds, COND_EQUAL **cond_equal, mem_root_deque< Table_ref * > *join_list, Item::cond_result *cond_value)
Optimize conditions by.
Definition: sql_optimizer.cc:10247
bool remove_eq_conds(THD *thd, Item *cond, Item **retcond, Item::cond_result *cond_value)
Removes const and eq items.
Definition: sql_optimizer.cc:10363
Access paths are a query planning structure that correspond 1:1 to iterators, in that an access path ...
Definition: access_path.h:189
Definition: sql_optimizer.h:169
Temp_table_param * temp_table_param
Definition: sql_optimizer.h:174
TABLE * table
Definition: sql_optimizer.h:170
Definition: lock.h:38
Definition: table.h:280
A position of table within a join order.
Definition: sql_select.h:351
Definition: sql_optimizer.h:82
Item ** arg_value
Definition: sql_optimizer.h:84
Field * field
Definition: sql_optimizer.h:83
uint num_values
Definition: sql_optimizer.h:85
Definition: table.h:1395
unsigned int uint
Definition: uca-dump.cc:29
#define PSI_NOT_INSTRUMENTED
Definition: validate_password_imp.cc:39
int n
Definition: xcom_base.cc:508