Query Executor

Enumerations
enum class	Substructure { Substructure::NONE , Substructure::OUTER_JOIN , Substructure::SEMIJOIN , Substructure::WEEDOUT }

Functions
static int	read_system (TABLE *table)
	Read a constant table when there is at most one matching row, using a table scan. More...
static bool	alloc_group_fields (JOIN *join, ORDER *group)
	Get a list of buffers for saving last group. More...
string	RefToString (const Index_lookup &ref, const KEY *key, bool include_nulls)
bool	has_rollup_result (Item *item)
	Checks if an item has a ROLLUP NULL which needs to be written to temp table. More...
bool	is_rollup_group_wrapper (const Item *item)
Item *	unwrap_rollup_group (Item *item)
bool	prepare_sum_aggregators (Item_sum **sum_funcs, bool need_distinct)
bool	setup_sum_funcs (THD *thd, Item_sum **func_ptr)
	Call setup() for all sum functions. More...
void	init_tmptable_sum_functions (Item_sum **func_ptr)
void	update_tmptable_sum_func (Item_sum **func_ptr, TABLE *tmp_table)
	Update record 0 in tmp_table from record 1. More...
bool	copy_funcs (Temp_table_param *param, const THD *thd, Copy_func_type type)
	Copy result of functions to record in tmp_table. More...
static bool	update_const_equal_items (THD *thd, Item *cond, JOIN_TAB *tab)
	Check appearance of new constant items in multiple equalities of a condition after reading a constant table. More...
void	setup_tmptable_write_func (QEP_TAB *tab, Opt_trace_object *trace)
	Setup write_func of QEP_tmp_table object. More...
static size_t	record_prefix_size (const TABLE *table)
	Find out how many bytes it takes to store the smallest prefix which covers all the columns that will be read from a table. More...
bool	set_record_buffer (TABLE *table, double expected_rows_to_fetch)
	Allocate a data buffer that the storage engine can use for fetching batches of records. More...
bool	ExtractConditions (Item *condition, Mem_root_array< Item * > *condition_parts)
	Split AND conditions into their constituent parts, recursively. More...
static bool	ContainsAnyMRRPaths (AccessPath *path)
	See if “path” has any MRR nodes; if so, we cannot optimize them away in PossiblyAttachFilter(), as the BKA iterator expects there to be a corresponding MRR iterator. More...
Item *	CreateConjunction (List< Item > *items)
	Create an AND conjunction of all given items. More...
AccessPath *	PossiblyAttachFilter (AccessPath *path, const vector< Item * > &conditions, THD *thd, table_map *conditions_depend_on_outer_tables)
	Return a new iterator that wraps "iterator" and that tests all of the given conditions (if any), ANDed together. More...
AccessPath *	CreateNestedLoopAccessPath (THD *thd, AccessPath *outer, AccessPath *inner, JoinType join_type, bool pfs_batch_mode)
static AccessPath *	NewInvalidatorAccessPathForTable (THD *thd, AccessPath *path, QEP_TAB *qep_tab, plan_idx table_index_to_invalidate)
static table_map	ConvertQepTabMapToTableMap (JOIN *join, qep_tab_map tables)
AccessPath *	CreateBKAAccessPath (THD *thd, JOIN *join, AccessPath *outer_path, qep_tab_map left_tables, AccessPath *inner_path, qep_tab_map right_tables, TABLE *table, Table_ref *table_list, Index_lookup *ref, JoinType join_type)
static AccessPath *	PossiblyAttachFilter (AccessPath *path, const vector< PendingCondition > &conditions, THD *thd, table_map *conditions_depend_on_outer_tables)
static Item_func_trig_cond *	GetTriggerCondOrNull (Item *item)
void	ConvertItemsToCopy (const mem_root_deque< Item * > &items, Field **fields, Temp_table_param *param)
	For historical reasons, derived table materialization and temporary table materialization didn't specify the fields to materialize in the same way. More...
static bool	IsJoinCondition (const Item *item, const QEP_TAB *qep_tab)
static Item *	GetInnermostCondition (Item *item)
static bool	CheckIfFieldsAvailableForCond (Item *item, table_map build_tables, table_map probe_tables)
static void	AttachSemiJoinCondition (Item *join_cond, vector< PendingCondition > *join_conditions, QEP_TAB *current_table, qep_tab_map left_tables, plan_idx semi_join_table_idx)
void	SplitConditions (Item *condition, QEP_TAB *current_table, vector< Item * > *predicates_below_join, vector< PendingCondition > *predicates_above_join, vector< PendingCondition > *join_conditions, plan_idx semi_join_table_idx, qep_tab_map left_tables)
static void	MarkUnhandledDuplicates (SJ_TMP_TABLE *weedout, plan_idx weedout_start, plan_idx weedout_end, qep_tab_map *unhandled_duplicates)
	For a given duplicate weedout operation, figure out which tables are supposed to be deduplicated by it, and add those to unhandled_duplicates. More...
static AccessPath *	CreateWeedoutOrLimitAccessPath (THD *thd, AccessPath *path, SJ_TMP_TABLE *weedout_table)
static AccessPath *	NewWeedoutAccessPathForTables (THD *thd, const qep_tab_map tables_to_deduplicate, QEP_TAB *qep_tabs, uint primary_tables, AccessPath *path)
static Substructure	FindSubstructure (QEP_TAB *qep_tabs, const plan_idx first_idx, const plan_idx this_idx, const plan_idx last_idx, CallingContext calling_context, bool *add_limit_1, plan_idx *substructure_end, qep_tab_map *unhandled_duplicates)
	Given a range of tables (where we assume that we've already handled first_idx..(this_idx-1) as inner joins), figure out whether this is a semijoin, an outer join or a weedout. More...
static bool	IsTableScan (AccessPath *path)
AccessPath *	GetAccessPathForDerivedTable (THD *thd, QEP_TAB *qep_tab, AccessPath *table_path)
static void	RecalculateTablePathCost (AccessPath *path, const Query_block &outer_query_block)
	Recalculate the cost of 'path'. More...
AccessPath *	MoveCompositeIteratorsFromTablePath (AccessPath *path, const Query_block &outer_query_block)
	For a MATERIALIZE access path, move any non-basic iterators (e.g. More...
AccessPath *	GetAccessPathForDerivedTable (THD *thd, Table_ref *table_ref, TABLE *table, bool rematerialize, Mem_root_array< const AccessPath * > *invalidators, bool need_rowid, AccessPath *table_path)
AccessPath *	GetTableAccessPath (THD *thd, QEP_TAB *qep_tab, QEP_TAB *qep_tabs)
	Get the RowIterator used for scanning the given table, with any required materialization operations done first. More...
void	SetCostOnTableAccessPath (const Cost_model_server &cost_model, const POSITION *pos, bool is_after_filter, AccessPath *path)
void	SetCostOnNestedLoopAccessPath (const Cost_model_server &cost_model, const POSITION *pos_inner, AccessPath *path)
void	SetCostOnHashJoinAccessPath (const Cost_model_server &cost_model, const POSITION *pos_outer, AccessPath *path)
static bool	ConditionIsAlwaysTrue (Item *item)
static AccessPath *	CreateHashJoinAccessPath (THD *thd, QEP_TAB *qep_tab, AccessPath *build_path, qep_tab_map build_tables, AccessPath *probe_path, qep_tab_map probe_tables, JoinType join_type, vector< Item * > *join_conditions, table_map *conditions_depend_on_outer_tables)
static void	ExtractJoinConditions (const QEP_TAB *current_table, vector< Item * > *predicates, vector< Item * > *join_conditions)
static bool	UseHashJoin (QEP_TAB *qep_tab)
static bool	UseBKA (QEP_TAB *qep_tab)
static bool	QueryMixesOuterBKAAndBNL (JOIN *join)
static bool	InsideOuterOrAntiJoin (QEP_TAB *qep_tab)
void	PickOutConditionsForTableIndex (int table_idx, vector< PendingCondition > *from, vector< PendingCondition > *to)
void	PickOutConditionsForTableIndex (int table_idx, vector< PendingCondition > *from, vector< Item * > *to)
AccessPath *	FinishPendingOperations (THD *thd, AccessPath *path, QEP_TAB *remove_duplicates_loose_scan_qep_tab, const vector< PendingCondition > &pending_conditions, table_map *conditions_depend_on_outer_tables)
AccessPath *	ConnectJoins (plan_idx upper_first_idx, plan_idx first_idx, plan_idx last_idx, QEP_TAB *qep_tabs, THD *thd, CallingContext calling_context, vector< PendingCondition > *pending_conditions, vector< PendingInvalidator > *pending_invalidators, vector< PendingCondition > *pending_join_conditions, qep_tab_map *unhandled_duplicates, table_map *conditions_depend_on_outer_tables)
	For a given slice of the table list, build up the iterator tree corresponding to the tables in that slice. More...
static table_map	get_update_or_delete_target_tables (const JOIN *join)
int	do_sj_dups_weedout (THD *thd, SJ_TMP_TABLE *sjtbl)
	SemiJoinDuplicateElimination: Weed out duplicate row combinations. More...
int	report_handler_error (TABLE *table, int error)
	Help function when we get some an error from the table handler. More...
int	join_read_const_table (JOIN_TAB *tab, POSITION *pos)
	Reads content of constant table. More...
int	read_const (TABLE *table, Index_lookup *ref)
static bool	cmp_field_value (Field *field, ptrdiff_t diff)
static bool	group_rec_cmp (ORDER *group, uchar *rec0, uchar *rec1)
	Compare GROUP BY in from tmp table's record[0] and record[1]. More...
static bool	table_rec_cmp (TABLE *table)
	Compare GROUP BY in from tmp table's record[0] and record[1]. More...
ulonglong	unique_hash (const Field *field, ulonglong *hash_val)
	Generate hash for a field. More...
static ulonglong	unique_hash_group (ORDER *group)
	Generate hash for unique constraint according to group-by list. More...
static ulonglong	unique_hash_fields (TABLE *table)
	Generate hash for unique_constraint for all visible fields of a table. More...
bool	check_unique_constraint (TABLE *table)
	Check unique_constraint. More...
bool	construct_lookup (THD *thd, TABLE *table, Index_lookup *ref)
	Copy the lookup key into the table ref's key buffer. More...
bool	make_group_fields (JOIN *main_join, JOIN *curr_join)
	allocate group fields or take prepared (cached). More...
int	update_item_cache_if_changed (List< Cached_item > &list)
static size_t	compute_ria_idx (const mem_root_deque< Item * > &fields, size_t i, size_t added_non_hidden_fields, size_t border)
	Compute the position mapping from fields to ref_item_array, cf. More...
bool	copy_fields (Temp_table_param *param, const THD *thd, bool reverse_copy)
	Make a copy of all simple SELECT'ed fields. More...
static bool	replace_embedded_rollup_references_with_tmp_fields (THD *thd, Item *item, mem_root_deque< Item * > *fields)
	For each rollup wrapper below the given item, replace it with a temporary field, e.g. More...
bool	change_to_use_tmp_fields (mem_root_deque< Item * > *fields, THD *thd, Ref_item_array ref_item_array, mem_root_deque< Item * > *res_fields, size_t added_non_hidden_fields)
	Change all funcs and sum_funcs to fields in tmp table, and create new list of all items. More...
static Item_rollup_group_item *	find_rollup_item_in_group_list (Item *item, Query_block *query_block)
bool	replace_contents_of_rollup_wrappers_with_tmp_fields (THD *thd, Query_block *select, Item *item_arg)
	For each rollup wrapper below the given item, replace its argument with a temporary field, e.g. More...
bool	change_to_use_tmp_fields_except_sums (mem_root_deque< Item * > *fields, THD *thd, Query_block *select, Ref_item_array ref_item_array, mem_root_deque< Item * > *res_fields, size_t added_non_hidden_fields)
	Change all sum_func refs to fields to point at fields in tmp table. More...
bool	JOIN::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. More...
void	JOIN::optimize_distinct ()
	Optimize distinct when used on a subset of the tables. More...
QEP_TAB::enum_op_type	JOIN::get_end_select_func ()
void	JOIN::create_access_paths ()
	Convert the executor structures to a set of access paths, storing the result in m_root_access_path. More...
void	JOIN::create_access_paths_for_index_subquery ()
bool	QEP_TAB::use_order () const
	Use ordering provided by chosen index? More...
AccessPath *	QEP_TAB::access_path ()
	Construct an access path for reading from this table in the query, using the access method that has been determined previously (e.g., table scan, ref access, optional sort afterwards, etc.). More...
bool	JOIN::clear_fields (table_map *save_nullinfo)
	Set all column values from all input tables to NULL. More...
void	JOIN::restore_fields (table_map save_nullinfo)
	Restore all result fields for all tables specified in save_nullinfo. More...
bool	QEP_TAB::pfs_batch_update (const JOIN *join) const

Variables
static constexpr size_t	MAX_RECORD_BUFFER_SIZE = 128 * 1024
	Maximum amount of space (in bytes) to allocate for a Record_buffer. More...
AccessPath *	JOIN::attach_access_path_for_update_or_delete (AccessPath *path)
AccessPath *	JOIN::create_root_access_path_for_join ()
AccessPath *	JOIN::attach_access_paths_for_having_and_limit (AccessPath *path)

Detailed Description

Enumeration Type Documentation

Substructure

enum class Substructure

strong

Enumerator
NONE
OUTER_JOIN
SEMIJOIN
WEEDOUT

Function Documentation

access_path()

AccessPath * QEP_TAB::access_path

(

)

Construct an access path for reading from this table in the query, using the access method that has been determined previously (e.g., table scan, ref access, optional sort afterwards, etc.).

alloc_group_fields()

static bool alloc_group_fields ( JOIN *  join, ORDER *  group  )

static

Get a list of buffers for saving last group.

Groups are saved in reverse order for easier check loop.

attach_access_path_for_update_or_delete()

AccessPath * JOIN::attach_access_path_for_update_or_delete ( AccessPath *  path )

private

attach_access_paths_for_having_and_limit()

AccessPath * JOIN::attach_access_paths_for_having_and_limit ( AccessPath *  path )

private

AttachSemiJoinCondition()

static void AttachSemiJoinCondition ( Item *  join_cond, vector< PendingCondition > *  join_conditions, QEP_TAB *  current_table, qep_tab_map  left_tables, plan_idx  semi_join_table_idx  )

static

change_to_use_tmp_fields()

bool change_to_use_tmp_fields	(	mem_root_deque< Item * > *	fields,
		THD *	thd,
		Ref_item_array	ref_item_array,
		mem_root_deque< Item * > *	res_fields,
		size_t	added_non_hidden_fields
	)

Change all funcs and sum_funcs to fields in tmp table, and create new list of all items.

Parameters

	fields	list of all fields; should really be const, but Item does not always respect constness
	thd	THD pointer
[out]	ref_item_array	array of pointers to top elements of filed list
[out]	res_fields	new list of all items
	added_non_hidden_fields	number of visible fields added by subquery to derived transformation

Returns

false if success, true if error

change_to_use_tmp_fields_except_sums()

bool change_to_use_tmp_fields_except_sums	(	mem_root_deque< Item * > *	fields,
		THD *	thd,
		Query_block *	select,
		Ref_item_array	ref_item_array,
		mem_root_deque< Item * > *	res_fields,
		size_t	added_non_hidden_fields
	)

Change all sum_func refs to fields to point at fields in tmp table.

Change all funcs to be fields in tmp table.

This is used when we set up a temporary table, but aggregate functions (sum_funcs) cannot be evaluated yet, for instance because data is not sorted in the right order. (Otherwise, change_to_use_tmp_fields() would be used.)

Parameters

	fields	list of all fields; should really be const, but Item does not always respect constness
	select	the query block we are doing this to
	thd	THD pointer
[out]	ref_item_array	array of pointers to top elements of filed list
[out]	res_fields	new list of items of select item list
	added_non_hidden_fields	number of visible fields added by subquery to derived transformation

Returns

false if success, true if error

check_unique_constraint()

bool check_unique_constraint

(

TABLE *

table

)

Check unique_constraint.

Calculates record's hash and checks whether the record given in table->record[0] is already present in the tmp table.

Parameters

table

JOIN_TAB of tmp table to check

Note

This function assumes record[0] is already filled by the caller. Depending on presence of table->group, it's or full list of table's fields are used to calculate hash.

Returns

false same record was found true record wasn't found

CheckIfFieldsAvailableForCond()

static bool CheckIfFieldsAvailableForCond ( Item *  item, table_map  build_tables, table_map  probe_tables  )

static

clear_fields()

bool JOIN::clear_fields

(

table_map *

save_nullinfo

)

Set all column values from all input tables to NULL.

This is used when no rows are found during grouping: for FROM clause, a result row of all NULL values will be output; then SELECT list expressions get evaluated. E.g. SUM() will be NULL (the special "clear" value) and thus SUM() IS NULL will be true.

Note

Setting field values for input tables is a destructive operation, since it overwrite the NULL value flags with 1 bits. Rows from const tables are never re-read, hence their NULL value flags must be saved by this function and later restored by JOIN::restore_fields(). This is generally not necessary for non-const tables, since field values are overwritten when new rows are read.

Parameters

[out]

save_nullinfo

Map of tables whose fields were set to NULL, and for which NULL values must be restored. Should be set to all zeroes on entry to function.

Returns

false if success, true if error

cmp_field_value()

static bool cmp_field_value ( Field *  field, ptrdiff_t  diff  )

static

compute_ria_idx()

static size_t compute_ria_idx ( const mem_root_deque< Item * > &  fields, size_t  i, size_t  added_non_hidden_fields, size_t  border  )

static

Compute the position mapping from fields to ref_item_array, cf.

detailed explanation in change_to_use_tmp_fields_except_sums

ConditionIsAlwaysTrue()

static bool ConditionIsAlwaysTrue ( Item *  item )

static

ConnectJoins()

AccessPath * ConnectJoins	(	plan_idx	upper_first_idx,
		plan_idx	first_idx,
		plan_idx	last_idx,
		QEP_TAB *	qep_tabs,
		THD *	thd,
		CallingContext	calling_context,
		vector< PendingCondition > *	pending_conditions,
		vector< PendingInvalidator > *	pending_invalidators,
		vector< PendingCondition > *	pending_join_conditions,
		qep_tab_map *	unhandled_duplicates,
		table_map *	conditions_depend_on_outer_tables
	)

For a given slice of the table list, build up the iterator tree corresponding to the tables in that slice.

It handles inner and outer joins, as well as semijoins (“first match”).

The join tree in MySQL is generally a left-deep tree of inner joins, so we can start at the left, make an inner join against the next table, join the result of that against the next table, etc.. However, a given sub-slice of the table list can be designated as an outer join, by setting first_inner() and last_inner() on the first table of said slice. (It is also set in some, but not all, of the other tables in the slice.) If so, we call ourselves recursively with that slice, put it as the right (inner) arm of an outer join, and then continue with our inner join.

Similarly, if a table N has set “first match” to table M (ie., jump back to table M whenever we see a non-filtered record in table N), then there is a subslice from [M+1,N] that we need to process recursively before putting it as the right side of a semijoin. Every semijoin can be implemented with a LIMIT 1, but for clarity and performance, we prefer to use a NestedLoopJoin with a special SEMI join type whenever possible. Sometimes, we have no choice, though (see the comments below). Note that we cannot use first_sj_inner() for detecting semijoins, as it is not updated when tables are reordered by the join optimizer. Outer joins and semijoins can nest, so we need to take some care to make sure that we pick the outermost structure to recurse on.

Conditions are a bit tricky. Conceptually, SQL evaluates conditions only after all tables have been joined; however, for efficiency reasons, we want to evaluate them as early as possible. As long as we are only dealing with inner joins, this is as soon as we've read all tables participating in the condition, but for outer joins, we need to wait until the join has happened. See pending_conditions below.

Parameters

	upper_first_idx	gives us the first table index of the other side of the join. Only valid if we are inside a substructure (outer join, semijoin or antijoin). I.e., if we are processing the right side of the query 't1 LEFT JOIN t2', upper_first_idx gives us the table index of 't1'. Used by hash join to determine the table map for each side of the join.
	first_idx	index of the first table in the slice we are creating a tree for (inclusive)
	last_idx	index of the last table in the slice we are creating a tree for (exclusive)
	qep_tabs	the full list of tables we are joining
	thd	the THD to allocate the iterators on
	calling_context	what situation we have immediately around is in the tree (ie., whether we are called to resolve the inner part of an outer join, a semijoin, etc.); mostly used to avoid infinite recursion where we would process e.g. the same semijoin over and over again
	pending_conditions	if nullptr, we are not at the right (inner) side of any outer join and can evaluate conditions immediately. If not, we need to push any WHERE predicates to that vector and evaluate them only after joins.
	pending_invalidators	a global list of CacheInvalidatorIterators we need to emit, but cannot yet due to pending outer joins. Note that unlike pending_conditions and pending_join_conditions, this is never nullptr, and is always the same pointer when recursing within the same JOIN.
	pending_join_conditions	if not nullptr, we are at the inner side of semijoin/antijoin. The join iterator is created at the outer side, so any join conditions at the inner side needs to be pushed to this vector so that they can be attached to the join iterator. Note that this is currently only used by hash join.
[out]	unhandled_duplicates	list of tables we should have deduplicated using duplicate weedout, but could not; append-only.
[out]	conditions_depend_on_outer_tables	For each condition we have applied on the inside of these iterators, their dependent tables are appended to this set. Thus, if conditions_depend_on_outer_tables contain something from outside the tables covered by [first_idx,last_idx) (ie., after translation from QEP_TAB indexes to table indexes), we cannot use a hash join, since the returned iterator depends on seeing outer rows when evaluating its conditions.

construct_lookup()

bool construct_lookup	(	THD *	thd,
		TABLE *	table,
		Index_lookup *	ref
	)

Copy the lookup key into the table ref's key buffer.

Parameters

thd	pointer to the THD object
table	the table to read
ref	information about the index lookup key

Return values

false	ref key copied successfully
true	error detected during copying of key

ContainsAnyMRRPaths()

static bool ContainsAnyMRRPaths ( AccessPath *  path )

static

See if “path” has any MRR nodes; if so, we cannot optimize them away in PossiblyAttachFilter(), as the BKA iterator expects there to be a corresponding MRR iterator.

(This is a very rare case, so all we care about is that it should not crash.)

ConvertItemsToCopy()

void ConvertItemsToCopy	(	const mem_root_deque< Item * > &	items,
		Field **	fields,
		Temp_table_param *	param
	)

For historical reasons, derived table materialization and temporary table materialization didn't specify the fields to materialize in the same way.

Temporary table materialization used copy_funcs() to get the data into the Field pointers of the temporary table to be written, storing the lists in items_to_copy. (Originally, there was also copy_fields(), but it is no longer used for this purpose.)

However, derived table materialization used JOIN::fields (which is a set of Item, not Field!) for the same purpose, calling fill_record() (which originally was meant for INSERT and UPDATE) instead. Thus, we have to rewrite one to the other, so that we can have only one MaterializeIterator. We choose to rewrite JOIN::fields to items_to_copy.

TODO: The optimizer should output just one kind of structure directly.

ConvertQepTabMapToTableMap()

static table_map ConvertQepTabMapToTableMap ( JOIN *  join, qep_tab_map  tables  )

static

copy_fields()

bool copy_fields	(	Temp_table_param *	param,
		const THD *	thd,
		bool	reverse_copy
	)

Make a copy of all simple SELECT'ed fields.

This is used in window functions, to copy fields to and from the frame buffer. (It used to be used in materialization, but now that is entirely done by copy_funcs(), even for Item_field.)

Parameters

param	Represents the current temporary file being produced
thd	The current thread
reverse_copy	If true, copies fields back from the frame buffer tmp table to the output table's buffer, cf. bring_back_frame_row.

Returns

false if OK, true on error.

copy_funcs()

bool copy_funcs	(	Temp_table_param *	param,
		const THD *	thd,
		Copy_func_type	type
	)

Copy result of functions to record in tmp_table.

Uses the thread pointer to check for errors in some of the val_xxx() methods called by the save_in_result_field() function. TODO: make the Item::val_xxx() return error code

Parameters

param	Copy functions of tmp table specified by param
thd	pointer to the current thread for error checking
type	type of function Items that need to be copied (used w.r.t windowing functions).

Return values

false	if OK
true	on error

create_access_paths()

void JOIN::create_access_paths ( )

private

Convert the executor structures to a set of access paths, storing the result in m_root_access_path.

create_access_paths_for_index_subquery()

void JOIN::create_access_paths_for_index_subquery ( )

private

create_intermediate_table()

bool JOIN::create_intermediate_table ( QEP_TAB *  tab, const mem_root_deque< Item * > &  tmp_table_fields, ORDER_with_src &  tmp_table_group, bool  save_sum_fields  )

private

Create a temporary table to be used for processing DISTINCT/ORDER BY/GROUP BY.

Note

Will modify JOIN object wrt sort/group attributes

Parameters

tab	the JOIN_TAB object to attach created table to
tmp_table_fields	List of items that will be used to define column types of the table.
tmp_table_group	Group key to use for temporary table, empty if none.
save_sum_fields	If true, do not replace Item_sum items in tmp_fields list with Item_field items referring to fields in temporary table.

Returns

false on success, true on failure

If this is a window's OUT table, any final DISTINCT, ORDER BY will lead to windows showing use of tmp table in the final windowing step, so no need to signal use of tmp table unless we are here for another tmp table.

create_root_access_path_for_join()

AccessPath * JOIN::create_root_access_path_for_join ( )

private

Helpers for create_access_paths.

CreateBKAAccessPath()

AccessPath * CreateBKAAccessPath	(	THD *	thd,
		JOIN *	join,
		AccessPath *	outer_path,
		qep_tab_map	left_tables,
		AccessPath *	inner_path,
		qep_tab_map	right_tables,
		TABLE *	table,
		Table_ref *	table_list,
		Index_lookup *	ref,
		JoinType	join_type
	)

CreateConjunction()

Item * CreateConjunction

(

List< Item > *

items

)

Create an AND conjunction of all given items.

If there are no items, returns nullptr. If there's only one item, returns that item.

CreateHashJoinAccessPath()

static AccessPath * CreateHashJoinAccessPath ( THD *  thd, QEP_TAB *  qep_tab, AccessPath *  build_path, qep_tab_map  build_tables, AccessPath *  probe_path, qep_tab_map  probe_tables, JoinType  join_type, vector< Item * > *  join_conditions, table_map *  conditions_depend_on_outer_tables  )

static

CreateNestedLoopAccessPath()

AccessPath * CreateNestedLoopAccessPath	(	THD *	thd,
		AccessPath *	outer,
		AccessPath *	inner,
		JoinType	join_type,
		bool	pfs_batch_mode
	)

CreateWeedoutOrLimitAccessPath()

static AccessPath * CreateWeedoutOrLimitAccessPath ( THD *  thd, AccessPath *  path, SJ_TMP_TABLE *  weedout_table  )

static

do_sj_dups_weedout()

int do_sj_dups_weedout	(	THD *	thd,
		SJ_TMP_TABLE *	sjtbl
	)

SemiJoinDuplicateElimination: Weed out duplicate row combinations.

SYNPOSIS do_sj_dups_weedout() thd Thread handle sjtbl Duplicate weedout table

DESCRIPTION Try storing current record combination of outer tables (i.e. their rowids) in the temporary table. This records the fact that we've seen this record combination and also tells us if we've seen it before.

RETURN -1 Error 1 The row combination is a duplicate (discard it) 0 The row combination is not a duplicate (continue)

ExtractConditions()

bool ExtractConditions	(	Item *	condition,
		Mem_root_array< Item * > *	condition_parts
	)

Split AND conditions into their constituent parts, recursively.

Conditions that are not AND conditions are appended unchanged onto condition_parts. E.g. if you have ((a AND b) AND c), condition_parts will contain [a, b, c], plus whatever it contained before the call.

ExtractJoinConditions()

static void ExtractJoinConditions ( const QEP_TAB *  current_table, vector< Item * > *  predicates, vector< Item * > *  join_conditions  )

static

find_rollup_item_in_group_list()

static Item_rollup_group_item * find_rollup_item_in_group_list ( Item *  item, Query_block *  query_block  )

static

FindSubstructure()

static Substructure FindSubstructure ( QEP_TAB *  qep_tabs, const plan_idx  first_idx, const plan_idx  this_idx, const plan_idx  last_idx, CallingContext  calling_context, bool *  add_limit_1, plan_idx *  substructure_end, qep_tab_map *  unhandled_duplicates  )

static

Given a range of tables (where we assume that we've already handled first_idx..(this_idx-1) as inner joins), figure out whether this is a semijoin, an outer join or a weedout.

In general, the outermost structure wins; if we are in one of the rare cases where there are e.g. coincident (first match) semijoins and weedouts, we do various forms of conflict resolution:

• Unhandled weedouts will add elements to unhandled_duplicates (to be handled at the top level of the query).
• Unhandled semijoins will either: Set add_limit_1 to true, which means a LIMIT 1 iterator should be added, or Add elements to unhandled_duplicates in situations that cannot be solved by a simple one-table, one-row LIMIT.

If not returning NONE, substructure_end will also be filled with where this sub-join ends (exclusive).

FinishPendingOperations()

AccessPath * FinishPendingOperations	(	THD *	thd,
		AccessPath *	path,
		QEP_TAB *	remove_duplicates_loose_scan_qep_tab,
		const vector< PendingCondition > &	pending_conditions,
		table_map *	conditions_depend_on_outer_tables
	)

get_end_select_func()

QEP_TAB::enum_op_type JOIN::get_end_select_func

(

)

Rows produced by a join sweep may end up in a temporary table or be sent to a client. Setup the function of the nested loop join algorithm which handles final fully constructed and matched records.

Returns

end_query_block function to use. This function can't fail.

get_update_or_delete_target_tables()

static table_map get_update_or_delete_target_tables ( const JOIN *  join )

static

GetAccessPathForDerivedTable() [1/2]

AccessPath * GetAccessPathForDerivedTable	(	THD *	thd,
		QEP_TAB *	qep_tab,
		AccessPath *	table_path
	)

GetAccessPathForDerivedTable() [2/2]

AccessPath * GetAccessPathForDerivedTable	(	THD *	thd,
		Table_ref *	table_ref,
		TABLE *	table,
		bool	rematerialize,
		Mem_root_array< const AccessPath * > *	invalidators,
		bool	need_rowid,
		AccessPath *	table_path
	)

GetInnermostCondition()

static Item * GetInnermostCondition ( Item *  item )

static

Returns

the innermost condition of a nested trigger condition. If the item is not a trigger condition, the item itself is returned.

GetTableAccessPath()

AccessPath * GetTableAccessPath	(	THD *	thd,
		QEP_TAB *	qep_tab,
		QEP_TAB *	qep_tabs
	)

Get the RowIterator used for scanning the given table, with any required materialization operations done first.

GetTriggerCondOrNull()

static Item_func_trig_cond * GetTriggerCondOrNull ( Item *  item )

static

group_rec_cmp()

static bool group_rec_cmp ( ORDER *  group, uchar *  rec0, uchar *  rec1  )

static

Compare GROUP BY in from tmp table's record[0] and record[1].

Returns

true records are different false records are the same

has_rollup_result()

bool has_rollup_result

(

Item *

item

)

Checks if an item has a ROLLUP NULL which needs to be written to temp table.

Parameters

item	Item for which we need to detect if ROLLUP NULL has to be written.

Returns

false if ROLLUP NULL need not be written for this item. true if it has to be written.

init_tmptable_sum_functions()

void init_tmptable_sum_functions

(

Item_sum **

func_ptr

)

InsideOuterOrAntiJoin()

static bool InsideOuterOrAntiJoin ( QEP_TAB *  qep_tab )

static

is_rollup_group_wrapper()

bool is_rollup_group_wrapper

(

const Item *

item

)

IsJoinCondition()

static bool IsJoinCondition ( const Item *  item, const QEP_TAB *  qep_tab  )

static

Parameters

item	The item we want to see if is a join condition.
qep_tab	The table we are joining in.

Returns

true if 'item' is a join condition for a join involving the given table (both equi-join and non-equi-join condition).

IsTableScan()

static bool IsTableScan ( AccessPath *  path )

static

join_read_const_table()

int join_read_const_table	(	JOIN_TAB *	tab,
		POSITION *	pos
	)

Reads content of constant table.

Parameters

tab	table
pos	position of table in query plan

Return values

0	ok, one row was found or one NULL-complemented row was created
-1	ok, no row was found and no NULL-complemented row was created
1	error

make_group_fields()

bool make_group_fields	(	JOIN *	main_join,
		JOIN *	curr_join
	)

allocate group fields or take prepared (cached).

Parameters

main_join	join of current select
curr_join	current join (join of current select or temporary copy of it)

Return values

0	ok
1	failed

MarkUnhandledDuplicates()

static void MarkUnhandledDuplicates ( SJ_TMP_TABLE *  weedout, plan_idx  weedout_start, plan_idx  weedout_end, qep_tab_map *  unhandled_duplicates  )

static

For a given duplicate weedout operation, figure out which tables are supposed to be deduplicated by it, and add those to unhandled_duplicates.

(SJ_TMP_TABLE contains the deduplication key, which is exactly the complement of the tables to be deduplicated.)

MoveCompositeIteratorsFromTablePath()

AccessPath * MoveCompositeIteratorsFromTablePath	(	AccessPath *	path,
		const Query_block &	query_block
	)

For a MATERIALIZE access path, move any non-basic iterators (e.g.

sorts and filters) from table_path to above the path, for easier EXPLAIN and generally simpler structure. Note the assert in CreateIteratorFromAccessPath() that we succeeded. (ALTERNATIVE counts as a basic iterator in this regard.)

We do this by finding the second-bottommost access path, and inserting our materialize node as its child. The bottommost one becomes the actual table access path.

If a ZERO_ROWS access path is materialized, we simply replace the MATERIALIZE path with the ZERO_ROWS path, since there is nothing to materialize.

Parameters

path	the MATERIALIZE path.
query_block	The query block in which 'path' belongs.

Returns

The new root of the set of AccessPaths formed by 'path' and its descendants.

NewInvalidatorAccessPathForTable()

static AccessPath * NewInvalidatorAccessPathForTable ( THD *  thd, AccessPath *  path, QEP_TAB *  qep_tab, plan_idx  table_index_to_invalidate  )

static

NewWeedoutAccessPathForTables()

static AccessPath * NewWeedoutAccessPathForTables ( THD *  thd, const qep_tab_map  tables_to_deduplicate, QEP_TAB *  qep_tabs, uint  primary_tables, AccessPath *  path  )

static

optimize_distinct()

void JOIN::optimize_distinct ( )

private

Optimize distinct when used on a subset of the tables.

E.g.,: SELECT DISTINCT t1.a FROM t1,t2 WHERE t1.b=t2.b In this case we can stop scanning t2 when we have found one t1.a

pfs_batch_update()

bool QEP_TAB::pfs_batch_update

(

const JOIN *

join

)

const

PickOutConditionsForTableIndex() [1/2]

void PickOutConditionsForTableIndex	(	int	table_idx,
		vector< PendingCondition > *	from,
		vector< Item * > *	to
	)

PickOutConditionsForTableIndex() [2/2]

void PickOutConditionsForTableIndex	(	int	table_idx,
		vector< PendingCondition > *	from,
		vector< PendingCondition > *	to
	)

PossiblyAttachFilter() [1/2]

AccessPath * PossiblyAttachFilter	(	AccessPath *	path,
		const vector< Item * > &	conditions,
		THD *	thd,
		table_map *	conditions_depend_on_outer_tables
	)

Return a new iterator that wraps "iterator" and that tests all of the given conditions (if any), ANDed together.

If there are no conditions, just return the given iterator back.

PossiblyAttachFilter() [2/2]

static AccessPath * PossiblyAttachFilter ( AccessPath *  path, const vector< PendingCondition > &  conditions, THD *  thd, table_map *  conditions_depend_on_outer_tables  )

static

prepare_sum_aggregators()

bool prepare_sum_aggregators	(	Item_sum **	sum_funcs,
		bool	need_distinct
	)

QueryMixesOuterBKAAndBNL()

static bool QueryMixesOuterBKAAndBNL ( JOIN *  join )

static

read_const()

int read_const	(	TABLE *	table,
		Index_lookup *	ref
	)

read_system()

static int read_system ( TABLE *  table )

static

Read a constant table when there is at most one matching row, using a table scan.

Parameters

table

Table to read

Return values

0	Row was found
-1	Row was not found
1	Got an error (other than row not found) during read

RecalculateTablePathCost()

static void RecalculateTablePathCost ( AccessPath *  path, const Query_block &  outer_query_block  )

static

Recalculate the cost of 'path'.

Parameters

path	the access path for which we update the cost numbers.
outer_query_block	the query block to which 'path belongs.

record_prefix_size()

static size_t record_prefix_size ( const TABLE *  table )

static

Find out how many bytes it takes to store the smallest prefix which covers all the columns that will be read from a table.

Parameters

table

the table to read

Returns

the size of the smallest prefix that covers all records to be read from the table

RefToString()

string RefToString	(	const Index_lookup &	ref,
		const KEY *	key,
		bool	include_nulls
	)

replace_contents_of_rollup_wrappers_with_tmp_fields()

bool replace_contents_of_rollup_wrappers_with_tmp_fields	(	THD *	thd,
		Query_block *	select,
		Item *	item_arg
	)

For each rollup wrapper below the given item, replace its argument with a temporary field, e.g.

1 + rollup_group_item(a) -> 1 + rollup_group_item(<temporary>.a).

Which temporary field to use is found by looking at the Query_block's group items, and looking up their (previously set) result fields.

replace_embedded_rollup_references_with_tmp_fields()

static bool replace_embedded_rollup_references_with_tmp_fields ( THD *  thd, Item *  item, mem_root_deque< Item * > *  fields  )

static

For each rollup wrapper below the given item, replace it with a temporary field, e.g.

1 + rollup_group_item(a) -> 1 + <temporary>.rollup_group_item(a)

Which temporary field to use is found by looking at the other fields; the rollup_group_item should already exist earlier in the list (and having a temporary table field set up), simply by virtue of being a group item.

report_handler_error()

int report_handler_error	(	TABLE *	table,
		int	error
	)

Help function when we get some an error from the table handler.

restore_fields()

void JOIN::restore_fields

(

table_map

save_nullinfo

)

Restore all result fields for all tables specified in save_nullinfo.

Parameters

save_nullinfo

Set of tables for which restore is necessary.

Note

Const tables must have their NULL value flags restored,

See also

JOIN::clear_fields().

set_record_buffer()

bool set_record_buffer	(	TABLE *	table,
		double	expected_rows_to_fetch
	)

Allocate a data buffer that the storage engine can use for fetching batches of records.

A buffer is only allocated if ha_is_record_buffer_wanted() returns true for the handler, and the scan in question is of a kind that could be expected to benefit from fetching records in batches.

Parameters

table	the table to read
expected_rows_to_fetch	number of rows the optimizer thinks we will be reading out of the table

Return values

true	if an error occurred when allocating the buffer
false	if a buffer was successfully allocated, or if a buffer was not attempted allocated

SetCostOnHashJoinAccessPath()

void SetCostOnHashJoinAccessPath	(	const Cost_model_server &	cost_model,
		const POSITION *	pos_outer,
		AccessPath *	path
	)

SetCostOnNestedLoopAccessPath()

void SetCostOnNestedLoopAccessPath	(	const Cost_model_server &	cost_model,
		const POSITION *	pos_inner,
		AccessPath *	path
	)

SetCostOnTableAccessPath()

void SetCostOnTableAccessPath	(	const Cost_model_server &	cost_model,
		const POSITION *	pos,
		bool	is_after_filter,
		AccessPath *	path
	)

setup_sum_funcs()

bool setup_sum_funcs	(	THD *	thd,
		Item_sum **	func_ptr
	)

Call setup() for all sum functions.

Parameters

thd	thread handler
func_ptr	sum function list

Return values

false	ok
true	error

setup_tmptable_write_func()

void setup_tmptable_write_func	(	QEP_TAB *	tab,
		Opt_trace_object *	trace
	)

Setup write_func of QEP_tmp_table object.

Parameters

tab	QEP_TAB of a tmp table
trace	Opt_trace_object to add to

Function sets up write_func according to how QEP_tmp_table object that is attached to the given join_tab will be used in the query.

SplitConditions()

void SplitConditions	(	Item *	condition,
		QEP_TAB *	current_table,
		vector< Item * > *	predicates_below_join,
		vector< PendingCondition > *	predicates_above_join,
		vector< PendingCondition > *	join_conditions,
		plan_idx	semi_join_table_idx,
		qep_tab_map	left_tables
	)

table_rec_cmp()

static bool table_rec_cmp ( TABLE *  table )

static

Compare GROUP BY in from tmp table's record[0] and record[1].

Returns

true records are different false records are the same

unique_hash()

ulonglong unique_hash	(	const Field *	field,
		ulonglong *	hash_val
	)

Generate hash for a field.

Returns

generated hash

unique_hash_fields()

static ulonglong unique_hash_fields ( TABLE *  table )

static

Generate hash for unique_constraint for all visible fields of a table.

Parameters

table

the table for which we want a hash of its fields

Returns

the hash value

unique_hash_group()

static ulonglong unique_hash_group ( ORDER *  group )

static

Generate hash for unique constraint according to group-by list.

This reads the values of the GROUP BY expressions from fields so assumes those expressions have been computed and stored into fields of a temporary table; in practice this means that copy_funcs() must have been called.

unwrap_rollup_group()

Item * unwrap_rollup_group

(

Item *

item

)

update_const_equal_items()

static bool update_const_equal_items ( THD *  thd, Item *  cond, JOIN_TAB *  tab  )

static

Check appearance of new constant items in multiple equalities of a condition after reading a constant table.

The function retrieves the cond condition and for each encountered multiple equality checks whether new constants have appeared after reading the constant (single row) table tab. If so it adjusts the multiple equality appropriately.

Parameters

thd	thread handler
cond	condition whose multiple equalities are to be checked
tab	constant table that has been read

update_item_cache_if_changed()

int update_item_cache_if_changed

(

List< Cached_item > &

list

)

update_tmptable_sum_func()

void update_tmptable_sum_func	(	Item_sum **	func_ptr,
		TABLE *	tmp_table
	)

Update record 0 in tmp_table from record 1.

use_order()

bool QEP_TAB::use_order

(

)

const

Use ordering provided by chosen index?

Check if access to this JOIN_TAB has to retrieve rows in sorted order as defined by the ordered index used to access this table.

UseBKA()

static bool UseBKA ( QEP_TAB *  qep_tab )

static

UseHashJoin()

static bool UseHashJoin ( QEP_TAB *  qep_tab )

static

Variable Documentation

MAX_RECORD_BUFFER_SIZE

constexpr size_t MAX_RECORD_BUFFER_SIZE = 128 * 1024

staticconstexpr

Maximum amount of space (in bytes) to allocate for a Record_buffer.