mariadb/sql/opt_index_cond_pushdown.cc

447 lines
14 KiB
C++
Raw Normal View History

/*
Copyright (c) 2009, 2012, Monty Program Ab
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; version 2 of the License.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */
#include "sql_select.h"
2011-04-25 17:22:25 +02:00
#include "sql_test.h"
/****************************************************************************
* Index Condition Pushdown code starts
***************************************************************************/
/*
Check if given expression uses only table fields covered by the given index
SYNOPSIS
uses_index_fields_only()
item Expression to check
tbl The table having the index
keyno The index number
other_tbls_ok TRUE <=> Fields of other non-const tables are allowed
DESCRIPTION
Check if given expression only uses fields covered by index #keyno in the
table tbl. The expression can use any fields in any other tables.
The expression is guaranteed not to be AND or OR - those constructs are
handled outside of this function.
RETURN
TRUE Yes
FALSE No
*/
bool uses_index_fields_only(Item *item, TABLE *tbl, uint keyno,
bool other_tbls_ok)
{
if (item->walk(&Item::limit_index_condition_pushdown_processor, FALSE, NULL))
{
return FALSE;
}
if (item->const_item())
return TRUE;
/*
Don't push down the triggered conditions. Nested outer joins execution
code may need to evaluate a condition several times (both triggered and
untriggered), and there is no way to put thi
TODO: Consider cloning the triggered condition and using the copies for:
1. push the first copy down, to have most restrictive index condition
possible
2. Put the second copy into tab->select_cond.
*/
if (item->type() == Item::FUNC_ITEM &&
((Item_func*)item)->functype() == Item_func::TRIG_COND_FUNC)
return FALSE;
if (!(item->used_tables() & tbl->map))
return other_tbls_ok;
Item::Type item_type= item->type();
switch (item_type) {
case Item::FUNC_ITEM:
{
/* This is a function, apply condition recursively to arguments */
Item_func *item_func= (Item_func*)item;
Item **child;
Item **item_end= (item_func->arguments()) + item_func->argument_count();
for (child= item_func->arguments(); child != item_end; child++)
{
if (!uses_index_fields_only(*child, tbl, keyno, other_tbls_ok))
return FALSE;
}
return TRUE;
}
case Item::COND_ITEM:
{
/*
This is a AND/OR condition. Regular AND/OR clauses are handled by
make_cond_for_index() which will chop off the part that can be
checked with index. This code is for handling non-top-level AND/ORs,
e.g. func(x AND y).
*/
List_iterator<Item> li(*((Item_cond*)item)->argument_list());
Item *item;
while ((item=li++))
{
if (!uses_index_fields_only(item, tbl, keyno, other_tbls_ok))
return FALSE;
}
return TRUE;
}
case Item::FIELD_ITEM:
{
Item_field *item_field= (Item_field*)item;
Field *field= item_field->field;
if (field->table != tbl)
return TRUE;
/*
The below is probably a repetition - the first part checks the
other two, but let's play it safe:
*/
if(!field->part_of_key.is_set(keyno) ||
field->type() == MYSQL_TYPE_GEOMETRY ||
field->type() == MYSQL_TYPE_BLOB)
return FALSE;
KEY *key_info= tbl->key_info + keyno;
KEY_PART_INFO *key_part= key_info->key_part;
KEY_PART_INFO *key_part_end= key_part + key_info->key_parts;
for ( ; key_part < key_part_end; key_part++)
{
if (field->eq(key_part->field))
return !(key_part->key_part_flag & HA_PART_KEY_SEG);
}
if ((tbl->file->ha_table_flags() & HA_PRIMARY_KEY_IN_READ_INDEX) &&
tbl->s->primary_key != MAX_KEY &&
tbl->s->primary_key != keyno)
{
key_info= tbl->key_info + tbl->s->primary_key;
key_part= key_info->key_part;
key_part_end= key_part + key_info->key_parts;
for ( ; key_part < key_part_end; key_part++)
{
/*
It does not make sense to use the fact that the engine can read in
a full field if the key if the index is built only over a part
of this field.
*/
if (field->eq(key_part->field))
return !(key_part->key_part_flag & HA_PART_KEY_SEG);
}
}
return FALSE;
}
case Item::REF_ITEM:
return uses_index_fields_only(item->real_item(), tbl, keyno,
other_tbls_ok);
default:
return FALSE; /* Play it safe, don't push unknown non-const items */
}
}
#define ICP_COND_USES_INDEX_ONLY 10
/*
Get a part of the condition that can be checked using only index fields
SYNOPSIS
make_cond_for_index()
cond The source condition
table The table that is partially available
keyno The index in the above table. Only fields covered by the index
are available
other_tbls_ok TRUE <=> Fields of other non-const tables are allowed
DESCRIPTION
Get a part of the condition that can be checked when for the given table
we have values only of fields covered by some index. The condition may
refer to other tables, it is assumed that we have values of all of their
fields.
Example:
make_cond_for_index(
"cond(t1.field) AND cond(t2.key1) AND cond(t2.non_key) AND cond(t2.key2)",
t2, keyno(t2.key1))
will return
"cond(t1.field) AND cond(t2.key2)"
RETURN
Index condition, or NULL if no condition could be inferred.
*/
Item *make_cond_for_index(Item *cond, TABLE *table, uint keyno,
bool other_tbls_ok)
{
if (!cond)
return NULL;
if (cond->type() == Item::COND_ITEM)
{
uint n_marked= 0;
if (((Item_cond*) cond)->functype() == Item_func::COND_AND_FUNC)
{
table_map used_tables= 0;
Item_cond_and *new_cond=new Item_cond_and;
if (!new_cond)
return (COND*) 0;
List_iterator<Item> li(*((Item_cond*) cond)->argument_list());
Item *item;
while ((item=li++))
{
Item *fix= make_cond_for_index(item, table, keyno, other_tbls_ok);
if (fix)
{
new_cond->argument_list()->push_back(fix);
used_tables|= fix->used_tables();
}
if (test(item->marker == ICP_COND_USES_INDEX_ONLY))
{
n_marked++;
item->marker= 0;
}
}
if (n_marked ==((Item_cond*)cond)->argument_list()->elements)
cond->marker= ICP_COND_USES_INDEX_ONLY;
switch (new_cond->argument_list()->elements) {
case 0:
return (COND*) 0;
case 1:
new_cond->used_tables_cache= used_tables;
return new_cond->argument_list()->head();
default:
new_cond->quick_fix_field();
new_cond->used_tables_cache= used_tables;
return new_cond;
}
}
else /* It's OR */
{
Item_cond_or *new_cond=new Item_cond_or;
if (!new_cond)
return (COND*) 0;
List_iterator<Item> li(*((Item_cond*) cond)->argument_list());
Item *item;
while ((item=li++))
{
Item *fix= make_cond_for_index(item, table, keyno, other_tbls_ok);
if (!fix)
return (COND*) 0;
new_cond->argument_list()->push_back(fix);
if (test(item->marker == ICP_COND_USES_INDEX_ONLY))
{
n_marked++;
item->marker= 0;
}
}
if (n_marked ==((Item_cond*)cond)->argument_list()->elements)
cond->marker= ICP_COND_USES_INDEX_ONLY;
new_cond->quick_fix_field();
new_cond->used_tables_cache= ((Item_cond_or*) cond)->used_tables_cache;
new_cond->top_level_item();
return new_cond;
}
}
if (!uses_index_fields_only(cond, table, keyno, other_tbls_ok))
return (COND*) 0;
cond->marker= ICP_COND_USES_INDEX_ONLY;
return cond;
}
Item *make_cond_remainder(Item *cond, TABLE *table, uint keyno,
bool other_tbls_ok, bool exclude_index)
{
if (cond->type() == Item::COND_ITEM)
{
table_map tbl_map= 0;
if (((Item_cond*) cond)->functype() == Item_func::COND_AND_FUNC)
{
/* Create new top level AND item */
Item_cond_and *new_cond=new Item_cond_and;
if (!new_cond)
return (COND*) 0;
List_iterator<Item> li(*((Item_cond*) cond)->argument_list());
Item *item;
while ((item=li++))
{
Item *fix= make_cond_remainder(item, table, keyno,
other_tbls_ok, exclude_index);
if (fix)
{
new_cond->argument_list()->push_back(fix);
tbl_map |= fix->used_tables();
}
}
switch (new_cond->argument_list()->elements) {
case 0:
return (COND*) 0;
case 1:
return new_cond->argument_list()->head();
default:
new_cond->quick_fix_field();
((Item_cond*)new_cond)->used_tables_cache= tbl_map;
return new_cond;
}
}
else /* It's OR */
{
Item_cond_or *new_cond=new Item_cond_or;
if (!new_cond)
return (COND*) 0;
List_iterator<Item> li(*((Item_cond*) cond)->argument_list());
Item *item;
while ((item=li++))
{
Item *fix= make_cond_remainder(item, table, keyno,
other_tbls_ok, FALSE);
if (!fix)
return (COND*) 0;
new_cond->argument_list()->push_back(fix);
tbl_map |= fix->used_tables();
}
new_cond->quick_fix_field();
((Item_cond*)new_cond)->used_tables_cache= tbl_map;
new_cond->top_level_item();
return new_cond;
}
}
else
{
if (exclude_index &&
uses_index_fields_only(cond, table, keyno, other_tbls_ok))
return 0;
else
return cond;
}
}
/*
Try to extract and push the index condition
SYNOPSIS
push_index_cond()
tab A join tab that has tab->table->file and its condition
in tab->select_cond
keyno Index for which extract and push the condition
DESCRIPTION
Try to extract and push the index condition down to table handler
*/
void push_index_cond(JOIN_TAB *tab, uint keyno)
{
DBUG_ENTER("push_index_cond");
Item *idx_cond;
Backport of: timestamp: Thu 2011-12-01 15:12:10 +0100 Fix for Bug#13430436 PERFORMANCE DEGRADATION IN SYSBENCH ON INNODB DUE TO ICP When running sysbench on InnoDB there is a performance degradation due to index condition pushdown (ICP). Several of the queries in sysbench have a WHERE condition that the optimizer uses for executing these queries as range scans. The upper and lower limit of the range scan will ensure that the WHERE condition is fulfilled. Still, the WHERE condition is part of the queries' condition and if ICP is enabled the condition will be pushed down to InnoDB as an index condition. Due to the range scan's upper and lower limits ensure that the WHERE condition is fulfilled, the pushed index condition will not filter out any records. As a result the use of ICP for these queries results in a performance overhead for sysbench. This overhead comes from using resources for determining the part of the condition that can be pushed down to InnoDB and overhead in InnoDB for executing the pushed index condition. With the default configuration for sysbench the range scans will use the primary key. This is a clustered index in InnoDB. Using ICP on a clustered index provides the lowest performance benefit since the entire record is part of the clustered index and in InnoDB it has the highest relative overhead for executing the pushed index condition. The fix for removing the overhead ICP introduces when running sysbench is to disable use of ICP when the index used by the query is a clustered index. When WL#6061 is implemented this change should be re-evaluated.
2012-02-16 17:15:57 +01:00
/*
Backported the following from MySQL 5.6:
6. The index is not a clustered index. The performance improvement
of pushing an index condition on a clustered key is much lower
than on a non-clustered key. This restriction should be
re-evaluated when WL#6061 is implemented.
*/
if ((tab->table->file->index_flags(keyno, 0, 1) &
HA_DO_INDEX_COND_PUSHDOWN) &&
optimizer_flag(tab->join->thd, OPTIMIZER_SWITCH_INDEX_COND_PUSHDOWN) &&
tab->join->thd->lex->sql_command != SQLCOM_UPDATE_MULTI &&
tab->join->thd->lex->sql_command != SQLCOM_DELETE_MULTI &&
Backport of: timestamp: Thu 2011-12-01 15:12:10 +0100 Fix for Bug#13430436 PERFORMANCE DEGRADATION IN SYSBENCH ON INNODB DUE TO ICP When running sysbench on InnoDB there is a performance degradation due to index condition pushdown (ICP). Several of the queries in sysbench have a WHERE condition that the optimizer uses for executing these queries as range scans. The upper and lower limit of the range scan will ensure that the WHERE condition is fulfilled. Still, the WHERE condition is part of the queries' condition and if ICP is enabled the condition will be pushed down to InnoDB as an index condition. Due to the range scan's upper and lower limits ensure that the WHERE condition is fulfilled, the pushed index condition will not filter out any records. As a result the use of ICP for these queries results in a performance overhead for sysbench. This overhead comes from using resources for determining the part of the condition that can be pushed down to InnoDB and overhead in InnoDB for executing the pushed index condition. With the default configuration for sysbench the range scans will use the primary key. This is a clustered index in InnoDB. Using ICP on a clustered index provides the lowest performance benefit since the entire record is part of the clustered index and in InnoDB it has the highest relative overhead for executing the pushed index condition. The fix for removing the overhead ICP introduces when running sysbench is to disable use of ICP when the index used by the query is a clustered index. When WL#6061 is implemented this change should be re-evaluated.
2012-02-16 17:15:57 +01:00
tab->type != JT_CONST && tab->type != JT_SYSTEM &&
!(keyno == tab->table->s->primary_key && // (6)
tab->table->file->primary_key_is_clustered())) // (6)
{
DBUG_EXECUTE("where",
print_where(tab->select_cond, "full cond", QT_ORDINARY););
idx_cond= make_cond_for_index(tab->select_cond, tab->table, keyno,
tab->icp_other_tables_ok);
DBUG_EXECUTE("where",
print_where(idx_cond, "idx cond", QT_ORDINARY););
if (idx_cond)
{
Item *idx_remainder_cond= 0;
2010-11-25 18:17:28 +01:00
tab->pre_idx_push_select_cond= tab->select_cond;
/*
For BKA cache we store condition to special BKA cache field
because evaluation of the condition requires additional operations
before the evaluation. This condition is used in
JOIN_CACHE_BKA[_UNIQUE]::skip_index_tuple() functions.
*/
if (tab->use_join_cache &&
/*
if cache is used then the value is TRUE only
for BKA[_UNIQUE] cache (see check_join_cache_usage func).
*/
tab->icp_other_tables_ok &&
(idx_cond->used_tables() &
~(tab->table->map | tab->join->const_table_map)))
tab->cache_idx_cond= idx_cond;
else
idx_remainder_cond= tab->table->file->idx_cond_push(keyno, idx_cond);
/*
Disable eq_ref's "lookup cache" if we've pushed down an index
condition.
TODO: This check happens to work on current ICP implementations, but
there may exist a compliant implementation that will not work
correctly with it. Sort this out when we stabilize the condition
pushdown APIs.
*/
if (idx_remainder_cond != idx_cond)
tab->ref.disable_cache= TRUE;
Item *row_cond= tab->idx_cond_fact_out ?
make_cond_remainder(tab->select_cond, tab->table, keyno,
tab->icp_other_tables_ok, TRUE) :
tab->pre_idx_push_select_cond;
DBUG_EXECUTE("where",
print_where(row_cond, "remainder cond", QT_ORDINARY););
if (row_cond)
{
if (!idx_remainder_cond)
tab->select_cond= row_cond;
else
{
COND *new_cond= new Item_cond_and(row_cond, idx_remainder_cond);
tab->select_cond= new_cond;
tab->select_cond->quick_fix_field();
((Item_cond_and*)tab->select_cond)->used_tables_cache=
row_cond->used_tables() | idx_remainder_cond->used_tables();
}
}
else
tab->select_cond= idx_remainder_cond;
if (tab->select)
{
DBUG_EXECUTE("where",
print_where(tab->select->cond,
"select_cond",
QT_ORDINARY););
tab->select->cond= tab->select_cond;
tab->select->pre_idx_push_select_cond= tab->pre_idx_push_select_cond;
}
}
}
DBUG_VOID_RETURN;
}