database/mariadb
1
0
Fork 0
mirror of https://github.com/MariaDB/server.git synced 2025-12-24 11:21:21 +03:00
Code Activity

fix for bug #1724 'WHERE ... IN() optimizer behaviour

Browse Source 
has changed since 4.0.14'
We need to calculate cost of RANGE
scan if it is present instead of cost of 
FULL scan.


mysql-test/r/order_by.result:
  more accurate row estimation
  for RANGE scan
mysql-test/r/range.result:
  added test case for bug #1724
  'WHERE ... IN() optimizer behaviour has changed since 4.0.14'
mysql-test/r/select.result:
  please ignore
mysql-test/t/range.test:
  added test case for bug #1724
  'WHERE ... IN() optimizer behaviour has changed since 4.0.14'
sql/sql_select.cc:
  fix for bug #1724 'WHERE ... IN() optimizer behaviour 
  has changed since 4.0.14'
  We need to calculate cost of RANGE
  scan instead of cost of FULL scan
  if RANGE is present
  Few comments cleaned up.
This commit is contained in:
unknown
2003-11-03 19:56:01 +03:00
parent 15cc0ad2ec
commit 22bc137c2e
5 changed files with 87 additions and 28 deletions
Download Patch File Download Diff File Expand all files Collapse all files

65
sql/sql_select.cc
View File

@@ -2145,8 +2145,6 @@ find_best(JOIN *join,table_map rest_tables,uint idx,double record_count,
!(s->table->force_index && best_key))
{ // Check full join
ha_rows rnd_records= s->found_records;
/* Estimate cost of reading table. */
tmp= s->table->file->scan_time();
/*
If there is a restriction on the table, assume that 25% of the
rows can be skipped on next part.
@@ -2156,36 +2154,57 @@ find_best(JOIN *join,table_map rest_tables,uint idx,double record_count,
if (found_constraint)
rnd_records-= rnd_records/4;
if (s->on_expr) // Can't use join cache
/*
Range optimizer never proposes a RANGE if it isn't better
than FULL: so if RANGE is present, it's always preferred to FULL.
Here we estimate its cost.
*/
if (s->quick)
{
/*
For each record we:
- read record range through 'quick'
- skip rows which does not satisfy WHERE constraints
*/
tmp= record_count *
/* We have to read the whole table for each record */
(tmp +
/*
And we have to skip rows which does not satisfy join
condition for each record.
*/
(s->records - rnd_records)/(double) TIME_FOR_COMPARE);
(s->quick->read_time +
(s->found_records - rnd_records)/(double) TIME_FOR_COMPARE);
}
else
{
/* We read the table as many times as join buffer becomes full. */
tmp*= (1.0 + floor((double) cache_record_length(join,idx) *
record_count /
(double) thd->variables.join_buff_size));
/*
We don't make full cartesian product between rows in the scanned
table and existing records because we skip all rows from the
scanned table, which does not satisfy join condition when
we read the table (see flush_cached_records for details). Here we
take into account cost to read and skip these records.
*/
tmp+= (s->records - rnd_records)/(double) TIME_FOR_COMPARE;
/* Estimate cost of reading table. */
tmp= s->table->file->scan_time();
if (s->on_expr) // Can't use join cache
{
/*
For each record we have to:
- read the whole table record
- skip rows which does not satisfy join condition
*/
tmp= record_count *
(tmp +
(s->records - rnd_records)/(double) TIME_FOR_COMPARE);
}
else
{
/* We read the table as many times as join buffer becomes full. */
tmp*= (1.0 + floor((double) cache_record_length(join,idx) *
record_count /
(double) thd->variables.join_buff_size));
/*
We don't make full cartesian product between rows in the scanned
table and existing records because we skip all rows from the
scanned table, which does not satisfy join condition when
we read the table (see flush_cached_records for details). Here we
take into account cost to read and skip these records.
*/
tmp+= (s->records - rnd_records)/(double) TIME_FOR_COMPARE;
}
}
/*
We estimate the cost of evaluating WHERE clause for found records
as record_count * rnd_records + TIME_FOR_COMPARE. This cost plus
as record_count * rnd_records / TIME_FOR_COMPARE. This cost plus
tmp give us total cost of using TABLE SCAN
*/
if (best == DBL_MAX ||