[MDEV-6041] ORDER BY+subqueries: subquery_table.key=outer_table.col is not recongized as binding - Jira

Details

Type: Bug
Status: Closed (View Workflow)
Priority: Major
Resolution: Fixed
Affects Version/s: 5.3.12, 5.5.36, 10.0.10
Fix Version/s: 10.0.11
Component/s: None
Labels:
- upstream
- upstream-5.5

Description

Create a test dataset:

create table t1 (a int);

insert into t1 select * from test.one_k;

create table tsubq(

  id int primary key,

  key1 int,

  col1 int,

  key(key1)

) engine=innodb;

insert into tsubq

  select A.a + B.a*1000, A.a, 123456 from test.one_k A, test.one_k B;

Then, check the plan:

explain select

   (SELECT

      concat(id, '-', key1, '-', col1)

    FROM tsubq

    WHERE tsubq.key1 = t1.a

    ORDER BY tsubq.id ASC LIMIT 1)

from

t1;

+------+--------------------+-------+-------+---------------+---------+---------+------+------+-------------+

| id   | select_type        | table | type  | possible_keys | key     | key_len | ref  | rows | Extra       |

+------+--------------------+-------+-------+---------------+---------+---------+------+------+-------------+

|    1 | PRIMARY            | t1    | ALL   | NULL          | NULL    | NULL    | NULL | 1000 |             |

|    2 | DEPENDENT SUBQUERY | tsubq | index | key1          | PRIMARY | 4       | NULL |    1 | Using where |

+------+--------------------+-------+-------+---------------+---------+---------+------+------+-------------+

The subquery uses "index" access, which is very inefficient. The estimate for #rows seems to come from the LIMIT clause and is very wrong in this case.

The table is InnoDB (with extended keys). The index KEY(key1) is actually KEY(key1, id). The query has a restriction on key1 which makes it constant (tsubq.key1 = t1.a). After that, ORDER BY tsubq.id is achieved automatically.

The problem seems to be specifically with references to outside of subquery. If I use a constant instead, the query plan is able to use key1:

explain select     (SELECT        concat(id, '-', key1, '-', col1)     FROM tsubq     WHERE tsubq.key1 = 333     ORDER BY tsubq.id ASC LIMIT 1) from    t1;

+------+-------------+-------+------+---------------+------+---------+-------+------+-------------+

| id   | select_type | table | type | possible_keys | key  | key_len | ref   | rows | Extra       |

+------+-------------+-------+------+---------------+------+---------+-------+------+-------------+

|    1 | PRIMARY     | t1    | ALL  | NULL          | NULL | NULL    | NULL  | 1000 |             |

|    2 | SUBQUERY    | tsubq | ref  | key1          | key1 | 5       | const |  999 | Using where |

+------+-------------+-------+------+---------------+------+---------+-------+------+-------------+

Attachments

Issue Links

relates to

MDEV-6081 ORDER BY+ref(const): selectivity is very incorrect (MySQL Bug#14338686)

Closed

Activity

Ascending order - Click to sort in descending order

Sergei Petrunia created issue - 2014-04-07 20:21

Sergei Petrunia made changes - 2014-04-07 20:21

Field	Original Value	New Value
Assignee		Sergei Petrunia [ psergey ]

Sergei Petrunia made changes - 2014-04-08 02:35

Description

{noformat}
create table t1 (a int);
insert into t1 select * from test.one_k;

create table tsubq(
  id int primary key,
  key1 int,
  col1 int,
  key(key1)
);

insert into tsubq
  select A.a + B.a*1000, A.a, 123456 from test.one_k A, test.one_k B;
{noformat}

{noformat}
explain select
   (SELECT
      concat(id, '-', key1, '-', col1)
    FROM tsubq
    WHERE tsubq.key1 = t1.a
    ORDER BY tsubq.id ASC LIMIT 1)
from
  t1;
{noformat}

{noformat}
+------+--------------------+-------+-------+---------------+---------+---------+------+------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+------+--------------------+-------+-------+---------------+---------+---------+------+------+-------------+
| 1 | PRIMARY | t1 | ALL | NULL | NULL | NULL | NULL | 1000 | |
| 2 | DEPENDENT SUBQUERY | tsubq | index | key1 | PRIMARY | 4 | NULL | 1 | Using where |
+------+--------------------+-------+-------+---------------+---------+---------+------+------+-------------+
{noformat}

The above is wrong. Note also a very wrong estimate for #rows.

If I change outer reference to constant, the EXPLAIN is more reasonable:

{noformat}
explain select (SELECT concat(id, '-', key1, '-', col1) FROM tsubq WHERE tsubq.key1 = 333 ORDER BY tsubq.id ASC LIMIT 1) from t1;
+------+-------------+-------+------+---------------+------+---------+-------+------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+------+-------------+-------+------+---------------+------+---------+-------+------+-------------+
| 1 | PRIMARY | t1 | ALL | NULL | NULL | NULL | NULL | 1000 | |
| 2 | SUBQUERY | tsubq | ref | key1 | key1 | 5 | const | 999 | Using where |
+------+-------------+-------+------+---------------+------+---------+-------+------+-------------+
{noformat}

Create a test dataset:
{noformat}
create table t1 (a int);
insert into t1 select * from test.one_k;

create table tsubq(
  id int primary key,
  key1 int,
  col1 int,
  key(key1)
);

insert into tsubq
  select A.a + B.a*1000, A.a, 123456 from test.one_k A, test.one_k B;
{noformat}

Then, check the plan:
{noformat}
explain select
   (SELECT
      concat(id, '-', key1, '-', col1)
    FROM tsubq
    WHERE tsubq.key1 = t1.a
    ORDER BY tsubq.id ASC LIMIT 1)
from
  t1;
{noformat}

{noformat}
+------+--------------------+-------+-------+---------------+---------+---------+------+------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+------+--------------------+-------+-------+---------------+---------+---------+------+------+-------------+
| 1 | PRIMARY | t1 | ALL | NULL | NULL | NULL | NULL | 1000 | |
| 2 | DEPENDENT SUBQUERY | tsubq | index | key1 | PRIMARY | 4 | NULL | 1 | Using where |
+------+--------------------+-------+-------+---------------+---------+---------+------+------+-------------+
{noformat}

The subquery uses "index" access, which is very inefficient. The estimate for #rows seems to come from the LIMIT clause and is very wrong in this case.

The query is not using a plan

If I change outer reference to constant, the EXPLAIN is more reasonable:

{noformat}
explain select (SELECT concat(id, '-', key1, '-', col1) FROM tsubq WHERE tsubq.key1 = 333 ORDER BY tsubq.id ASC LIMIT 1) from t1;
+------+-------------+-------+------+---------------+------+---------+-------+------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+------+-------------+-------+------+---------------+------+---------+-------+------+-------------+
| 1 | PRIMARY | t1 | ALL | NULL | NULL | NULL | NULL | 1000 | |
| 2 | SUBQUERY | tsubq | ref | key1 | key1 | 5 | const | 999 | Using where |
+------+-------------+-------+------+---------------+------+---------+-------+------+-------------+
{noformat}

Sergei Petrunia made changes - 2014-04-08 02:39

Description

Create a test dataset:
{noformat}
create table t1 (a int);
insert into t1 select * from test.one_k;

create table tsubq(
  id int primary key,
  key1 int,
  col1 int,
  key(key1)
) engine=innodb;

insert into tsubq
  select A.a + B.a*1000, A.a, 123456 from test.one_k A, test.one_k B;
{noformat}

Then, check the plan:
{noformat}
explain select
   (SELECT
      concat(id, '-', key1, '-', col1)
    FROM tsubq
    WHERE tsubq.key1 = t1.a
    ORDER BY tsubq.id ASC LIMIT 1)
from
  t1;
{noformat}

{noformat}
+------+--------------------+-------+-------+---------------+---------+---------+------+------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+------+--------------------+-------+-------+---------------+---------+---------+------+------+-------------+
| 1 | PRIMARY | t1 | ALL | NULL | NULL | NULL | NULL | 1000 | |
| 2 | DEPENDENT SUBQUERY | tsubq | index | key1 | PRIMARY | 4 | NULL | 1 | Using where |
+------+--------------------+-------+-------+---------------+---------+---------+------+------+-------------+
{noformat}

The subquery uses "index" access, which is very inefficient. The estimate for #rows seems to come from the LIMIT clause and is very wrong in this case.

The table is InnoDB (with extended keys). The index KEY(key1) is actually KEY(key1, id). The query has a restriction on key1 which makes it constant (tsubq.key1 = t1.a). After that, ORDER BY tsubq.id is achieved automatically.

The problem seems to be specifically with references to outside of subquery. If I use a constant instead, the query plan is able to use key1:

{noformat}
explain select (SELECT concat(id, '-', key1, '-', col1) FROM tsubq WHERE tsubq.key1 = 333 ORDER BY tsubq.id ASC LIMIT 1) from t1;
+------+-------------+-------+------+---------------+------+---------+-------+------+-------------+
| id | select_type | table | type | possible_keys | key | key_len | ref | rows | Extra |
+------+-------------+-------+------+---------------+------+---------+-------+------+-------------+
| 1 | PRIMARY | t1 | ALL | NULL | NULL | NULL | NULL | 1000 | |
| 2 | SUBQUERY | tsubq | ref | key1 | key1 | 5 | const | 999 | Using where |
+------+-------------+-------+------+---------------+------+---------+-------+------+-------------+
{noformat}

Sergei Petrunia made changes - 2014-04-09 03:19

Affects Version/s		5.3.12 [ 12000 ]
Affects Version/s		5.5.36 [ 14600 ]

Sergei Petrunia made changes - 2014-04-09 03:19

Fix Version/s

10.0.11 [ 15200 ]

Sergei Petrunia made changes - 2014-04-09 03:19

Labels

upstream, upstream-5.5

Sergei Petrunia made changes - 2014-04-09 03:20

Labels

upstream, upstream-5.5

upstream upstream-5.5

Sergei Petrunia made changes - 2014-04-09 03:20

Status

Open [ 1 ]

In Progress [ 3 ]

Sergei Petrunia made changes - 2014-04-11 23:08

Link

This issue relates to ~~MDEV-6081~~ [ ~~MDEV-6081~~ ]

Sergei Petrunia made changes - 2014-04-12 00:10

Resolution		Fixed [ 1 ]
Status	In Progress [ 3 ]	Closed [ 6 ]

Sergei Golubchik made changes - 2014-06-13 15:06

Workflow

defaullt [ 38410 ]

MariaDB v2 [ 43782 ]

Rasmus Johansson (Inactive) made changes - 2015-05-18 17:51

Workflow

MariaDB v2 [ 43782 ]

MariaDB v3 [ 62972 ]

Sergei Golubchik made changes - 2021-12-06 21:39

Workflow

MariaDB v3 [ 62972 ]

MariaDB v4 [ 147757 ]

People

Assignee:: Sergei Petrunia

Reporter:: Sergei Petrunia

Votes:: 0 Vote for this issue

Watchers:: 2 Start watching this issue

Dates

Created:: 2014-04-07 20:21

Updated:: 2014-04-12 00:10

Resolved:: 2014-04-12 00:10

Git Integration

Error rendering 'com.xiplink.jira.git.jira_git_plugin:git-issue-webpanel'. Please contact your Jira administrators.

MariaDB Server

Details

Description

Attachments

Issue Links

Activity

People

Dates

Git Integration