Help Center/ GaussDB/ Developer Guide(Centralized_8.x)/ SQL Optimization/ Typical SQL Optimization Methods/ Optimizing Subqueries
Updated on 2024-08-20 GMT+08:00
View PDF
Share

Optimizing Subqueries

Context

When an application runs an SQL statement to operate the database, a large number of subqueries are used because they are more clear than table join. Especially in complicated query statements, subqueries have more complete and independent semantics, which makes SQL statements clearer and easier to understand. Therefore, subqueries are widely used.

In GaussDB, subqueries can also be called sublinks based on the location of subqueries in SQL statements.

  • Subquery: corresponds to a range table (RangeTblEntry) in the query parse tree. That is, a subquery is a SELECT statement following immediately after the FROM keyword.
  • Sublink: corresponds to an expression in the query parsing tree. That is, a sublink is a statement in the WHERE or ON clause or in the target list.
    In conclusion, a subquery is a range table and a sublink is an expression in the query parsing tree. A sublink can be found in constraint conditions and expressions. In GaussDB, sublinks can be classified into the following types:
    • exist_sublink: corresponds to the EXIST and NOT EXIST statements.
    • any_sublink: corresponds to the op ANY(SELECT...) statement. op can be the <, >, or = operator. IN/NOT IN (SELECT...) also belongs to this type.
    • all_sublink: corresponds to the op ALL(SELECT...) statement. op can be the <, >, or = operator.
    • rowcompare_sublink: corresponds to the RECORD op (SELECT...) statement.
    • expr_sublink: corresponds to the (SELECT with a single target list item...) statement.
    • array_sublink: corresponds to the ARRAY(SELECT...) statement.
    • cte_sublink: corresponds to the WITH(...) query statement.

    The exist_sublink and any_sublink are pulled up by the optimization engine of GaussDB. In addition, expr_sublink can also be pulled up. However, because of the flexible use of subqueries in SQL statements, complex subqueries may affect query performance. If you do not want to pull up expr_sublink, set the GUC parameter rewrite_rule. Subqueries are classified into non-correlated subqueries and correlated subqueries.

    • Non-correlated subqueries

      The execution of a subquery is independent from attributes of the outer query. In this way, a subquery can be executed before outer queries.

      Example:

       1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
      		 
      gaussdb=# EXPLAIN SELECT t1.c1,t1.c2
FROM t1
WHERE t1.c1 IN (
    SELECT c2
    FROM t2
    WHERE t2.c2 IN (2,3,4)
);
                              QUERY PLAN
----------------------------------------------------------------
 Hash Join 
   Hash Cond: (t1.c1 = t2.c2)
   ->  Seq Scan on t1 
         Filter: (c1 = ANY ('{2,3,4}'::integer[]))
   ->  Hash 
         ->  HashAggregate 
               Group By Key: t2.c2
               ->  Seq Scan on t2  
                     Filter: (c2 = ANY ('{2,3,4}'::integer[]))
(9 rows)

    • Correlated subqueries

      The execution of a subquery depends on some attributes (used as AND conditions of the subquery) of outer queries. In the following example, t1.c1 in the t2.c1 = t1.c1 condition is a correlated attribute. Such a subquery depends on outer queries and needs to be executed once for each outer query.

      Example:

       1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
      		 
      gaussdb=# EXPLAIN SELECT t1.c1,t1.c2
FROM t1
WHERE t1.c1 IN (
    SELECT c2
    FROM t2
    WHERE t2.c1 = t1.c1 AND t2.c2 IN (2,3,4)
);
                               QUERY PLAN
------------------------------------------------------------------------
 Seq Scan on t1
   Filter: (SubPlan 1)
   SubPlan 1
     ->  Seq Scan on t2
           Filter: ((c1 = t1.c1) AND (c2 = ANY ('{2,3,4}'::integer[])))
(5 rows)

Sublink Optimization on GaussDB

To optimize a sublink, a subquery is pulled up to join with tables in outer queries. You can run the EXPLAIN statement to check whether a sublink is converted into such a plan.

Example:

gaussdb=# EXPLAIN SELECT t1.c1, t1.c2 FROM t1 WHERE t1.c1 IN(SELECT c2 FROM t2 WHERE t2.c1 = t1.c1);
           QUERY PLAN           
--------------------------------
 Seq Scan on t1
   Filter: (SubPlan 1)
   SubPlan 1
     ->  Seq Scan on t2
           Filter: (c1 = t1.c1)
(5 rows)
  • Sublink-release scenarios supported by GaussDB
    • Pulling up the IN sublink
      • The subquery cannot contain columns in the outer query (columns in more outer queries are allowed).
      • The subquery cannot contain volatile functions.

        Example:

      gaussdb=# EXPLAIN SELECT t1.c1, t1.c2 FROM t1 WHERE t1.c1 IN(SELECT c2 FROM t2 WHERE t2.c1 = 1);
              QUERY PLAN              
--------------------------------------
 Hash Join
   Hash Cond: (t1.c1 = t2.c2)
   ->  Seq Scan on t1
   ->  Hash
         ->  HashAggregate
               Group By Key: t2.c2
               ->  Seq Scan on t2
                     Filter: (c1 = 1)
(8 rows)
    • Pulling up the EXISTS sublink

      The WHERE clause must contain a column in the outer query. Other parts of the subquery cannot contain the column. Other restrictions are as follows:

      • The subquery must contain the FROM clause.
      • The subquery cannot contain the WITH clause.
      • The subquery cannot contain aggregate functions.
      • The subquery cannot contain a SET, SORT, LIMIT, WindowAgg, or HAVING operation.
      • The subquery cannot contain volatile functions.

        Example:

      Replace the execution plan on the right of the arrow with the following execution plan:

                  QUERY PLAN             
-----------------------------------
 Hash Join
   Hash Cond: (t1.c1 = t2.c1)
   ->  Seq Scan on t1
   ->  Hash
         ->  HashAggregate
               Group By Key: t2.c1
               ->  Seq Scan on t2
(7 rows)
    • Pulling up an equivalent correlated query containing aggregate functions

      The WHERE condition of the subquery must contain a column from the outer query. Equivalence comparison must be performed between this column and related columns in tables of the subquery. These conditions must be connected using AND. Other parts of the subquery cannot contain the column. Other restrictions are as follows:

      • The columns in the expression in the WHERE condition of the subquery must exist in tables.
      • After the SELECT keyword of the subquery, there must be only one output column. The output column must be an aggregate function (for example, MAX), and the parameter (for example, t2.c2) of the aggregate function cannot be columns of a table (for example, t1) in outer queries. The aggregate function cannot be COUNT.
        For example, the following subquery can be pulled up: 
        1
2
3
        		 
        SELECT * FROM t1 WHERE c1 >(
       SELECT max(t2.c1) FROM t2 WHERE t2.c1=t1.c1
);

        The following subquery cannot be pulled up because the subquery has no aggregate function:

        1
2
3
        		 
        SELECT * FROM t1 WHERE c1 >(
       SELECT  t2.c1 FROM t2 WHERE t2.c1=t1.c1
);

        The following subquery cannot be pulled up because the subquery has two output columns:

        1
2
3
        		 
        SELECT * FROM t1 WHERE (c1,c2) >(
       SELECT max(t2.c1),min(t2.c2) FROM t2 WHERE t2.c1=t1.c1
);
      • The subquery must be a FROM clause.
      • The subquery cannot contain a GROUP BY, HAVING, or SET operation.
      • The subquery can only be an INNER JOIN.
        For example, the following subquery cannot be pulled up: 
        1
2
3
        		 
        SELECT * FROM t1 WHERE c1 >(
       SELECT max(t2.c1) FROM t2 FULL JOIN t3 ON (t2.c2=t3.c2) WHERE t2.c1=t1.c1
);
      • The target list of the subquery cannot contain the function that returns a set.
      • The WHERE condition of the subquery must contain a column from the outer query. Equivalence comparison must be performed between this column and related columns in tables of the subquery. These conditions must be connected using AND. Other parts of the subquery cannot contain the column. For example, the following subquery can be pulled up: 
        1
2
3
4
5
        		 
        SELECT * FROM t3 WHERE t3.c1=(
        SELECT t1.c1
        FROM t1 WHERE c1 >(
                SELECT max(t2.c1) FROM t2 WHERE t2.c1=t1.c1 
));

        If another condition is added to the subquery in the previous example, the subquery cannot be pulled up because the subquery references to the column in the outer query. Example:

        1
2
3
4
5
6
        		 
        SELECT * FROM t3 WHERE t3.c1=(
        SELECT t1.c1
        FROM t1 WHERE c1 >(
               SELECT max(t2.c1) FROM t2 WHERE t2.c1=t1.c1 AND t3.c1>t2.c2

));
    • Pulling up a sublink in the OR clause

      If the WHERE condition contains an EXIST correlated sublink connected by OR, the following is an example:

      1
2
3
      		 
      SELECT a, c FROM t1
WHERE t1.a = (SELECT avg(a) FROM t3 WHERE t1.b = t3.b) OR
EXISTS (SELECT * FROM t4 WHERE t1.c = t4.c);

      The process of pulling up such a sublink is as follows:

      1. Extract opExpr from the OR clause in the WHERE condition. The value is t1.a = (select avg(a) from t3 where t1.b = t3.b).
      2. The opExpr contains a subquery. If the subquery can be pulled up, the subquery is rewritten as SELECT avg(a), t3.b FROM t3 GROUP BY t3.b, generating the NOT NULL condition t3.b IS NOT NULL. The opExpr is replaced by this NOT NULL condition. In this case, the SQL statement changes to: 
        1
2
3
        		 
        SELECT a, c
FROM t1 LEFT JOIN (SELECT avg(a) avg, t3.b FROM t3 GROUP BY t3.b) AS t3 ON (t1.a = avg and t1.b = t3.b)
WHERE t3.b IS NOT NULL OR exists (SELECT * FROM t4 WHERE t1.c = t4.c);
      3. Extract the EXISTS sublink exists (select * from t4 where t1.c = t4.c) from the OR clause to check whether the sublink can be pulled up. If it can be pulled up, the subquery is converted into select t4.c from t4 group by t4.c, generating the NOT NULL condition t4.c is not null. In this case, the SQL statement changes to: 
        1
        		 
        SELECT t1.a, t1.c FROM t1 LEFT JOIN (SELECT avg(a) avg, t3.b FROM t3 GROUP BY t3.b) AS t3 ON (t1.a = avg AND t1.b = t3.b) LEFT JOIN (SELECT t5.c FROM t5 GROUP BY t5.c) AS t5 ON (t1.c = t5.c) WHERE t3.b IS NOT NULL OR t5.c IS NOT NULL;
  • Sublink-release scenarios not supported by GaussDB

    Except the sublinks described above, all the other sublinks cannot be pulled up. In this case, a join subquery is planned as the combination of subplans and broadcast. As a result, if inner tables have a large amount of data, query performance may be poor.

    If a correlated subquery joins with two tables in outer queries, the subquery cannot be pulled up. You need to change the parent query into a WITH clause and then perform the join.

    Example:

    1
2
    		 
    SELECT distinct t1.a, t2.a
FROM t1 LEFT JOIN t2 ON t1.a=t2.a AND NOT EXISTS (SELECT a,b FROM test1 WHERE test1.a=t1.a AND test1.b=t2.a);

    The rewriting is as follows:

    1
2
3
4
5
6
7
8
    		 
    WITH temp AS
(
        SELECT * FROM (SELECT t1.a AS a, t2.a AS b FROM t1 LEFT JOIN t2 ON t1.a=t2.a)

)
SELECT distinct a,b
FROM temp
WHERE NOT EXISTS (SELECT a,b FROM test1 WHERE temp.a=test1.a AND temp.b=test1.b);
    • The subquery (without COUNT) in the target list cannot be pulled up.

      Example:

      1
2
3
4
      		 
      gaussdb=# EXPLAIN (costs off)
SELECT (SELECT c2 FROM t2 WHERE t1.c1 = t2.c1) ssq, t1.c2
FROM t1
WHERE t1.c2 > 10;

      The execution plan is as follows:

       1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
      		 
      gaussdb=# EXPLAIN (costs off)
SELECT (SELECT c2 FROM t2 WHERE t1.c1 = t2.c1) ssq, t1.c2
FROM t1
WHERE t1.c2 > 10;
           QUERY PLAN
--------------------------------
 Seq Scan on t1
   Filter: (c2 > 10)
   SubPlan 1
     ->  Seq Scan on t2
           Filter: (t1.c1 = c1)
(5 rows)

      The correlated subquery is displayed in the target list (query return list). Values need to be returned even if the condition t1.c1=t2.c1 is not met. Therefore, use a right outer join to join t2 and t1 so that the SSQ can return padding values when the condition t1.c1=t2.c1 is not met.

      ScalarSubQuery (SSQ) and Correlated-ScalarSubQuery (CSSQ) are described as follows:

      • SSQ: a sublink that returns a scalar value of a single row with a single column
      • CSSQ: an SSQ containing correlation conditions

      The preceding SQL statement can be changed into:

      1
2
3
4
5
      		 
      WITH ssq AS
(
    SELECT * FROM t1 WHERE t1.c2 >10
)
SELECT t2.c2,ssq.c2 FROM t2 RIGHT JOIN ssq ON ssq.c1 = t2.c1;

      The execution plan after the change is as follows:

      1
2
3
4
5
6
7
8
9
      		 
                 QUERY PLAN            
---------------------------------
 Hash Right Join
   Hash Cond: (t2.c1 = t1.c1)
   ->  Seq Scan on t2
   ->  Hash
         ->  Seq Scan on t1
               Filter: (c2 > 10)
(6 rows)

      In the preceding example, the SSQ in the target list is pulled up to right join, preventing poor performance caused by the plan involving subplans when the table (t2) in the subquery is too large.

    • The subquery (with COUNT) in the target list cannot be pulled up.

      Example:

      1
2
3
      		 
      SELECT (SELECT count(*) FROM t2 WHERE t2.c1=t1.c1) cnt, t1.c1, t3.c1
FROM t1,t3
WHERE t1.c1=t3.c1 ORDER BY cnt, t1.c1;

      The execution plan is as follows:

       1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
      		 
                       QUERY PLAN
--------------------------------------------
 Sort
   Sort Key: ((SubPlan 1)), t1.c1
   ->  Hash Join
         Hash Cond: (t1.c1 = t3.c1)
         ->  Seq Scan on t1
         ->  Hash
               ->  Seq Scan on t3
         SubPlan 1
           ->  Aggregate
                 ->  Seq Scan on t2
                       Filter: (c1 = t1.c1)
(11 rows)

      The correlated subquery is displayed in the target list (query return list). Values need to be returned even if the condition t1.c1=t2.c1 is not met. Therefore, use left outer join to join T1 and T2 so that SSQ can return padding values when the condition t1.c1=t2.c1 is not met. However, COUNT is used, which requires that 0 is returned when the condition is not met. Therefore, case-when NULL then 0 else count(*) can be used.

      The preceding SQL statement can be changed into:

       1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
      		 
      WITH ssq AS
(
    SELECT count(*) cnt, c1 FROM t2 GROUP BY c1
)
SELECT case WHEN
            ssq.cnt IS NULL THEN 0
            ELSE ssq.cnt
       END cnt, t1.c1, t3.c1
FROM t1 LEFT JOIN ssq ON ssq.c1 = t1.c1,t3
WHERE t1.c1 = t3.c1
ORDER BY ssq.cnt, t1.c1;

      The execution plan after the change is as follows:

       1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
      		 
                      QUERY PLAN
-------------------------------------------
 Sort
   Sort Key: ssq.cnt, t1.c1
   CTE ssq
     ->  HashAggregate
           Group By Key: t2.c1
           ->  Seq Scan on t2
   ->  Hash Join
         Hash Cond: (t1.c1 = t3.c1)
         ->  Hash Left Join
               Hash Cond: (t1.c1 = ssq.c1)
               ->  Seq Scan on t1
               ->  Hash
                     ->  CTE Scan on ssq
         ->  Hash
               ->  Seq Scan on t3
(15 rows)
    • Non-equivalent correlated subqueries cannot be pulled up.

      Example:

      1
2
3
      		 
      SELECT t1.c1, t1.c2
FROM t1
WHERE t1.c1 = (SELECT agg() FROM t2.c2 > t1.c2);

      Non-equivalent correlated subqueries cannot be pulled up. You can perform join twice (one CorrelationKey and one rownum self-join) to rewrite the statement.

      You can rewrite the statement in either of the following ways:

      • Subquery rewriting: 
        1
2
3
4
5
6
7
        		 
        SELECT t1.c1, t1.c2
FROM t1, (
    SELECT t1.rowid, agg() aggref
    FROM t1,t2
    WHERE t1.c2 > t2.c2 GROUP BY t1.rowid
) dt /* derived table */
WHERE t1.rowid = dt.rowid AND t1.c1 = dt.aggref;
      • CTE rewriting: 
         1
 2
 3
 4
 5
 6
 7
 8
 9
10
        		 
        WITH dt as
(
    SELECT t1.rowid, agg() aggref
    FROM t1,t2
    WHERE t1.c2 > t2.c2 GROUP BY t1.rowid
)
SELECT t1.c1, t1.c2
FROM t1, dt
WHERE t1.rowid = dt.rowid AND
t1.c1 = dt.aggref;
    • If the AGG type is COUNT(*), 0 is used for data padding when CASE-WHEN is not matched. If the type is not COUNT(*), NULL is used.
    • CTE rewriting works better by using sharescan.

More Optimization Examples

Example: Modify the SELECT statement by modifying the subquery to a join relationship between the main table and the parent query or modifying the subquery to improve the query performance. Ensure that the subquery to be used is semantically correct.

1
2
3
4
5
6
7
8
9
 
gaussdb=# explain (costs off) select * from t1 where t1.c1 in (select t2.c1 from t2 where t1.c1 = t2.c2);
           QUERY PLAN
--------------------------------
 Seq Scan on t1
   Filter: (SubPlan 1)
   SubPlan 1
     ->  Seq Scan on t2
           Filter: (t1.c1 = c2)
(5 rows)

In the preceding example, a subplan is used. To remove the subplan, you can modify the statement as follows:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
 
gaussdb=# explain (costs off) select * from t1 where exists (select t2.c1 from t2 where t1.c1 = t2.c2 and t1.c1 = t2.c1);
                QUERY PLAN
------------------------------------------
 Hash Join
   Hash Cond: (t1.c1 = t2.c2)
   ->  Seq Scan on t1
   ->  Hash
         ->  HashAggregate
               Group By Key: t2.c2, t2.c1
               ->  Seq Scan on t2
                     Filter: (c2 = c1)
(8 rows)

In this way, the subplan is replaced by the hash-join between the two tables, greatly improving the execution efficiency.