• Stable 函数不能修改数据库，单个Query中所有行给定同样的参数确保返回相同的结果。这种稳定级别允许优化器将多次函数调用转换为一次。在索引扫描的条件中使用这种函数是可行的，因为索引扫描只计算一次比较值(comparison value),而不是每行都计算一次。
• Immutable 函数不能修改数据库，在任何情况下，只要输入参数相同，返回结果就相同。这种级别的函数，优化器可以提前进行计算，在查询过程中作为常量参数。比如：SELECT…WHERE x=2+2 可以简化为SELECT…WHERE x=4。

以下以例子说明二者的差异。

1、准备数据

create table t1(id1 integer,id2 integer);
insert into t1 select generate_series(1,10000000),generate_series(1,10000000);

test=# \timing on
Timing is on.
test=# select count(*) from t1;

count

10000000
(1 row)

Time: 681.445 ms

2、创建immutable 和 stable 函数

create or replace function f001()
returns bigint
immutable
language sql
as
$$ select count(*) from t1 $$ ;

create or replace function f002()
returns bigint
stable
language sql
as
$$ select count(*) from t1 $$ ;

3、单独explain 函数

可以看到对于 immutable 函数，在 explain 时，实际会去执行的；而stable 函数，explain 时则不会实际执行。

test=# explain select f001();

QUERY PLAN

Result (cost=0.00..0.01 rows=1 width=8)
(1 row)

Time: 450.572 ms
test=# explain select f002();

QUERY PLAN

Result (cost=0.00..0.26 rows=1 width=8)
(1 row)

Time: 0.641 ms

test=# select f001();

f001

10000000
(1 row)

Time: 448.720 ms
test=# select f002();

f002

10000000
(1 row)

Time: 426.745 ms

4、例子一

可以看到 immutable 函数执行时间主要花在planning上，也就是在制定执行计划前，就已经取得函数的值；而 stable 函数，则在语句解析和执行时，都要执行函数，而且，针对语句的访问的每个tuple，都要执行一次函数调用。

test=# explain analyze select * from (select * from t1 limit 10) a where a.id1=f001();

QUERY PLAN

Subquery Scan on a (cost=0.00..0.27 rows=1 width=8) (actual time=0.012..0.013 rows=0 loops=1)
Filter: (a.id1 = '10000000'::bigint)
Rows Removed by Filter: 10
-> Limit (cost=0.00..0.15 rows=10 width=8) (actual time=0.009..0.010 rows=10 loops=1)
-> Seq Scan on t1 (cost=0.00..148609.21 rows=10007621 width=8) (actual time=0.008..0.009 rows=10 loops=1)
Planning Time: 413.963 ms
Execution Time: 0.026 ms
(7 rows)

test=# explain analyze select * from (select * from t1 limit 10) a where a.id1=f002();

QUERY PLAN

Subquery Scan on a (cost=0.00..2.77 rows=1 width=8) (actual time=3691.788..3691.788 rows=0 loops=1)
Filter: (a.id1 = f002())
Rows Removed by Filter: 10
-> Limit (cost=0.00..0.15 rows=10 width=8) (actual time=0.012..0.028 rows=10 loops=1)
-> Seq Scan on t1 (cost=0.00..148609.21 rows=10007621 width=8) (actual time=0.011..0.021 rows=10 loops=1)
Planning Time: 364.233 ms
Execution Time: 3691.807 ms
(7 rows)

Time: 4056.907 ms (00:04.057)

test=# explain analyze select * from (select * from t1 where 1=2) a where a.id1=f002();

QUERY PLAN

Result (cost=0.00..2675533.51 rows=1 width=8) (actual time=0.001..0.001 rows=0 loops=1)
One-Time Filter: false
-> Seq Scan on t1 (cost=0.00..2675533.51 rows=1 width=8) (never executed)
Filter: (id1 = f002())
Planning Time: 490.720 ms
Execution Time: 0.017 ms
(6 rows)

test=# explain analyze select * from (select * from t1 limit 1) a where a.id1=f002();

QUERY PLAN

Subquery Scan on a (cost=0.00..0.28 rows=1 width=8) (actual time=390.833..390.834 rows=0 loops=1)
Filter: (a.id1 = f002())
Rows Removed by Filter: 1
-> Limit (cost=0.00..0.01 rows=1 width=8) (actual time=0.024..0.053 rows=1 loops=1)
-> Seq Scan on t1 (cost=0.00..148609.21 rows=10007621 width=8) (actual time=0.023..0.023 rows=1 loops=1)
Planning Time: 426.693 ms
Execution Time: 390.852 ms
(7 rows)

5、例子二

sysdate 函数 为 stable 时：sysdate 函数在同一事务内多次执行返回的结果都相同，跨事务则每次结果不同。

test=# begin
test-# for i in 1..5 loop
test-# raise notice '%', sysdate();
test-# perform sys_sleep(1);
test-# commit;
test-# end loop;
test-# end;
test-# /
NOTICE: 2022-06-29 19:53:18
NOTICE: 2022-06-29 19:53:19
NOTICE: 2022-06-29 19:53:20
NOTICE: 2022-06-29 19:53:21
NOTICE: 2022-06-29 19:53:22
ANONYMOUS BLOCK
Time: 5011.031 ms (00:05.011)

test=# begin
test-# for i in 1..5 loop
test-# raise notice '%', sysdate();
test-# perform sys_sleep(1);
test-# end loop;
test-# end;
test-# /
NOTICE: 2022-06-29 19:54:14
NOTICE: 2022-06-29 19:54:14
NOTICE: 2022-06-29 19:54:14
NOTICE: 2022-06-29 19:54:14
NOTICE: 2022-06-29 19:54:14
ANONYMOUS BLOCK
Time: 5005.724 ms (00:05.006)

sysdate 函数为 immutable 时：不管是否跨事务，sysdate 函数结果都相同。

test=# alter function sysdate immutable;
ALTER FUNCTION
Time: 6.276 ms

test=# begin
test-# for i in 1..5 loop
test-# raise notice '%', sysdate();
test-# perform sys_sleep(1);
test-# commit;
test-# end loop;
test-# end;
test-# /
NOTICE: 2022-06-29 19:54:58
NOTICE: 2022-06-29 19:54:58
NOTICE: 2022-06-29 19:54:58
NOTICE: 2022-06-29 19:54:58
NOTICE: 2022-06-29 19:54:58
ANONYMOUS BLOCK
Time: 5007.899 ms (00:05.008)

test=# begin
test-# for i in 1..5 loop
test-# raise notice '%', sysdate();
test-# perform sys_sleep(1);
test-# end loop;
test-# end;
test-# /
NOTICE: 2022-06-29 19:55:11
NOTICE: 2022-06-29 19:55:11
NOTICE: 2022-06-29 19:55:11
NOTICE: 2022-06-29 19:55:11
NOTICE: 2022-06-29 19:55:11
ANONYMOUS BLOCK
Time: 5007.694 ms (00:05.008)

1、创建函数

create or replace function f001
return integer
deterministic
as
cnt integer;
begin
for i in 1..10 loop
select count(*) into cnt from t1 ;
end loop;
return cnt;
end ;

create or replace function f002
return integer
as
cnt integer;
begin
for i in 1..10 loop
select count(*) into cnt from t1 ;
end loop;
return cnt;
end ;

2、单独explain 函数

可以看到两个函数都不会执行，不管是 deterministic，还是 volatile

SQL> explain plan for select f001() from dual;

Explained.

Elapsed: 00:00:00.00
SQL> explain plan for select f002() from dual;

Explained.

Elapsed: 00:00:00.00

3、实际执行

deterministic 只需执行一次，但是在SQL执行时才执行函数调用，而非explain时；volatile 解析时，不需要调用函数，而针对每个tuple 都必须要调用一次，如果没有记录，则无需调用。

SQL> select * from (select * from t1 where rownum<11) where id1=f001();

no rows selected

Elapsed: 00:00:00.48

SQL> select * from (select * from t1 where rownum<11) where id1=f002();

no rows selected

Elapsed: 00:00:05.01

SQL> select * from (select * from t1 where 1=2) where id1=f002();

no rows selected

Elapsed: 00:00:00.00