案例：改写SQL消除子查询（案例1）

现象描述

1. select
2. 1,
3. (select count(*) from normal_date n where n.id = a.id) as GZCS
4. from normal_date a;

此SQL性能较差，查看发现执行计划中存在SubPlan，具体如下：

1. QUERY PLAN
2. ---------------------------------------------------------------------------------------------------------------------------------------
3. Seq Scan on normal_date a  (cost=0.00..888118.42 rows=5129 width=4) (actual time=2.394..22194.907 rows=10000 loops=1)
4. SubPlan 1
5. ->  Aggregate  (cost=173.12..173.12 rows=1 width=8) (actual time=22179.496..22179.942 rows=10000 loops=10000)
6. ->  Seq Scan on normal_date n  (cost=0.00..173.11 rows=1 width=0) (actual time=11279.349..22159.608 rows=10000 loops=10000)
7. Filter: (id = a.id)
8. Rows Removed by Filter: 99990000
9. Total runtime: 22196.415 ms
10. (7 rows)

优化说明

此优化的核心就是消除子查询。分析业务场景发现_a**.**id_不为null，那么从SQL语义出发，可以等价改写SQL为：

1. select
2. count(*)
3. from normal_date n, normal_date a
4. where n.id = a.id
5. group by  a.id;
6. 计划如下：
7. QUERY PLAN
8. ----------------------------------------------------------------------------------------------------------------------------------
9. HashAggregate  (cost=480.86..532.15 rows=5129 width=12) (actual time=21.539..24.356 rows=10000 loops=1)
10. Group By Key: a.id
11. ->  Hash Join  (cost=224.40..455.22 rows=5129 width=4) (actual time=6.402..13.484 rows=10000 loops=1)
12. Hash Cond: (n.id = a.id)
13. ->  Seq Scan on normal_date n  (cost=0.00..160.29 rows=5129 width=4) (actual time=0.087..1.459 rows=10000 loops=1)
14. ->  Hash  (cost=160.29..160.29 rows=5129 width=4) (actual time=6.065..6.065 rows=10000 loops=1)
15. Buckets: 32768  Batches: 1  Memory Usage: 352kB
16. ->  Seq Scan on normal_date a  (cost=0.00..160.29 rows=5129 width=4) (actual time=0.046..2.738 rows=10000 loops=1)
17. Total runtime: 26.844 ms
18. (9 rows)

说明： 为了保证改写的等效性，在_normal_date.id_加了_not null_约束。