教你使用常用的逻辑公式和恒等式等价改写SQL-526互联

今天同事给我一条2秒的SQL看看能不能优化。

原始SQL：

SELECT pk_dept
FROM aaaa
WHERE 1 = 1
  AND ((pk_group = '0001A110000000000JQ6' AND pk_org IN ('0001A110000000001M09')))
  AND (PK_DEPT IN (SELECT t1.ORGID
                   FROM xxxxx t1
                            INNER JOIN (SELECT (CASE WHEN ORGID3 IS NULL THEN ORGID2 ELSE ORGID3 END) orgid
                                        FROM xxxxx
                                        WHERE ORGID = '1001A110000000001U8S') t2
                                       ON t1.ORGID2 = t2.orgid OR t1.ORGID3 = t2.orgid))
  AND (enablestate IN (2))
ORDER BY code;

4 loops=1)
                                                                                                               Heap Fetches: 0
 Planning Time: 258.024 ms
 Execution Time: 2493.882 ms
(1583 rows)

xxxxx 表是一张非常复杂的视图，上面SQL执行计划大概2千行左右，所以本案例就不放整体的计划了。

如果经常做优化的同学对于简单的SQL，相信可以使用瞪眼大法基本定位到语句慢的位置?

子查询慢SQL：

  SELECT t1.ORGID
                   FROM xxxxx t1
                            INNER JOIN (SELECT (CASE WHEN ORGID3 IS NULL THEN ORGID2 ELSE ORGID3 END) orgid
                                        FROM xxxxx
                                        WHERE ORGID = '1001A110000000001U8S') t2
                                       ON t1.ORGID2 = t2.orgid OR t1.ORGID3 = t2.orgid

主要是慢在连接列的 OR 关系运算符上。

我们可以仔细看看这条SQL，xxxxx 分别作为t1、t2 表内连接关联2次，关联条件为：t1.ORGID2 = t2.orgid OR t1.ORGID3 = t2.orgid

这种关联条件的逻辑关系可以简化为： t1.ORGID2 = t1.ORGID3

SQL语句可以进行以下精简：

SELECT t1.ORGID FROM xxxxx t1 where  ORGID = '1001A110000000001U8S';

验证等价逻辑：

ncc=#  SELECT t1.ORGID FROM xxxxx t1 WHERE EXISTS (
ncc(# SELECT 1 FROM
ncc(# (SELECT (CASE WHEN ORGID3 IS NULL THEN ORGID2 ELSE ORGID3 END) orgid FROM xxxxx WHERE ORGID = '1001A110000000001U8S') t2
ncc(# WHERE t1.ORGID2 = t2.orgid  OR   t1.ORGID3 = t2.orgid
ncc(#  );
        orgid
----------------------
 1001A110000000001U8S
(1 row)

ncc=#   SELECT t1.ORGID FROM xxxxx t1 where  ORGID = '1001A110000000001U8S';
        orgid
----------------------
 1001A110000000001U8S
(1 row)

ncc=#

可以看到子查询SQL进行改写后是结果是等价的，使用逻辑公式进行化繁为简。

改写优化SQL后执行计划：

SELECT pk_dept
FROM org_dept
WHERE 1 = 1
  AND ((pk_group = '0001A110000000000JQ6' AND pk_org IN ('0001A110000000001M09')))
  AND (PK_DEPT IN (SELECT t1.ORGID
                   FROM EHR_ORG_DEPT t1
                   where ORGID = '1001A110000000001U8S')
    )
  AND (enablestate IN (2))
ORDER BY code;

                                                                                 ->  Seq Scan on org_dept d1_35  (cost=0.00..639.33 rows=7133 width=116) (actual time=0.047..1.480 rows=7133 loops=1)
                                                                                 ->  Hash  (cost=120.64..120.64 rows=3224 width=21) (actual time=1.315..1.315 rows=3224 loops=1)
                                                                                       Buckets: 4096  Batches: 1  Memory Usage: 199kB
                                                                                       ->  Index Only Scan using pk_bd_region on bd_region bg_35  (cost=0.28..120.64 rows=3224 width=21) (actual time=0.024..0.651 rows=3224 loops=1)
                                                                                             Heap Fetches: 0
 Planning Time: 86.684 ms
 Execution Time: 625.439 ms

2.4秒的SQL最后通过化繁为简最终优化到625毫秒就能跑出结果。

总结：逻辑思维课程还是挺重要的，建议大家有空的时候去看看。