SQL查询和优化（十二）

下等连接的标量子查询的改写
上一篇的例子中，主表中没有过滤条件。当主表中有过滤条件时，按上一篇的方法就会产生不必要的访问量

SELECT s.stkcode,
        t.mktcode,
        t.stype,
        t.sname,
        (SELECT SUM(c.hsl)
        FROM c 
        WHERE c.stkcode = s.stkcode
        AND c.mktcode = t.mktcode
        AND c.calcdate BETWEEN to_char(to_date(s.tdate,'yyyymmdd')-365,'yyyymmdd')AND s.tdate)AS f1,
        (SELECT decode(COUNT(C.CALCDATE),0,null,SUM(c.hsl)/COUNT(c.calcdate))
        FROM c
        WHERE c.stkcode = s.stkcode
        AND c.mktcode = t.mktcode
        AND c.calcdate BETWEEN to_char(to_date(s.tdate,'yyyymmdd')-365,'yyyymmdd')AND s.tdate)AS f2,
        s.tdate
    FROM s,t
    WHERE s.stkcode = t.scode
      AND t.status = 1
      AND t.stype = 2
      AND s.tdate >= to_number(to_char(SYSDATE -3,'YYYYMMDD'));

这各标量有以下两种必定方法：
1、把标题数据单独提出来与主表进行JOIN操作，因有区间条件，这时，主表做再次JOIN
2、主表直接与标量进行JOIN操作，然后汇总数据
两种改写方法都需要把主查询放在WITH语句中
WITH t0 AS
 (SELECT rownum AS sn,s.stkcode, t.mktcode,t.stype,t.sname,s.tdate
 FROM s,t
 WHERE s.stkcode = t.scode
      AND t.status = 1
      AND t.stype = 2
      AND s.tdate >= to_number(to_char(SYSDATE -3,'YYYYMMDD')))
这个语句中有两个表，不能返回rowid，所以使用rownum as sn 当作唯一标识
改写一：先分组汇总，再关联
WITH t0 AS
 (SELECT rownum AS sn,s.stkcode, t.mktcode,t.stype,t.sname,s.tdate
 FROM s,t
 WHERE s.stkcode = t.scode
      AND t.status = 1
      AND t.stype = 2
      AND s.tdate >= to_number(to_char(SYSDATE -3,'YYYYMMDD')))
SELECT t.stkcode, t.mktcode, t.stype,t.sname,t.tdate,c.f2,c.f2
  FROM t0 t 
  LEFT JOIN (SELECT t.sn,
                  SUM(c.hsl) AS f1,
                    decode(COUNT(C.CALCDATE),
                            null,
                            SUM(c.hsl)/COUNT(c.calcdate)) AS f2
            FROM c
            INNER JOIN t0 t ON (c.stkcode = t.stkcode
                                AND c.mktcode = t.mktcode)
            WHERE c.calcdate BETWEEN to_char(to_date(t.tdate,'yyyymmdd')-365,'yyyymmdd')AND s.tdate)
            GROUP BY t.sn) c
            ON (c.sn = t.sn);

先关联，再汇总
WITH t0 AS
 (SELECT rownum AS sn,s.stkcode, t.mktcode,t.stype,t.sname,s.tdate
 FROM s,t
 WHERE s.stkcode = t.scode
      AND t.status = 1
      AND t.stype = 2
      AND s.tdate >= to_number(to_char(SYSDATE -3,'YYYYMMDD')))
SELECT t.stkcode, 
        t.mktcode, 
        t.stype,
        t.sname,
        t.tdate,
        SUM(c.hsl) AS f1,
        decode(COUNT(C.CALCDATE),0,null,SUM(c.hsl)/COUNT(c.calcdate))AS f2
FROM TO t
LEFT JOIN c ON (c.stkcode = t.stkcode AND c.mktcode = t.mktcode)
WHERE c.calcdate BETWEE to_char(to_date(t.tdate,'yyyymmdd')-365,'yyyymmdd')
                AND s.tdate)
GROUP BY t.sn,t.skcode,t.mktcode,t.stype,t.sname,t,tdate;