原文:
通过调整表union all的顺序优化SQL
操作系统：Windows XP
数据库版本：SQL Server 2005
今天遇到一个SQL，过滤条件是自动生成的，因此，没法通过调整SQL的谓词达到优化的目的，只能去找SQL中的“大表”。有一个视图返回的结果集比较大，如果能调整的话，也只能调整该视图了。
看了一下该视图的结构，里面还套用了另一层视图，直接看最里层视图的查询SQL。
SELECT a.dfeesum_no ,
a.opr_amt - ISNULL(b.dec_pay, 0) - ISNULL(b.dec_corrpay, 0)
- ISNULL(b.dec_deduamt, 0) dec_amt ,
a.dec_camt - ISNULL(b.dec_pay, 0) - a.dec_comprate
* ISNULL(b.dec_deduamt, 0) dec_compamt ,
a.dec_ramt - ISNULL(b.dec_corrpay, 0) - ( a.dec_comprate - 1 )
* ISNULL(b.dec_deduamt, 0) dec_corramt ,
a.dec_qty - ISNULL(b.dec_qty, 0) - ISNULL(b.dec_deduqty, 0) opr_qty ,
ISNULL(b.dec_pay, 0) dec_pay ,
ISNULL(b.dec_corrpay, 0) dec_corrpay ,
ISNULL(b.dec_deduqty, 0) dec_deduqty ,
ISNULL(b.dec_deduamt, 0) dec_deduamt ,
ISNULL(b.dec_qty, 0) dec_qty
FROM ctlm8686 a
LEFT JOIN ( SELECT dfeesum_no ,
SUM(dec_ramt) dec_pay ,
SUM(dec_corramt) dec_corrpay ,
SUM(dec_qty) dec_qty ,
SUM(CASE WHEN flag_dedu = '1' THEN dec_deduamt
ELSE 0
END) dec_deduamt ,
SUM(CASE WHEN flag_dedu = '1' THEN dec_deduqty
ELSE 0
END) dec_deduqty
FROM dfeepay_03
GROUP BY dfeesum_no
) b ON a.dfeesum_no = b.dfeesum_no
UNION ALL
SELECT a.dfeesum_no ,
a.dec_amt - ISNULL(b.dec_pay, 0) - ISNULL(b.dec_corrpay, 0)
- ISNULL(b.dec_deduamt, 0) dec_amt ,
a.dec_compamt - ISNULL(b.dec_pay, 0) - a.dec_comprate
* ISNULL(b.dec_deduamt, 0) dec_compamt ,
a.dec_corramt - ISNULL(b.dec_corrpay, 0) - ( a.dec_comprate - 1 )
* ISNULL(b.dec_deduamt, 0) dec_corramt ,
a.opr_qty - ISNULL(b.dec_qty, 0) - ISNULL(b.dec_deduqty, 0) opr_qty ,
ISNULL(b.dec_pay, 0) dec_pay ,
ISNULL(b.dec_corrpay, 0) dec_corrpay ,
ISNULL(b.dec_deduqty, 0) dec_deduqty ,
ISNULL(b.dec_deduamt, 0) dec_deduamt ,
ISNULL(b.dec_qty, 0) dec_qty
FROM dfeeapp_03 a
LEFT JOIN ( SELECT dfeesum_no ,
SUM(dec_ramt) dec_pay ,
SUM(dec_corramt) dec_corrpay ,
SUM(dec_qty) dec_qty ,
SUM(CASE WHEN flag_dedu = '1' THEN dec_deduamt
ELSE 0
END) dec_deduamt ,
SUM(CASE WHEN flag_dedu = '1' THEN dec_deduqty
ELSE 0
END) dec_deduqty
FROM dfeepay_03
GROUP BY dfeesum_no
) b ON a.dfeesum_no = b.dfeesum_no
返回结果集有1433891行，其中
SELECT COUNT(*) FROM dfeepay_03 --1103914
SELECT COUNT(*) FROM ctlm8686 --1131586
SELECT COUNT(*) FROM dfeeapp_03--302305
上述SQL脚本中，子查询是相同的，即对子查询进行了两次扫描，可以考虑先让dfeeapp_03和ctlm8686union all，再left join dfeepay_03 。同时，对于子查询，先让dfeepay_03 表先查询出flag_dedu = '1'的数据，就不用再进行case when判断了。
改写后的SQL如下
SELECT a.dfeesum_no ,
a.opr_amt - ISNULL(b.dec_pay, 0) - ISNULL(b.dec_corrpay, 0)
- ISNULL(b.dec_deduamt, 0) dec_amt ,
a.dec_camt - ISNULL(b.dec_pay, 0) - a.dec_comprate
* ISNULL(b.dec_deduamt, 0) dec_compamt ,
a.dec_ramt - ISNULL(b.dec_corrpay, 0) - ( a.dec_comprate - 1 )
* ISNULL(b.dec_deduamt, 0) dec_corramt ,
a.dec_qty - ISNULL(b.dec_qty, 0) - ISNULL(b.dec_deduqty, 0) opr_qty ,
ISNULL(b.dec_pay, 0) dec_pay ,
ISNULL(b.dec_corrpay, 0) dec_corrpay ,
ISNULL(b.dec_deduqty, 0) dec_deduqty ,
ISNULL(b.dec_deduamt, 0) dec_deduamt ,
ISNULL(b.dec_qty, 0) dec_qty
FROM ( SELECT a.dfeesum_no ,
a.opr_amt ,
a.dec_camt ,
a.dec_comprate ,
a.dec_ramt ,
a.dec_qty
FROM ctlm8686 a
UNION ALL
SELECT a.dfeesum_no ,
a.dec_amt ,
a.dec_compamt ,
a.dec_comprate ,
a.dec_corramt ,
a.opr_qty
FROM dfeeapp_03 a
) a
LEFT JOIN ( SELECT dfeesum_no ,
SUM(dec_ramt) dec_pay ,
SUM(dec_corramt) dec_corrpay ,
SUM(dec_qty) dec_qty ,
SUM(dec_deduamt) dec_deduamt,
SUM(dec_deduqty) dec_deduqty
FROM dfeepay_03
WHERE flag_dedu = '1'
GROUP BY dfeesum_no
) b ON a.dfeesum_no = b.dfeesum_no
跑这个视图的查询语句，从原来的一分半钟降到一分钟，对于整个SQL而言，则从原来跑几分钟的直接10S出结果。
