SQL优化（一）
优化SQL (一)
今天开发同事提出一些SQL运行较慢，我一看执行结果很少。执行时间近14秒！不能忍受，对其进行优化！
|
|
执行时间 |
优化结果 |
| 优化前 |
Elapsed: 00:00:13.97 |
|
| 优化后 |
Elapsed: 00:00:00.07 |
性能提升近200倍 |
优化思想：
1、 驱动表数据最小化原则（from最后的表为驱动表在RBO规则中，CBO中会自动选择最小代价的表为驱动表）
2、 过滤条件最大化原则（where后过滤的数据依次越来越多）
3、 建立索引走索引
背景说明：
OLTP系统，对这几张有频繁的insert,update。所以暂时不对系统加索引优化。所以系统只进行SQL写法的调优。
实验：
下面在测试服务器上完全相同的数据及结构进行优化测试。
一、尝试优化SQL的写法
SQL1
select lf.*,ld.answer as tick_number,lt.answer as major_code,a.state from
(select lt.*,sf.state from skills.ns2012_state_first sf join skills.resultedit lt on lt.aclass='101' and sf.serialno=lt.serialno and lt.answer='12867' and lt.question=1) a
join skills.resultedit ld on ld.aclass=a.aclass and a.serialno=ld.serialno and ld.question=21
left join skills.resultedit lt on lt.question=20 and lt.aclass=a.aclass and a.serialno=lt.serialno
join skills.logininfo lf on lf.aclass=a.aclass and a.serialno=lf.serialno where lf.sign is null
and
lower(ld.answer) in (
select tick_number from (
select count(c.answer)as counts,(c.answer) as tick_number ,max(c.serialno)as serialno from (
select b.* from (select lt.* from skills.ns2012_state_first sf join skills.resultedit lt on lt.serialno=sf.serialno and lt.question='1' and lt.aclass='101' and lt.answer='12867') a
join (select lt.aclass,lt.serialno,lt.question,lower(lt.answer) as answer from skills.ns2012_state_first sf join skills.resultedit lt on lt.serialno=sf.serialno and lt.question='21' and lt.aclass='101' and lt.answer is not null ) b
on a.serialno=b.serialno and a.aclass=b.aclass
join skills.logininfo lf on lf.aclass=a.aclass and a.serialno=lf.serialno where lf.sign is null
)c group by c.answer
) where counts>=2
)
order by lower(ld.answer) desc;
优化后
SQL2
select a.indate,a.serialno,a.sip, ld.answer as tick_number,lt.answer as major_code,a.state from
(select lt.aclass,lt.serialno,lf.indate,lf.sip,sf.state from skills.ns2012_state_first sf
join skills.resultedit lt on lt.aclass=101 and sf.serialno=lt.serialno and lt.answer='12867' and lt.question=1
join skills.logininfo lf on lf.aclass=lt.aclass and lf.serialno=lt.serialno where lf.sign is null and to_char(lf.indate,'yyyy-MM-dd HH24:Mi:SS') between '2012-07-01 9:00:00' and '2012-07-30 09:00:00'
) a
join skills.resultedit ld on ld.aclass=a.aclass and a.serialno=ld.serialno and ld.question=21
left join skills.resultedit lt on lt.question=20 and lt.aclass=a.aclass and a.serialno=lt.serialno
join(
select tick_number from (
select count(c.answer)as counts,(c.answer) as tick_number ,max(c.serialno)as serialno from (
select x.aclass,x.serialno,n.answer from (select lt.aclass,lt.question,lt.serialno from skills.ns2012_state_first sf join skills.resultedit lt on lt.serialno=sf.serialno and lt.question='1' and lt.aclass=101 and lt.answer='12867') x
join (select lt.aclass,lt.serialno,lt.question,lower(lt.answer) as answer from skills.ns2012_state_first sf join skills.resultedit lt on lt.serialno=sf.serialno and lt.question='21' and lt.aclass=101 and lt.answer is not null ) n
on x.serialno=n.serialno and x.aclass=n.aclass
join skills.logininfo lf on lf.aclass=n.aclass and lf.serialno=n.serialno where lf.sign is null and to_char(lf.indate,'yyyy-MM-dd HH24:Mi:SS') between '2012-07-01 9:00:00' and '2012-07-30 09:00:00'
)c group by c.answer
) where counts>=2
) m on lower(ld.answer)=m.tick_number
order by lower(ld.answer) desc;
执行计划很多，这里只提取我们所需要的部分，执行计划均是2次执行后的记录。这个是频繁操作的！
| SQL |
SQL1 |
SQL2（优化写法后） |
| time |
Elapsed: 00:00:09.76 |
Elapsed: 00:00:00.08 |
| 2次执行的执行计划 |
0 recursive calls 0 db block gets 3113107 consistent gets 0 physical reads 0 redo size 19011 bytes sent via SQL*Net to client 689 bytes received via SQL*Net from client 22 SQL*Net roundtrips to/from client 1 sorts (memory) 0 sorts (disk) 315 rows processed
|
0 recursive calls 0 db block gets 12714 consistent gets 0 physical reads 0 redo size 16918 bytes sent via SQL*Net to client 741 bytes received via SQL*Net from client 22 SQL*Net roundtrips to/from client 1 sorts (memory) 0 sorts (disk) 315 rows processed
|
二、添加索引
针对aclass里面出现最多的列，添加了索引进行尝试。并寻找重复最少的列进行加索引。
SQL> select index_name,table_name,status,ITYP_NAME from user_indexes where table_owner='SKILLS';
INDEX_NAME TABLE_NAME STATUS ITYP_NAME
-------------------- -------------------- ---------- ----------
PK_RESULTEDIT RESULTEDIT VALID
INDEX_2012 NS2012_STATE_FIRST VALID
LOG_INDEX LOGININFO VALID
红色部分为后来添加的索引
添加索引后，然后执行SQL1和SQL2，看看执行效果，从执行时间上看我们没有优化。
这个也说明了根据执行计划来建立索引进行优化效果明显，但是对于一个OLTP系统来说，以后用来维护索引的资源也是必须要考虑的，所以最后还是采用了优化写法，没有对系统加索引。
| SQL |
SQL1 |
SQL2（优化写法后） |
| time |
Elapsed: 00:00:00.05 |
Elapsed: 00:00:00.09 |
| 2次执行的执行计划 |
0 recursive calls 0 db block gets 9154 consistent gets 0 physical reads 0 redo size 18957 bytes sent via SQL*Net to client 689 bytes received via SQL*Net from client 22 SQL*Net roundtrips to/from client 1 sorts (memory) 0 sorts (disk) 315 rows processed
|
0 recursive calls 0 db block gets 20655 consistent gets 0 physical reads 0 redo size 16914 bytes sent via SQL*Net to client 741 bytes received via SQL*Net from client 22 SQL*Net roundtrips to/from client 1 sorts (memory) 0 sorts (disk) 315 rows processed
|
总结SQL优化：
1、 优化SQL先优化写法及逻辑思维。（尽量减少不必要访问表数据和排序！根据filter，考虑过滤条件）
2、 添加相应的索引（根据access，考虑加索引）