今天闲的 看awr，发现一条SQL 每次执行40多秒，语句如下
?
1
SELECT a.bill_class AS pol_code , b.bill_name AS pol_name , a.bill_no AS card_no , '网站' AS buy_path , a.rev_clerk_code AS agent_code , a.rev_clerk_type AS agent_type , to_char(a.regist_date,'yyyy-mm-dd') AS recip_date , to_char(a.chk_date,'yyyy-mm-dd') AS pay_date , to_char(a.invalid_date,'yyyy-mm-dd') AS invalid_date , CASE WHEN chk_stat = '0' THEN '未核销' WHEN chk_stat = '1' AND autochkflag = '1' THEN '已自动核销' WHEN chk_stat = '1' AND autochkflag = '0' THEN '已人工核销' WHEN chk_stat = '2' THEN '丢失' WHEN chk_stat = '3' THEN '作废' WHEN chk_stat = '4' THEN '回缴核销' WHEN chk_stat = '5' THEN '已销毁' WHEN chk_stat = 'A' THEN '待核销' END AS recip_stat , rev_branch_no ,b.bill_literal_price AS prem FROM shcvms.bill_grant_check a ,shcvms.bill_class b WHERE a.regist_date IS NOT NULL AND a.bill_class = b.bill_class AND a.bill_class IN ( '1001', '1093', '1096', '1097', '1098', '1099', '1100', '1302', '1303', '1305', '1306', '1910', '1911', '1912', '1913', '1914', '1915', '1916', '1917' , '1918', '1919', '1922', '1923', '1924', '1925', '1926', '1927', '1928', '1929', '1930', '1931', '1935', '1936', '1937', '1938', '1939', '1940', '1941', '1942', '1943', '1944', '1945', '5232', '5233', '5234', '5252', '5255', '5258', '5265', '5260', '5276', '5277', '5278', '5285', '5290', '5301', '5304', '5309', '5311', '5329', '5330', '5359' , '5376', '5507', '5512', '5514', '5516', '5122', '5126', '5130', '5132', '5141', '5142', '5156', '5157', '5161', '5162', '5163', '5164', '5165', '5166', '5167', '5168', '5558', '5564', '5507' )
看这种SQL，一种方法，是通过PL/SQL DEV直接格式化下，第二种方法，放到txt文件中，搜索FROM，直接把前面的全删了(非标量子查询时)。
SELECT ……
FROM shcvms.bill_grant_check a , shcvms.bill_class b
WHERE a.regist_date IS NOT NULL AND
a.bill_class = b.bill_class
AND a.bill_class IN ( ……);
这里也没有绑定变量，所以explain plan for的执行计划是没有误差的
当前执行计划如下
?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
PLAN_TABLE_OUTPUT
-----------------------------------------------------------------------------------------------------------------------------------------
Plan hash value: 2990887684
---------------------------------------------------------------------------------------
| Id| Operation| Name| Rows| Bytes | Cost (%CPU)| Time|
---------------------------------------------------------------------------------------
|0 | SELECT STATEMENT||6452K|553M|151K(5)| 00:30:19 |
|*1 |HASH JOIN||6452K|553M|151K(5)| 00:30:19 |
|*2 |TABLE ACCESS FULL| BILL_CLASS|206 |5150 |3(0)| 00:00:01 |
|*3 |TABLE ACCESS FULL| BILL_GRANT_CHECK |7321K|453M|151K(5)| 00:30:19 |
---------------------------------------------------------------------------------------
Predicate Information (identifiedby operation id):
---------------------------------------------------
1 - access("A"."BILL_CLASS"="B"."BILL_CLASS")
2 - filter("B"."BILL_CLASS"='1001' OR "B"."BILL_CLASS"='1093' OR
"B"."BILL_CLASS"='1096' OR "B"."BILL_CLASS"='1097' OR "B"."BILL_CLASS"='1098'
OR "B"."BILL_CLASS"='1099' OR "B"."BILL_CLASS"='1100' OR
"B"."BILL_CLASS"='1302' OR "B"."BILL_CLASS"='1303' OR "B"."BILL_CLASS"='1
。
。
。
。
这个看起来很简单的，7321k=7.3M 7.3M/453M =1.6%这个结果集，完全可以走索引，而且这里使用的in，并且里面用的是常量，完全可以走 INList ITERATOR
我创建一个虚索引，看一下执行计划，如果效果确实不错，再提个申请，真实创建这个索引。
?
1
2
3
4
5
6
7
8
SQL>
create index billgrant_class on shcvms.bill_grant_check(bill_class)nosegment;
Index created.
SQL> alter session set "_use_nosegment_indexes"=true;
Sessionaltered.
然后使用explain plan for 来解析执行计划
?
1
2
3
4
5
explain plan for SELECT /*+ index(a billgrant_class )*/ N多个列
FROM shcvms.bill_grant_check a ,shcvms.bill_class b
WHERE a.regist_date IS NOT NULL AND
a.bill_class = b.bill_class
AND a.bill_class IN (......N多常量)
我们看看执行计划
?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
PLAN_TABLE_OUTPUT
-----------------------------------------------------------------------------------------------------------------------------------------
Plan hash value: 2116188717
--------------------------------------------------------------------------------------------------
| Id| Operation|Name| Rows| Bytes | Cost (%CPU)| Time|
--------------------------------------------------------------------------------------------------
|0 | SELECT STATEMENT||6452K|553M|207(1)| 00:00:03 |
|1 |MERGE JOIN||6452K|553M|207(1)|00:00:03 |
|2 |INLIST ITERATOR||||||
|*3 |TABLE ACCESS BY INDEX ROWID| BILL_GRANT_CHECK |7321K|453M|203(1)| 00:00:03 |
|*4 |INDEX RANGE SCAN| BILLGRANT_CLASS|7321K||16(0)| 00:00:01 |
|*5 |SORT JOIN||206 |5150 |4(25)|00:00:01 |
|*6 |TABLE ACCESS FULL| BILL_CLASS|206 |5150 |3(0)| 00:00:01 |
--------------------------------------------------------------------------------------------------
Predicate Information (identifiedby operation id):
---------------------------------------------------
3 - filter("A"."REGIST_DATE" IS NOT NULL)
4 - access("A"."BILL_CLASS"='1001' OR "A"."BILL_CLASS"='1093' OR
"A"."BILL_CLASS"='1096' OR "A"."BILL_CLASS"='1097' OR "A"."BILL_CLASS"='1098' OR
"A"."BILL_CLASS"='1099' OR "A"."BILL_CLASS"='1100' OR "A"."BILL_CLASS"='1302' OR
看Cost。从之前的151k 降到了现在的207，速度提升为原来的1/1000。 在统计信息正确的情况下cost是具有参考价值的(不正确的话就别看cost了)
现在可以提申请了。真实创建这个索引。
一个简单的例子完事，(虽然这么简单的，不想往这贴，但是我blog中“Oracle优化之SQL 优化”分支太空了，写点东西充实一下它)
