SQL优化一例

网络整理 - 08-18

　　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 (identified by 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; Session altered.

　　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 (identified by 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)