在索引列上使用函数使得索引失效的是常见的索引失效原因之一,因此尽可能的避免在索引列上使用函数。尽管可以使用基于函数的索引来解决索引失效的问题,但如此一来带来的比如磁盘空间的占用以及列上过多的索引导致dml性能的下降。本文描述的是一个索引列上使用函数使其失效的案例。
一、数据版本与原始语句及相关信息
1.版本信息
sql> select * from v$version; banner ---------------------------------------------------------------- oracle database 10g release 10.2.0.3.0 - 64bit production pl/sql release 10.2.0.3.0 - production core 10.2.0.3.0 production tns for linux: version 10.2.0.3.0 - production nlsrtl version 10.2.0.3.0 - production
2.原始语句与其执行计划
sql> set autotrace traceonly exp; select acc_num, curr_cd, decode('20110728', (select to_char(last_day(to_date('20110728', 'yyyymmdd')), 'yyyymmdd') from dual), 0, adj_credit_int_lv1_amt + adj_credit_int_lv2_amt - adj_debit_int_lv1_amt - adj_debit_int_lv2_amt) as interest from acc_pos_int_tbl acc_pos_int_tbl1 where substr(business_date, 1, 6) = substr('20110728', 1, 6) and business_date <= '20110728'; execution plan ---------------------------------------------------------- plan hash value: 3114115399 ------------------------------------------------------------------------------------- | id | operation | name | rows | bytes | cost (%cpu)| time | ------------------------------------------------------------------------------------- | 0 | select statement | | 336k| 12m| 96399 (1)| 00:19:17 | | 1 | fast dual | | 1 | | 2 (0)| 00:00:01 | |* 2 | table access full| acc_pos_int_tbl | 336k| 12m| 96399 (1)| 00:19:17 | ------------------------------------------------------------------------------------- predicate information (identified by operation id): --------------------------------------------------- 2 - filter(substr("business_date",1,6)='201107' and "business_date"<='20110728')
从执行计划可以看出,sql语句使用了全表扫描,而where 子句中只有唯一的一列business_date
3.表上的索引信息
sql> set autotrace off; sql> set linesize 190 sql> @idx_info enter value for owner: goex_admin old 10: and owner = upper('&owner') new 10: and owner = upper('goex_admin') enter value for table_name: acc_pos_int_tbl old 11: and a.table_name = upper('&table_name') new 11: and a.table_name = upper('acc_pos_int_tbl') table_name index_name col_nam cl_pos status idx_typ dscd ------------------ ------------------------ -------------------- ------ -------- --------------- ---- acc_pos_int_tbl acc_pos_int_10dig_idx sys_nc00032$ 1 valid function-based asc normal acc_pos_int_tbl acc_pos_int_10dig_idx business_date 2 valid function-based asc normal acc_pos_int_tbl acc_pos_int_10dig_idx curr_cd 3 valid function-based asc normal acc_pos_int_tbl pk_acc_pos_int_tbl acc_num 1 valid normal asc acc_pos_int_tbl pk_acc_pos_int_tbl business_date 2 valid normal asc
从索引的情况上来看有一个基于主键的索引包含了business_date列,而查询语句并没有走索引而是选择的全表扫描,而且预估所返回的行rows与bytes也是大的惊人,cost的值96399,接近10w。
二、分析与改造sql语句
1.原始的sql语句分析
sql语句中where子句的business_date列实现对记录过滤
business_date <= ‘20110728’条件不会限制索引的使用
substr(business_date, 1, 6) = substr(‘20110728’, 1, 6)使用了substr函数,限制了优化器选择索引
基于business_date列来建立索引函数,从已存在的索引来看,必要性不大
2.改造sql语句
substr(business_date, 1, 6) = substr(‘20110728’, 1, 6)的实质是等于当月,即限制返回的行为从2011.7.1日至2011.7.28
因此其返回的记录大于等于2011.7.1,且小于2011.7.28
做如下改造
business_date >=to_char(last_day(add_months(to_date(‘20110728′,’yyyymmdd’),-1)) + 1,’yyyymmdd’)
3.改造后的sql语句
select acc_num, curr_cd, decode('20110728', (select to_char(last_day(to_date('20110728', 'yyyymmdd')), 'yyyymmdd') from dual), 0, adj_credit_int_lv1_amt + adj_credit_int_lv2_amt - adj_debit_int_lv1_amt - adj_debit_int_lv2_amt) as interest from acc_pos_int_tbl acc_pos_int_tbl1 where business_date >= to_char(last_day(add_months(to_date('20110728', 'yyyymmdd'), -1)) + 1, 'yyyymmdd') and business_date <= '20110728';
4.改造后的执行计划
execution plan ---------------------------------------------------------- plan hash value: 66267922 -------------------------------------------------------------------------------------------------- | id | operation | name | rows | bytes | cost (%cpu)| time | -------------------------------------------------------------------------------------------------- | 0 | select statement | | 1065k| 39m| 75043 (1)| 00:15:01 | | 1 | fast dual | | 1 | | 2 (0)| 00:00:01 | | 2 | table access by index rowid| acc_pos_int_tbl | 1065k| 39m| 75043 (1)| 00:15:01 | |* 3 | index skip scan | pk_acc_pos_int_tbl | 33730 | | 41180 (1)| 00:08:15 | -------------------------------------------------------------------------------------------------- predicate information (identified by operation id): --------------------------------------------------- 3 - access("business_date">='20110701' and "business_date"<='20110728') filter("business_date">='20110701' and "business_date"<='20110728')
改造后可以看到sql语句的执行计划已经由原来的全表扫描改为执行index skip scan,但其cost也并没有降低多少
三、进一步分析
1.表的相关信息
sql> @tab_stat enter value for input_table_name: acc_pos_int_tbl old 11: where table_name = upper('&input_table_name') new 11: where table_name = upper('acc_pos_int_tbl') enter value for input_owner: goex_admin old 12: and owner = upper('&input_owner') new 12: and owner = upper('goex_admin') num_rows blks em_blks avg_space chain_cnt avg_row_len avg_rows_per_block lst_anly sta ---------- ---------- ---------- ---------- ---------- ----------- ------------------ --------- --- 33659947 437206 1322 855 0 99 77 27-sep-11 no
2.索引的相关信息
sql> @idx_stat enter value for input_table_name: acc_pos_int_tbl old 11: where table_name = upper('&input_table_name') new 11: where table_name = upper('acc_pos_int_tbl') enter value for input_owner: goex_admin old 12: and owner = upper('&input_owner') new 12: and owner = upper('goex_admin') blev idx_name lf_blks dst_keys num_rows lf_per_key dat_blk_per_key clus_fct lst_anly ---- ------------------------------ ---------- ---------- ---------- ---------- --------------- ---------- --------- 3 pk_acc_pos_int_tbl 155658 33777720 33777720 1 1 33777447 27-sep-11 3 acc_pos_int_10dig_idx 160247 32850596 32850596 1 1 32763921 27-sep-11
3.尝试在business_date列上创建索引
sql> create index i_acc_pos_int_tbl_bs_dt on acc_pos_int_tbl(business_date) tablespace tbs_tmp nologging; index created. sql> @idx_stat enter value for input_table_name: acc_pos_int_tbl old 11: where table_name = upper('&input_table_name') new 11: where table_name = upper('acc_pos_int_tbl') enter value for input_owner: goex_admin old 12: and owner = upper('&input_owner') new 12: and owner = upper('goex_admin') blev idx_name lf_blks dst_keys num_rows lf_per_key dat_blk_per_key clus_fct lst_anly ---- ------------------------------ ---------- ---------- ---------- ---------- --------------- ---------- --------- 2 i_acc_pos_int_tbl_bs_dt 93761 908 33659855 103 506 460007 30-sep-11 3 pk_acc_pos_int_tbl 155658 33777720 33777720 1 1 33777447 27-sep-11 3 acc_pos_int_10dig_idx 160247 32850596 32850596 1 1 32763921 27-sep-11
建立索引后聚簇因子较小,差不多接近表上块的数量
4.使用新创建索引后的执行计划
execution plan ---------------------------------------------------------- plan hash value: 2183566226 ------------------------------------------------------------------------------------------------------- | id | operation | name | rows | bytes | cost (%cpu)| time | ------------------------------------------------------------------------------------------------------- | 0 | select statement | | 1065k| 39m| 17586 (1)| 00:03:32 | | 1 | fast dual | | 1 | | 2 (0)| 00:00:01 | | 2 | table access by index rowid| acc_pos_int_tbl | 1065k| 39m| 17586 (1)| 00:03:32 | |* 3 | index range scan | i_acc_pos_int_tbl_bs_dt | 1065k| | 2984 (1)| 00:00:36 | ------------------------------------------------------------------------------------------------------- predicate information (identified by operation id): --------------------------------------------------- 3 - access("business_date">='20110701' and "business_date"<='20110728')
从上面的执行计划看出,sql语句已经选择了新建的索引尽管返回的rows,bytes没有明显的变化,但cost已经少了近7倍。
以上所述是www.887551.com给大家介绍的oracle函数使索引列失效的解决办法,希望对大家有所帮助。在此也非常感谢大家对www.887551.com网站的支持!