MSQL优化基础(frequence直方图选择率)
2015-07-03 09:24
330 查看
准备数据
drop table t1; create table t1(n1 number); begin for i in 1..100 loop for j in i*5..100 loop insert into t1 values(i); end loop; end loop; end ; / commit; //收集数据 exec dbms_stats.gather_table_stats(null,'T1',method_opt=>'for all columns size 20'); //查询直方图 select ENDPOINT_VALUE, ENDPOINT_NUMBER, ENDPOINT_NUMBER-nvl(lag(ENDPOINT_NUMBER) over(order by endpoint_value),0) backet_size from user_histograms where table_name='T1' and column_name='N1'; ENDPOINT_VALUE ENDPOINT_NUMBER BACKET_SIZE -------------- --------------- ----------- 1 96 96 2 187 91 3 273 86 4 354 81 5 430 76 6 501 71 7 567 66 8 628 61 9 684 56 10 735 51 11 781 46 12 822 41 13 858 36 14 889 31 15 915 26 16 936 21 17 952 16 18 963 11 19 969 6 20 970 1 20 rows selected. // 查询行数 SQL> select num_rows from user_tables where table_name='T1'; NUM_ROWS ----------
970 //查询最大最小值 select column_name,num_distinct,low_value,high_value,num_nulls,HISTOGRAM from user_tab_columns where table_name='T1' and COLUMN_NAME='N1'; COLUMN_NAM NUM_DISTINCT LOW_VALUE HIGH_VALUE NUM_NULLS HISTOGRAM ---------- ------------ ---------- ---------- --------- ---------- N1 20 C102(1) C115(20) 0 FREQUENCY
注意：null值不会进直方图
案例一 = (in-range) frequency直方图
有自己的bucket 选择率 =BucketSize /NumRows没有自己的bucket 选择率 =最小BucketSize/2/NumRows
最小BucketSize如果很小的话 出的问题很高
select * from t1 where n1=12; select 41/970*970 = 41 -------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | -------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 41 | 123 | 2 (0)| 00:00:01 | |* 1 | TABLE ACCESS FULL| T1 | 41 | 123 | 2 (0)| 00:00:01 | -------------------------------------------------------------------------- select * from t1 where n1=12.5; select 1/2/970*970 from dual; -------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | -------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1 | 3 | 2 (0)| 00:00:01 | |* 1 | TABLE ACCESS FULL| T1 | 1 | 3 | 2 (0)| 00:00:01 | --------------------------------------------------------------------------
案例二 >,<,<=,>= (in-range) frequency直方图
>,<,<=,>=选择率： (sum(BucketSize)) /NumRowsbetween 选择率： (sum(Bucketsize)) /Numrows
<pre class="sql" name="code">select * from t1 where n1> 17 -------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | -------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 17 | 51 | 2 (0)| 00:00:01 | |* 1 | TABLE ACCESS FULL| T1 | 17 | 51 | 2 (0)| 00:00:01 | -------------------------------------------------------------------------- select (11+6+1 )/970 * 970 from dual; 18 差一点点 情况不清， 最多差1 问题不大
案例三 如何验证是否bind peek
var b1 number; exec :b1:=50 alter session set events '10053 trace name context forever ,level 1' select * from t1 where n1=:b1; alter session set events '10053 trace name context off'; SQL> show parameter dump ; NAME TYPE VALUE ------------------------------------ ----------- ------------------------------ background_core_dump string partial background_dump_dest string /opt/ora10g/admin/RACDB/bdump core_dump_dest string /opt/ora10g/admin/RACDB/cdump max_dump_file_size string UNLIMITED shadow_core_dump string partial user_dump_dest string /opt/ora10g/admin/RACDB/udump/opt/ora10g/admin/RACDB/udump cd /opt/ora10g/admin/RACDB/udump/opt/ora10g/admin/RACDB/udump 找到 trace 文件 搜索 bind Peeked values of the binds in SQL statement ******************************************* kkscoacd Bind#0 oacdty=02 mxl=22(22) mxlc=00 mal=00 scl=00 pre=00 oacflg=03 fl2=1000000 frm=00 csi=00 siz=24 off=0 kxsbbbfp=b7ed88d0 bln=22 avl=02 flg=05 value=50 从中可以看出 绑定 的值是50 ,这一段说明了 做了bind peek alter session set "_optim_peek_user_binds"=false; alter session set events '10053 trace name context forever ,level 1' select /****/ * from t1 where n1=:b1; alter session set events '10053 trace name context off' trace 文件中 这一节要看一下，这里列出来了 被改过的优化器参数 ************************************* PARAMETERS WITH ALTERED VALUES ****************************** _optim_peek_user_binds = false ************************************* PARAMETERS WITH DEFAULT VALUES 以下内容说明木有做bind peeks ******************************************* Peeked values of the binds in SQL statement ******************************************* kkscoacd Bind#0 oacdty=02 mxl=22(22) mxlc=00 mal=00 scl=00 pre=00 oacflg=03 fl2=1000000 frm=00 csi=00 siz=24 off=0 No bind buffers allocated *************************************** select * from t1 where n1=50; -------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | -------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1 | 3 | 2 (0)| 00:00:01 | |* 1 | TABLE ACCESS FULL| T1 | 1 | 3 | 2 (0)| 00:00:01 | -------------------------------------------------------------------------- alter session set "_optim_peek_user_binds"=false; var b1 number; exec :b1:=50 select * from t1 where n1=:b1; -------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | -------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 49 | 147 | 2 (0)| 00:00:01 | |* 1 | TABLE ACCESS FULL| T1 | 49 | 147 | 2 (0)| 00:00:01 | -------------------------------------------------------------------------- 不管b1绑定什么 都是 49 禁用了bind peek 都是这样来算， 不知道 用直方图， 不知道out-range 的情况
案例四 不做 bind peeking
=谓词 (1/NDV)*非空率>,<,<=,>= 0.05*非空率
between 谓词 0.0025*非空率
not 谓词 (1-0.05)*非空率
alter session set "_optim_peek_user_binds"=false;
select * from t1 where n1=:b1; -------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | -------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 49 | 147 | 15 (0)| 00:00:01 | |* 1 | TABLE ACCESS FULL| T1 | 49 | 147 | 15 (0)| 00:00:01 | -------------------------------------------------------------------------- select 0.05 * 970 from dual; 48.5 select * from t1 where n1>:b1 -------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | -------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 49 | 147 | 15 (0)| 00:00:01 | |* 1 | TABLE ACCESS FULL| T1 | 49 | 147 | 15 (0)| 00:00:01 | -------------------------------------------------------------------------- select 0.05 * 970 from dual; 48.5 select * from t1 where n1 between :b1 and :b2; --------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | --------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 2 | 6 | 15 (0)| 00:00:01 | |* 1 | FILTER | | | | | | |* 2 | TABLE ACCESS FULL| T1 | 2 | 6 | 15 (0)| 00:00:01 | --------------------------------------------------------------------------- select * from t1 where n1<> :b1 ; (1-0.05)*1*970 = 921.5 -------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | -------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 922 | 2766 | 15 (0)| 00:00:01 | |* 1 | TABLE ACCESS FULL| T1 | 922 | 2766 | 15 (0)| 00:00:01 | --------------------------------------------------------------------------
案例五 out-of-range
MIN(BucketSize)/2/NumRows * ((1-(val-high_value)/(high_value-low_value)))//准备数据 begin for i in 1..20 loop insert into t1 values(19); end loop; end; commit; exec dbms_stats.gather_table_stats(null,'T1',method_opt=>'for all columns size 20'); select num_rows from user_tables where table_name='T1'; NUM_ROWS ---------- 2970 //查询直方图 select ENDPOINT_VALUE, ENDPOINT_NUMBER, ENDPOINT_NUMBER-nvl(lag(ENDPOINT_NUMBER) over(order by endpoint_value),0) backet_size from user_histograms where table_name='T1' and column_name='N1'; ENDPOINT_VALUE ENDPOINT_NUMBER BACKET_SIZE -------------- --------------- ----------- 1 96 96 2 187 91 3 273 86 4 354 81 5 430 76 6 501 71 7 567 66 8 628 61 9 684 56 10 735 51 11 781 46 12 822 41 13 858 36 14 889 31 15 915 26 16 936 21 17 952 16 18 963 11 19 969 6 20 2970 2001 select /***^^^****/* from t1 where n1=25; -------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | -------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 1 | 3 | 3 (0)| 00:00:01 | |* 1 | TABLE ACCESS FULL| T1 | 1 | 3 | 3 (0)| 00:00:01 | -------------------------------------------------------------------------- 按照公式是算得2 ，不知是不是和版本有关， 测试版本是 10.2g (6/2/2970)* ((1-(25-20)/(20-1))) * 2970=2.21052632
补充 函数索引
drop table t2; create table t2 as select * from dba_objects; select distinct status from t2; --两个值 STATUS ------- INVALID VALID select count(*) , status from t2 group by status; COUNT(*) STATUS ---------- ------- 1 INVALID 46423 VALID exec dbms_stats.gather_table_stats(null, 'T2'); //如果要查询 INVALID 的值, INVALID 的值又少，可以考虑建函数索引 create index indx_t2 on t2 (status); //常规索引 create index indx1_t2 on t2 (decode (status, 'VALID',null,'INVALID')); // 函数索引 select bytes /1024/1024 from user_segments where segment_name ='INDX_T2'; //常规索引占的大小 BYTES/1024/1024 --------------- .9375 select bytes /1024/1024 from user_segments where segment_name ='INDX1_T2'; //函数索引占的大小 BYTES/1024/1024 --------------- .0625 drop index indx_t2 //删掉 常规索引 select * from t2 where decode(status, 'VALID',null,'INVALID')='INVALID' ---------------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | ---------------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 464 | 43152 | 2 (0)| 00:00:01 | | 1 | TABLE ACCESS BY INDEX ROWID| T2 | 464 | 43152 | 2 (0)| 00:00:01 | |* 2 | INDEX RANGE SCAN | INDX1_T2 | 1 | | 1 (0)| 00:00:01 | ---------------------------------------------------------------------------------------- 走了函数索引
应用中 尽量少用 truncate , 使用truncate ，使这张表相关打开的共享游标都失效！！
频繁使用 truncate , 别人又在查询时候，可能会使数据库hang住应用中 尽量少用 truncate , 使用truncate ，使这张表相关打开的共享游标都失效！！
频繁使用 truncate , 别人又在查询时候，可能会使数据库hang住
optizer_max_permutations 各种访问路径 尝试最多的次数
in(12,13,14,15...) 会转换成 or-expansion
内容来自用户分享和网络整理，不保证内容的准确性，如有侵权内容，可联系管理员处理 
相关文章推荐
- Kendo UI k-template 的使用
- projecteuler---->problem=14----Longest Collatz sequence
- 详解 UIView 的 Tint Color 属性
- 用 OpenCV 实现 Guided Filter
- StringBuilder is mutable
- UINavigationController纯代码实现细节提示
- 记录:sea.js和require.js配置 与 性能对比
- NGUI长按事件
- Google interview question: count bounded slices(min/max queue)
- 黑马程序员----java的GUI图形用户界面及编程基础
- LeetCode Unique Binary Search Trees II
- LeetCode Unique Binary Search Trees
- UIView用户事件响应
- Liferay 6.1开发学习（十九）：Liferay ServiceBuilder之自定义查询
- Liferay 6.1开发学习（二十）：Dynamic Query高级查询
- 去除UITableView多余分割线
- UILabel 字体超出 frame 自动缩小
- 解决iOS7和iOS8的UITableViewCell的分割线右移问题
- UITableView去除多余分割线
- ARP协议详解之Gratuitous ARP（免费ARP）