如何让in/exists 子查询(半连接)作为驱动表？
一哥们问我，怎么才能让子查询作为驱动表？ SQL如下：
select rowid rid
from its_car_pass7 v
where 1 = 1
and pass_datetime >=
to_date('2013-07-06 :17:46:04', 'yyyy-mm-dd hh24:mi:ss')
and pass_datetime <=
to_date('2013-07-06 :18:46:06', 'yyyy-mm-dd hh24:mi:ss')
and v.pass_device_unid in
(select unid
from its_base_device
where dev_bay_unid in ('01685EFE4658C19D59C4DDAAEDD37393')
and dev_type = '1'
and dev_chk_flag = '1'
and dev_delete_flag = 'N')
order by v.pass_datetime asc
/
执行计划如下：
Execution Plan
----------------------------------------------------------
Plan hash value: 3634433140 --------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | Pstart| Pstop |
--------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 111 | 2 (50)| 00:00:01 | | |
| 1 | SORT ORDER BY | | 1 | 111 | 2 (50)| 00:00:01 | | |
| 2 | NESTED LOOPS | | | | | | | |
| 3 | NESTED LOOPS | | 1 | 111 | 1 (0)| 00:00:01 | | |
| 4 | PARTITION RANGE SINGLE | | 1 | 39 | 1 (0)| 00:00:01 | 1284 | 1284 |
|* 5 | INDEX SKIP SCAN | IDX_VT7_DEVICEID | 1 | 39 | 1 (0)| 00:00:01 | 1284 | 1284 |
|* 6 | INDEX UNIQUE SCAN | PK_ITS_BASE_DEVICE | 1 | | 0 (0)| 00:00:01 | | |
|* 7 | TABLE ACCESS BY INDEX ROWID| ITS_BASE_DEVICE | 1 | 72 | 0 (0)| 00:00:01 | | |
-------------------------------------------------------------------------------------------------------------------- Predicate Information (identified by operation id):
--------------------------------------------------- 5 - access("PASS_DATETIME">=TO_DATE(' 2013-07-06 17:46:04', 'syyyy-mm-dd hh24:mi:ss') AND
"PASS_DATETIME"<=TO_DATE(' 2013-07-06 18:46:06', 'syyyy-mm-dd hh24:mi:ss'))
filter("PASS_DATETIME">=TO_DATE(' 2013-07-06 17:46:04', 'syyyy-mm-dd hh24:mi:ss') AND
"PASS_DATETIME"<=TO_DATE(' 2013-07-06 18:46:06', 'syyyy-mm-dd hh24:mi:ss'))
6 - access("V"."PASS_DEVICE_UNID"="UNID")
7 - filter("DEV_BAY_UNID"='01685EFE4658C19D59C4DDAAEDD37393' AND "DEV_TYPE"='1' AND
"DEV_DELETE_FLAG"='N' AND "DEV_CHK_FLAG"='1') Statistics
----------------------------------------------------------
1 recursive calls
0 db block gets
110973 consistent gets
0 physical reads
0 redo size
47861 bytes sent via SQL*Net to client
1656 bytes received via SQL*Net from client
105 SQL*Net roundtrips to/from client
1 sorts (memory)
0 sorts (disk)
1560 rows processed
这里我们就不管统计信息是否准确了，也不管SQL优化的问题，就单单讨论哥们问的问题吧。
那哥们说，怎么才能让子查询作为驱动表呢？他自己试了很多方法就是搞不定。 那我们来亲自搞搞吧
explain plan for select rowid rid
from its_car_pass7 v
where 1 = 1
and pass_datetime >=
to_date('2013-07-06 :17:46:04', 'yyyy-mm-dd hh24:mi:ss')
and pass_datetime <=
to_date('2013-07-06 :18:46:06', 'yyyy-mm-dd hh24:mi:ss')
and v.pass_device_unid in
(select unid
from its_base_device
where dev_bay_unid in ('01685EFE4658C19D59C4DDAAEDD37393')
and dev_type = '1'
and dev_chk_flag = '1'
and dev_delete_flag = 'N')
order by v.pass_datetime asc
/
执行计划如下
SELECT * FROM TABLE(DBMS_XPLAN.DISPLAY(NULL, NULL, 'ADVANCED -PROJECTION')); -----------------------------------------------------------
Plan hash value: 2191740724
---------------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | Pstart| Pstop |
---------------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 111 | 2092K (1)| 06:58:26 | | |
| 1 | NESTED LOOPS | | | | | | | |
| 2 | NESTED LOOPS | | 1 | 111 | 2092K (1)| 06:58:26 | | |
| 3 | PARTITION RANGE SINGLE | | 1 | 39 | 2092K (1)| 06:58:26 | 1284 | 1284 |
| 4 | TABLE ACCESS BY LOCAL INDEX ROWID| ITS_CAR_PASS7 | 1 | 39 | 2092K (1)| 06:58:26 | 1284 | 1284 |
|* 5 | INDEX RANGE SCAN | IDX_VT7_DATETIME | 1 | | 6029 (1)| 00:01:13 | 1284 | 1284 |
|* 6 | INDEX UNIQUE SCAN | PK_ITS_BASE_DEVICE | 1 | | 0 (0)| 00:00:01 | | |
|* 7 | TABLE ACCESS BY INDEX ROWID | ITS_BASE_DEVICE | 1 | 72 | 0 (0)| 00:00:01 | | |
--------------------------------------------------------------------------------------------------------------------------- Query Block Name / Object Alias (identified by operation id):
------------------------------------------------------------- 1 - SEL$5DA710D3
4 - SEL$5DA710D3 / V@SEL$1
5 - SEL$5DA710D3 / V@SEL$1
6 - SEL$5DA710D3 / ITS_BASE_DEVICE@SEL$2
7 - SEL$5DA710D3 / ITS_BASE_DEVICE@SEL$2 Outline Data
------------- /*+
BEGIN_OUTLINE_DATA
NLJ_BATCHING(@"SEL$5DA710D3" "ITS_BASE_DEVICE"@"SEL$2")
USE_NL(@"SEL$5DA710D3" "ITS_BASE_DEVICE"@"SEL$2")
LEADING(@"SEL$5DA710D3" "V"@"SEL$1" "ITS_BASE_DEVICE"@"SEL$2")
INDEX(@"SEL$5DA710D3" "ITS_BASE_DEVICE"@"SEL$2" ("ITS_BASE_DEVICE"."UNID"))
INDEX_RS_ASC(@"SEL$5DA710D3" "V"@"SEL$1" ("ITS_CAR_PASS7"."PASS_DATETIME"))
OUTLINE(@"SEL$2")
OUTLINE(@"SEL$1")
UNNEST(@"SEL$2")
OUTLINE_LEAF(@"SEL$5DA710D3")
FIRST_ROWS
DB_VERSION('11.2.0.3')
OPTIMIZER_FEATURES_ENABLE('11.2.0.3')
IGNORE_OPTIM_EMBEDDED_HINTS
END_OUTLINE_DATA
*/ Predicate Information (identified by operation id):
--------------------------------------------------- 5 - access("PASS_DATETIME">=TO_DATE(' 2013-07-06 17:46:04', 'syyyy-mm-dd hh24:mi:ss') AND
"PASS_DATETIME"<=TO_DATE(' 2013-07-06 18:46:06', 'syyyy-mm-dd hh24:mi:ss'))
6 - access("V"."PASS_DEVICE_UNID"="UNID")
7 - filter("DEV_BAY_UNID"='01685EFE4658C19D59C4DDAAEDD37393' AND "DEV_TYPE"='1' AND "DEV_DELETE_FLAG"='N' AND
"DEV_CHK_FLAG"='1')
TMD 执行计划又变了，我们也先别管执行计划为啥变了，驱动表仍然是 ITS_CAR_PASS7，现在我们来改变驱动表
select /*+ leading(ITS_BASE_DEVICE@SEL$2) */ rowid rid
from its_car_pass7 v
where 1 = 1
and pass_datetime >=
to_date('2013-07-06 :17:46:04', 'yyyy-mm-dd hh24:mi:ss')
and pass_datetime <=
to_date('2013-07-06 :18:46:06', 'yyyy-mm-dd hh24:mi:ss')
and v.pass_device_unid in
(select unid
from its_base_device
where dev_bay_unid in ('01685EFE4658C19D59C4DDAAEDD37393')
and dev_type = '1'
and dev_chk_flag = '1'
and dev_delete_flag = 'N')
order by v.pass_datetime asc
/ Execution Plan
----------------------------------------------------------
Plan hash value: 712001411 -------------------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | Pstart| Pstop |
-------------------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 111 | 25 (4)| 00:00:01 | | |
| 1 | SORT ORDER BY | | 1 | 111 | 25 (4)| 00:00:01 | | |
|* 2 | TABLE ACCESS BY LOCAL INDEX ROWID| ITS_CAR_PASS7 | 1 | 39 | 2 (0)| 00:00:01 | 1284 | 1284 |
| 3 | NESTED LOOPS | | 1 | 111 | 24 (0)| 00:00:01 | | |
|* 4 | TABLE ACCESS BY INDEX ROWID | ITS_BASE_DEVICE | 6 | 432 | 12 (0)| 00:00:01 | | |
|* 5 | INDEX RANGE SCAN | IDX_DEVICE_DEV_BAY_UNID | 7 | | 1 (0)| 00:00:01 | | |
| 6 | PARTITION RANGE SINGLE | | 44M| | 2 (0)| 00:00:01 | 1284 | 1284 |
|* 7 | INDEX RANGE SCAN | IDX_VT7_PASS_DEVICE_UNID | 44M| | 2 (0)| 00:00:01 | 1284 | 1284 |
------------------------------------------------------------------------------------------------------------------------------- Predicate Information (identified by operation id):
--------------------------------------------------- 2 - filter("PASS_DATETIME">=TO_DATE(' 2013-07-06 17:46:04', 'syyyy-mm-dd hh24:mi:ss') AND
"PASS_DATETIME"<=TO_DATE(' 2013-07-06 18:46:06', 'syyyy-mm-dd hh24:mi:ss'))
4 - filter("DEV_TYPE"='1' AND "DEV_DELETE_FLAG"='N' AND "DEV_CHK_FLAG"='1')
5 - access("DEV_BAY_UNID"='01685EFE4658C19D59C4DDAAEDD37393')
7 - access("V"."PASS_DEVICE_UNID"="UNID") Statistics
----------------------------------------------------------
1 recursive calls
0 db block gets
18645 consistent gets
130 physical reads
0 redo size
47861 bytes sent via SQL*Net to client
1657 bytes received via SQL*Net from client
105 SQL*Net roundtrips to/from client
1 sorts (memory)
0 sorts (disk)
1560 rows processed
驱动表改了之后，逻辑读从11W降低为1.8W
如何让in/exists 子查询(半连接)作为驱动表？的更多相关文章
- mysql---where子查询、form子查询、exists子查询
1.什么是子查询? 当一个查询是另一个查询的条件时,称之为子查询. 2.子查询有什么好处? 子查询可以使用几个简单命令构造功能强大的复合命令. 那么,现在让我们一起来学习子查询. 3.where型的子 ...
- 关于T-SQL中exists或者not exists子查询的“伪优化”的做法
问题起源 在使用t-sql中的exists(或者not exists)子查询的时候,不知道什么时候开始,发现一小部分人存在一种“伪优化”的一些做法,并且向不明真相的群众传递这一种写法“优越性”,实在看 ...
- 7-14 EXISTS子查询
EXISTS: 只注重于子查询是否有返回行,如果查有返回行返回结果为值,否则为假 并不使用子查询的结果,仅用于测试子查询是否有返回结果. 语法: IF EXISTS (子查询) BEGIN 语句块 E ...
- JOIN与EXISTS(子查询)的效率研究
使用MySQL提供的Sample数据库Sakila 现将profiling打开,用来一会查看sql执行时间 set profiling=1; exists 子查询与 join联接效率的对比,功能:查看 ...
- in型子查询陷阱,exists子查询
in 型子查询引出的陷阱 select goods_id from goods where cat_id in (1,2,3) 直接用id,不包含子查询,不会中陷阱 题: 在ecshop商城表中,查询 ...
- mysql优化---in型子查询，exists子查询，from 型子查询
in型子查询引出的陷阱:(扫更少的行,不要临时表,不要文件排序就快) 题: 在ecshop商城表中,查询6号栏目的商品, (注,6号是一个大栏目) 最直观的: mysql); 误区: 给我们的感觉是, ...
- Mysql常用sql语句（19）- in / exists 子查询
测试必备的Mysql常用sql语句系列 https://www.cnblogs.com/poloyy/category/1683347.html 前言 子查询在我们查询方法中是比较常用的,通过子查询可 ...
- [慢查优化]联表查询注意谁是驱动表 & 你搞不清楚谁join谁更好时请放手让mysql自行判定
写在前面的话: 不要求每个人一定理解 联表查询(join/left join/inner join等)时的mysql运算过程: 不要求每个人一定知道线上(现在或未来)哪张表数据量大,哪张表数据量小: ...
- 【转】[慢查优化]联表查询注意谁是驱动表 & 你搞不清楚谁join谁更好时请放手让mysql自行判定
转自:http://zhengyun-ustc.iteye.com/blog/1942797 写在前面的话: 不要求每个人一定理解 联表查询(join/left join/inner join等)时的 ...
随机推荐
- 关于 SSD 的接口和相关名词（2019-09-10）
关于 SSD 的接口和相关名词 了解了很多天的 SSD,太多的名词. 先记录一下. SATA MSATA M2 NVME NGFF U2 TODO: 后续收集相关信息.
- 连接池dbcp
连接池dbcp DBCP:apache组织 使用步骤: 1.导入jar包(commons-dbcp-1.4.jar和commons-pool-1.5.6.jar.commons-logging-1.2 ...
- python 代码中的类和对象
- 【JZOJ4832】【NOIP2016提高A组集训第3场10.31】高维宇宙
题目描述 数据范围 解法 由于大于4的素数只有可能由奇数和偶数的和得出. 所以根据数的奇偶性可以分出两类数奇数和偶数. 奇数之间不会相互匹配,偶数之间也不会相互匹配. 那么原问题转化为二分图最大匹配. ...
- IDEA入门(1)--lombok和Junit generator2插件的运用
前言 最近在慕课网看到了一些视频,准备从0开始做一个电商网站.视频中的大牛用的java的IDE都是IDEA,让我很纠结.从as到MyEclipse,好不容易稍微熟悉了一下MyEclipse的基本操作, ...
- MyEclipse2014，java文件无法编译，run as上是none applicable，不是文件本身的问题
1.配置一下JDK目录 2.window -> Preferences -> java -> Installed JREs -> Add
- 寒哥教你学iOS - 经验漫谈
http://www.jianshu.com/p/cb54054d3add 寒哥教你学iOS - 经验漫谈 字数2848 阅读1896 评论19 喜欢43 顺便来个广告 iOS开发者 群1734993 ...
- 在Swift中检查API的可用性
http://www.cocoachina.com/swift/20150901/13283.html 本文由CocoaChina译者ALEX吴浩文翻译自Use Your Loaf博客 原文:Chec ...
- 火狐自动填写表单autofill forms的用法和注意事项
1.安装后,打开要表单页面,右键
- jQuery show hide方法 二级菜单
<!DOCTYPE html> <html lang="en"> <head> <meta charset="UTF-8&quo ...