树形查询SQL优化一例
上周五一哥们发了条SQL，让我看看，代码如下：
SELECT COUNT(1)
FROM (select m.sheet_id
from cpm_main_sheet_history m, cpm_service_warn_config s
where m.sheet_type_id in
(select t.row_id
from tbl_class_trees t
start with t.row_id = s.sheet_type_id
connect by t.parent_row_id = prior t.row_id)
and m.service_type in
(select tt.row_id
from tbl_class_trees tt
start with tt.row_id = s.business_type_id
connect by tt.parent_row_id = prior tt.row_id)
and m.accept_time >=
TO_CHAR(SYSDATE - time_interval / 24, 'yyyy-mm-dd hh24:mi:ss')
and s.row_id = 'AS170904165251'
and m.no_area in ('0000')) c__
--执行计划
PLAN_TABLE_OUTPUT
Plan hash value: 2710926849
----------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
----------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 143 | 96246 (1)| 00:19:15 |
|* 1 | FILTER | | | | | |
| 2 | NESTED LOOPS | | 1681 | 234K| 746 (1)| 00:00:09 |
| 3 | TABLE ACCESS BY INDEX ROWID | CPM_SERVICE_WARN_CONFIG | 1 | 56 | 1 (0)| 00:00:01 |
|* 4 | INDEX UNIQUE SCAN | PK_CONFIG_ROW_ID | 1 | | 0 (0)| 00:00:01 |
| 5 | TABLE ACCESS BY INDEX ROWID | CPM_MAIN_SHEET_HISTORY | 1681 | 142K| 745 (1)| 00:00:09 |
|* 6 | INDEX RANGE SCAN | IDX_CPM_NO_AREA_TIME1 | 449 | | 62 (0)| 00:00:01 |
|* 7 | FILTER | | | | | |
|* 8 | CONNECT BY NO FILTERING WITH SW (UNIQUE)| | | | | |
| 9 | TABLE ACCESS FULL | TBL_CLASS_TREES | 9527 | 493K| 113 (0)| 00:00:02 |
|* 10 | FILTER | | | | | |
|* 11 | CONNECT BY NO FILTERING WITH SW (UNIQUE)| | | | | |
| 12 | TABLE ACCESS FULL | TBL_CLASS_TREES | 9527 | 493K| 113 (0)| 00:00:02 |
----------------------------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter( EXISTS (SELECT 0 FROM "TBL_CLASS_TREES" "T" WHERE "T"."ROW_ID"=:B1 START WITH "T"."ROW_ID"=:B2
CONNECT BY "T"."PARENT_ROW_ID"=PRIOR "T"."ROW_ID") AND EXISTS (SELECT 0 FROM "TBL_CLASS_TREES" "TT" WHERE
"TT"."ROW_ID"=:B3 START WITH "TT"."ROW_ID"=:B4 CONNECT BY "TT"."PARENT_ROW_ID"=PRIOR "TT"."ROW_ID"))
4 - access("S"."ROW_ID"='AS170904165251')
6 - access("M"."ACCEPT_TIME">=TO_CHAR(SYSDATE@!-"TIME_INTERVAL"/24,'yyyy-mm-dd hh24:mi:ss') AND
"M"."NO_AREA"='0000' AND "M"."ACCEPT_TIME" IS NOT NULL)
filter("M"."NO_AREA"='0000')
7 - filter("T"."ROW_ID"=:B1)
8 - access("T"."PARENT_ROW_ID"=PRIOR "T"."ROW_ID")
filter("T"."ROW_ID"=:B1)
10 - filter("TT"."ROW_ID"=:B1)
11 - access("TT"."PARENT_ROW_ID"=PRIOR "TT"."ROW_ID")
filter("TT"."ROW_ID"=:B1) SQL优化前：
耗时：20s
count(1)返回: 147条数据
分析执行计划，执行计划中有filter关键字且有3个子级，这种sql是最容易引起性能问题的，所以第一时间是反应是sql有没有走索引，能不能改写。
尝试1：
建索引优化：
在TBL_CLASS_TREES表(row_id,parent_row_id)上建索引
执行计划：
PLAN_TABLE_OUTPUT
Plan hash value: 135779572
----------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
----------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 155 | 34147 (1)| 00:06:50 |
|* 1 | FILTER | | | | | |
| 2 | NESTED LOOPS | | 3021 | 457K| 816 (1)| 00:00:10 |
| 3 | TABLE ACCESS BY INDEX ROWID | CPM_SERVICE_WARN_CONFIG | 1 | 62 | 1 (0)| 00:00:01 |
|* 4 | INDEX UNIQUE SCAN | PK_CONFIG_ROW_ID | 1 | | 0 (0)| 00:00:01 |
| 5 | TABLE ACCESS BY INDEX ROWID | CPM_MAIN_SHEET_HISTORY | 3021 | 274K| 815 (1)| 00:00:10 |
|* 6 | INDEX RANGE SCAN | IDX_CPM_NO_AREA_TIME1 | 563 | | 136 (0)| 00:00:02 |
|* 7 | FILTER | | | | | |
|* 8 | CONNECT BY NO FILTERING WITH SW (UNIQUE)| | | | | |
| 9 | INDEX FAST FULL SCAN | IDX_ROW_ID | 9527 | 493K| 20 (0)| 00:00:01 |
|* 10 | FILTER | | | | | |
|* 11 | CONNECT BY NO FILTERING WITH SW (UNIQUE)| | | | | |
| 12 | INDEX FAST FULL SCAN | IDX_ROW_ID | 9527 | 493K| 20 (0)| 00:00:01 |
----------------------------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter( EXISTS (SELECT 0 FROM "TBL_CLASS_TREES" "T" WHERE "T"."ROW_ID"=:B1 START WITH "T"."ROW_ID"=:B2
CONNECT BY "T"."PARENT_ROW_ID"=PRIOR "T"."ROW_ID") AND EXISTS (SELECT 0 FROM "TBL_CLASS_TREES" "TT" WHERE
"TT"."ROW_ID"=:B3 START WITH "TT"."ROW_ID"=:B4 CONNECT BY "TT"."PARENT_ROW_ID"=PRIOR "TT"."ROW_ID"))
4 - access("S"."ROW_ID"='AS170904165251')
6 - access("M"."ACCEPT_TIME">=TO_CHAR(SYSDATE@!-"TIME_INTERVAL"/24,'yyyy-mm-dd hh24:mi:ss') AND
"M"."NO_AREA"='0000' AND "M"."ACCEPT_TIME" IS NOT NULL)
filter("M"."NO_AREA"='0000')
7 - filter("T"."ROW_ID"=:B1)
8 - access("T"."PARENT_ROW_ID"=PRIOR "T"."ROW_ID")
filter("T"."ROW_ID"=:B1)
10 - filter("TT"."ROW_ID"=:B1)
11 - access("TT"."PARENT_ROW_ID"=PRIOR "TT"."ROW_ID")
filter("TT"."ROW_ID"=:B1)
--效果还是一样慢，此优化失败。尝试2：
利用with改写sql优化
with t as (select /*+ materialize */ row_id,parent_row_id from tbl_class_trees)
SELECT COUNT(1)
FROM (select m.sheet_id
from cpm_main_sheet_history m,cpm_service_warn_config s
where m.sheet_type_id in
(select t.row_id
from t
start with t.row_id = s.sheet_type_id
connect by t.parent_row_id = prior t.row_id)
and m.service_type in
(select t.row_id
from t
start with t.row_id = s.business_type_id
connect by t.parent_row_id = prior t.row_id)
and m.accept_time >=
TO_CHAR(SYSDATE - time_interval / 24, 'yyyy-mm-dd hh24:mi:ss')
and s.row_id = 'AS170904165251'
and m.no_area in ('0000')) c__
--效果还是一样慢，此优化失败。 再次分析原SQL执行计划：
id=8,id=11的执行计划关键词是：CONNECT BY NO FILTERING WITH SW (UNIQUE)。
这个为树形查询在11g中的新特性，尝试让sql不使用这个新特性。
于是使用以下hint：/*+ connect_by_filtering */ 进行优化：
SELECT COUNT(1)
FROM (select m.sheet_id
from cpm_main_sheet_history m, cpm_service_warn_config s
where m.sheet_type_id in
(select /*+ connect_by_filtering */ t.row_id
from tbl_class_trees t
start with t.row_id = s.sheet_type_id
connect by t.parent_row_id = prior t.row_id)
and m.service_type in
(select /*+ connect_by_filtering */ tt.row_id
from tbl_class_trees tt
start with tt.row_id = s.business_type_id
connect by tt.parent_row_id = prior tt.row_id)
and m.accept_time >=
TO_CHAR(SYSDATE - time_interval / 24, 'yyyy-mm-dd hh24:mi:ss')
and s.row_id = 'AS170904165251'
and m.no_area in ('0000')) c__
PLAN_TABLE_OUTPUT
Plan hash value: 2824841339
----------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time |
----------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 246 | 188K (1)| 00:37:47 |
|* 1 | FILTER | | | | | |
| 2 | NESTED LOOPS | | 3021 | 725K| 816 (1)| 00:00:10 |
| 3 | TABLE ACCESS BY INDEX ROWID | CPM_SERVICE_WARN_CONFIG | 1 | 106 | 1 (0)| 00:00:01 |
|* 4 | INDEX UNIQUE SCAN | PK_CONFIG_ROW_ID | 1 | | 0 (0)| 00:00:01 |
| 5 | TABLE ACCESS BY INDEX ROWID | CPM_MAIN_SHEET_HISTORY | 3021 | 413K| 815 (1)| 00:00:10 |
|* 6 | INDEX RANGE SCAN | IDX_CPM_NO_AREA_TIME1 | 563 | | 136 (0)| 00:00:02 |
|* 7 | FILTER | | | | | |
|* 8 | CONNECT BY WITH FILTERING | | | | | |
| 9 | TABLE ACCESS BY INDEX ROWID| TBL_CLASS_TREES | 1 | 107 | 2 (0)| 00:00:01 |
|* 10 | INDEX UNIQUE SCAN | PK_TBL_CLASS_TREESS | 1 | | 1 (0)| 00:00:01 |
|* 11 | HASH JOIN | | | | | |
| 12 | CONNECT BY PUMP | | | | | |
| 13 | TABLE ACCESS FULL | TBL_CLASS_TREES | 6 | 318 | 113 (0)| 00:00:02 |
|* 14 | FILTER | | | | | |
|* 15 | CONNECT BY WITH FILTERING | | | | | |
| 16 | TABLE ACCESS BY INDEX ROWID| TBL_CLASS_TREES | 1 | 107 | 2 (0)| 00:00:01 |
|* 17 | INDEX UNIQUE SCAN | PK_TBL_CLASS_TREESS | 1 | | 1 (0)| 00:00:01 |
|* 18 | HASH JOIN | | | | | |
| 19 | CONNECT BY PUMP | | | | | |
| 20 | TABLE ACCESS FULL | TBL_CLASS_TREES | 6 | 318 | 113 (0)| 00:00:02 |
----------------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
1 - filter( EXISTS (SELECT /*+ CONNECT_BY_FILTERING */ 0 FROM "TBL_CLASS_TREES" "T" WHERE
"T"."ROW_ID"=:B1 START WITH "T"."ROW_ID"=:B2) AND EXISTS (SELECT /*+ CONNECT_BY_FILTERING */ 0
FROM "TBL_CLASS_TREES" "TT" WHERE "TT"."ROW_ID"=:B3 START WITH "TT"."ROW_ID"=:B4))
4 - access("S"."ROW_ID"='AS170904165251')
6 - access("M"."ACCEPT_TIME">=TO_CHAR(SYSDATE@!-"TIME_INTERVAL"/24,'yyyy-mm-dd hh24:mi:ss')
AND "M"."NO_AREA"='0000' AND "M"."ACCEPT_TIME" IS NOT NULL)
filter("M"."NO_AREA"='0000')
7 - filter("T"."ROW_ID"=:B1)
8 - access("T"."PARENT_ROW_ID"=PRIOR "T"."ROW_ID")
10 - access("T"."ROW_ID"=:B1)
11 - access("T"."PARENT_ROW_ID"=PRIOR "T"."ROW_ID")
14 - filter("TT"."ROW_ID"=:B1)
15 - access("TT"."PARENT_ROW_ID"=PRIOR "TT"."ROW_ID")
17 - access("TT"."ROW_ID"=:B1)
18 - access("TT"."PARENT_ROW_ID"=PRIOR "TT"."ROW_ID")
--优化后，SQL能在5s返回结果树形查询SQL优化一例的更多相关文章
- 跨服务器查询sql语句样例
若2个数据库在同一台机器上:insert into DataBase_A..Table1(col1,col2,col3----)select col11,col22,col33-- from Data ...
- 跨服务器查询sql语句样例（转）
若2个数据库在同一台机器上: insert into DataBase_A..Table1(col1,col2,col3----) select col11,col22,col33-- from Da ...
- oracle 11g亿级复杂SQL优化一例（数量级性能提升）
自从16年之后,因为工作原因,项目中就没有再使用oracle了,最近最近支持一个项目,又要开始负责这块事情了.最近在跑性能测试,配置全部调好之后,不少sql还存在性能低下的问题,主要涉及执行计划的不合 ...
- 查询SQL优化
SQL优化的一般步骤 通过show status命令了解各种SQL的执行频率定位执行效率较低的SQL语句,重点select通过explain分析低效率的SQL确定问题并采取相应的优化措施 优化措施 s ...
- 反连接NOT EXISTS子查询中有or 谓词连接条件SQL优化一例
背景 今天在日常数据库检查中,发现一SQL运行时间特别长,于是抓取出来,进行优化. 优化前: 耗时:503s 返回:0 SQL代码 SELECT * FROM MM_PAYABLEMONEY_TD P ...
- 1 min 数据查询 SQL 优化
问题 前几天线上数据库 IOPS 飙升,一直居高不下,最近并没有升级.遂查看数据库正在执行的 SQL 语句,发现有个查询离线设备的语句极其缓慢. 探寻原因 SELECT o.* FROM ( SELE ...
- mysql联合查询sql优化
我们在使用mysql数据库时,经常会使用到mysql的联合查询,联合查询分为内连接和外连接,内连接查询结果是联合的表都存在匹配才会有结果,外连接则根据驱动表是否存在匹配来生成结果集. 这里使用mysq ...
- oracle查询SQL优化相当重要
如果表中的时间字段是索引,那么时间字段不要使用函数,函数会使索引失效. 例如: select * from mytable where trunc(createtime)=trunc(sysdate) ...
- Mysql 分页查询sql优化
先查下数据表的总条数: SELECT COUNT(id) FROM ts_translation_send_address 执行分页界SQL 查看使用时间2.210s SELECT * FROM ts ...
随机推荐
- 【黑金教程笔记之003】【建模篇】akuei2的Verilog hdl心路
Verilog hdl不是“编程”是“建模” Verilog hdl语言是一种富有“形状”的语言. 如果着手以“建模”去理解Verilog hdl语言,以“形状”去完成Verilog hdl语言的设计 ...
- bzoj 1597: [Usaco2008 Mar]土地购买【斜率优化】
按xy降序排序,把能被完全包含的去掉 然后就得到了x升序y降序的一个数组 然后方程就显然了:f[i]=min(f[j]+y[j+1]x[i]) 斜率优化转移 说起来我还不会斜率优化呢是不是该学一下了 ...
- A+B Problem——经典中的经典
A+B Problem,这道题,吸收了天地的精华,是当之无愧的经典中的经典中的经典.自古以来OIer都会经过它的历练(这不是白说吗?),下面就有我herobrine来讲讲这道题的各种做法. 好吧,同志 ...
- 【TIDB】2、TIDB进阶
0.TIDB优势 1.和MySql相比,具备OLAP能力.省去了很多数据仓库搭建成本和学习成本.这在业务层是非常受欢迎的.可以在其他分库分表业务中,通过 syncer 同步,进行合并,然后进行统计分析 ...
- VS 2017 产品密钥
Visual Studio 2017(VS2017) 企业版 Enterprise 注册码:NJVYC-BMHX2-G77MM-4XJMR-6Q8QFVisual Studio 2017(VS2017 ...
- BP神经网络算法改进
周志华机器学习BP改进 试设计一个算法,能通过动态调整学习率显著提升收敛速度,编程实现该算法,并选择两个UCI数据集与标准的BP算法进行实验比较. 1.方法设计 传统的BP算法改进主要有两类: - 启 ...
- 转】MongoDB 自动分片 auto sharding
原博文出自于: http://blog.fens.me/category/%E6%95%B0%E6%8D%AE%E5%BA%93/page/4/ 感谢! MongoDB 自动分片 auto shard ...
- 学JAVA第二十三天，List类型和Set类型
数组,是我们最常用的,但是有时候,我们要用数组,但是又不知道数组的类的长度的时候, 我们java就有一个很好用的工具Collection,这都是java的爸爸的用心良苦,Collection中包含Li ...
- git push代码时的'git did not exit cleanly (exit code 1)'问题解决
在利用git管理本地发布的galleryLeftOrRight插件项目时,按照git的使用方法:先commit→master,再 push,发现提示错误git did not exit cleanly ...
- Selenium--Python环境部署
本文引读:一二为python环境安装:三为selenium安装同时介绍了pip:四为PyCharm安装:五为验证SE可以正常使用 一.下载python安装包 我这里安装的是python3.6.5,官网 ...