通过使用hint unnest调优sql语句(r4笔记第38天)
生产环境中有一条sql语句通过sql_monitor看到执行的时间实在是太惊人了,竟然达到了13个小时,而且还没有执行完。
Session |
APPC (20015:7013) |
---|---|
SQL ID |
74pzzzjddkyd4 |
SQL Execution ID |
16777242 |
Execution Started |
2/2/2015 10:52 |
First Refresh Time |
2/2/2015 10:52 |
Last Refresh Time |
2/3/2015 0:05 |
Duration |
47669s |
Module/Action |
bfi@ccbdbpr1 (TNS V1-V3)/- |
Service |
XXXXX |
Program |
bfi@xxx (TNS V1-V3) |
sql语句如下:
SELECT NVL(SUM(CUST.WEIGHT), 0) TOTAL_WEIGHT
FROM BL1_CUSTOMER CUST, BL1_CYCLE_CUSTOMERS CYC_CUST
WHERE CYC_CUST.PERIOD_KEY = :periodKey
AND CYC_CUST.CYCLE_SEQ_NO = :cycleSeqNo
AND CYC_CUST.CUSTOMER_NO = CUST.CUSTOMER_ID
AND (CYC_CUST.UNDO_REQ_TYPE = 'N' OR CYC_CUST.UNDO_REQ_TYPE IS NULL)
AND EXISTS
(SELECT 1
FROM BL1_CYC_PAYER_POP PAYER, BL1_DOCUMENT DOC
WHERE PAYER.PERIOD_KEY = :periodKey
AND PAYER.CYCLE_SEQ_NO = :cycleSeqNo
AND PAYER.CYCLE_SEQ_RUN = :cycleSeqRun
AND PAYER.CUSTOMER_NO = CYC_CUST.CUSTOMER_NO
AND PAYER.DB_STATUS = 'BL'
AND (PAYER.UNDO_REQ_TYPE = 'N' OR PAYER.UNDO_REQ_TYPE IS NULL)
AND PAYER.FORMAT_EXT_DATE IS NULL
AND DOC.PERIOD_KEY = :periodKey
AND DOC.CYCLE_SEQ_NO = :cycleSeqNo
AND DOC.CYCLE_SEQ_RUN = :cycleSeqRun
AND PAYER.BA_NO = DOC.BA_NO
AND doc.DOC_PRODUCE_IND IN ('Y', 'E'))
查看执行计划没有发现很严重的资源消耗。但是实际的执行情况怎么和执行计划相差甚远。预计8分钟,实际上十多个小时还没有执行完。
Plan hash value: 3506320481
--------------------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | Pstart| Pstop |
--------------------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 24 | 42048 (1)| 00:08:25 | | |
| 1 | SORT AGGREGATE | | 1 | 24 | | | | |
|* 2 | FILTER | | | | | | | |
| 3 | NESTED LOOPS | | | | | | | |
| 4 | NESTED LOOPS | | 1 | 24 | 42046 (1)| 00:08:25 | | |
| 5 | PARTITION RANGE SINGLE | | 1 | 15 | 42045 (1)| 00:08:25 | 171 | 171 |
|* 6 | TABLE ACCESS FULL | BL1_CYCLE_CUSTOMERS | 1 | 15 | 42045 (1)| 00:08:25 | 171 | 171 |
|* 7 | INDEX UNIQUE SCAN | BL1_CUSTOMER_PK | 1 | | 1 (0)| 00:00:01 | | |
| 8 | TABLE ACCESS BY INDEX ROWID | BL1_CUSTOMER | 1 | 9 | 1 (0)| 00:00:01 | | |
| 9 | NESTED LOOPS | | | | | | | |
| 10 | NESTED LOOPS | | 1 | 52 | 2 (0)| 00:00:01 | | |
| 11 | PARTITION RANGE SINGLE | | 1 | 34 | 1 (0)| 00:00:01 | 171 | 171 |
|* 12 | TABLE ACCESS BY LOCAL INDEX ROWID| BL1_CYC_PAYER_POP | 1 | 34 | 1 (0)| 00:00:01 | 171 | 171 |
|* 13 | INDEX RANGE SCAN | BL1_CYC_PAYER_POP_1IX | 3 | | 1 (0)| 00:00:01 | 171 | 171 |
| 14 | PARTITION RANGE SINGLE | | 7 | | 1 (0)| 00:00:01 | 171 | 171 |
|* 15 | INDEX RANGE SCAN | BL1_DOCUMENT_1IX | 7 | | 1 (0)| 00:00:01 | 171 | 171 |
|* 16 | TABLE ACCESS BY LOCAL INDEX ROWID | BL1_DOCUMENT | 1 | 18 | 1 (0)| 00:00:01 | 171 | 171 |
--------------------------------------------------------------------------------------------------------------------------------
这个时候可以通过sql monitor得到一个相对比较准确的资源使用情况。
Buffer Gets |
IO Requests |
Database Time |
Wait Activity |
||||
---|---|---|---|---|---|---|---|
. |
96M |
. |
21M |
. |
48518s |
. |
100% |
一看IO请求达21M次,约等于160.9G左右的数据量。 从sql语句的执行计划可以看出,语句可以分为两大部分,一部分是exist字句上面的部分,两个大表做了关联,得到了相关的customer_no然后在exists字句中继续关联。 大量的IO请求都消耗在BL1_CUSTOMER,其实这个表实际上数据量近千万,还没有80G多G,但是发送的IO请求累计的数据量却已经超过了80G,占到了整个IO请求数的一半以上。消耗的CPU资源也在73%以上
Id |
Operation |
Name |
Estimated |
Cost |
Execs |
Rows |
IO Requests |
CPU Activity |
|||
---|---|---|---|---|---|---|---|---|---|---|---|
Rows | |||||||||||
. |
0 |
SELECT STATEMENT |
. |
. |
. |
1 |
. |
. |
|||
-> |
1 |
. SORT AGGREGATE |
. |
1 |
. |
1 |
0 |
. |
|||
-> |
2 |
.. FILTER |
. |
. |
. |
1 |
403K |
. |
|||
-> |
3 |
... NESTED LOOPS |
. |
. |
. |
1 |
562K |
. |
|||
-> |
4 |
.... NESTED LOOPS |
. |
1682 |
45038 |
1 |
562K |
. |
|||
-> |
5 |
..... PARTITION RANGE ITERATOR |
. |
1682 |
44869 |
1 |
562K |
. |
|||
. |
6 |
...... TABLE ACCESS FULL |
BL1_CYCLE_CUSTOMERS |
1682 |
44869 |
1 |
562K |
7736 (<0.1%) |
|||
. |
7 |
..... INDEX UNIQUE SCAN |
BL1_CUSTOMER_PK |
1 |
1 |
949K |
562K |
. |
673K (3.3%) |
### |
|
-> |
8 |
.... TABLE ACCESS BY INDEX ROWID |
BL1_CUSTOMER |
1 |
1 |
990K |
562K |
. |
11M (52%) |
. |
73% |
-> |
9 |
... NESTED LOOPS |
. |
. |
. |
562K |
403K |
. |
|||
-> |
10 |
.... NESTED LOOPS |
. |
1 |
74 |
562K |
6M |
. |
|||
-> |
11 |
..... PARTITION RANGE ITERATOR |
. |
1 |
37 |
562K |
497K |
. |
|||
. |
12 |
...... TABLE ACCESS BY LOCAL INDEX ROWID |
BL1_CYC_PAYER_POP |
1 |
37 |
562K |
497K |
. |
774K (3.8%) |
. |
0.99% |
. |
13 |
....... INDEX RANGE SCAN |
BL1_CYC_PAYER_POP_1IX |
3 |
36 |
562K |
4M |
. |
864K (4.2%) |
. |
1.90% |
-> |
14 |
..... PARTITION RANGE ITERATOR |
. |
14 |
36 |
497K |
6M |
. |
|||
. |
15 |
...... INDEX RANGE SCAN |
BL1_DOCUMENT_1IX |
14 |
36 |
497K |
6M |
. |
4M (20%) |
. |
21% |
. |
16 |
.... TABLE ACCESS BY LOCAL INDEX ROWID |
BL1_DOCUMENT |
1 |
37 |
6M |
403K |
. |
3M (15%) |
. |
1.20% |
可以通过禁用子查询解嵌套来做为一种调优思路,优先从子查询中先输出数据来。 而BL1_CYCLE_PAYER_POP表作为一个重要的关联表。子查询中的条件AND PAYER.CUSTOMER_NO = CYC_CUST.CUSTOMER_NO和外部查询相关联。 可以优先查询这个表,考虑到执行的频率和性能,添加了并行hint。 这样sql语句就变为
SELECT NVL(SUM(CUST.WEIGHT), 0) TOTAL_WEIGHT
FROM BL1_CUSTOMER CUST, BL1_CYCLE_CUSTOMERS CYC_CUST
WHERE CYC_CUST.PERIOD_KEY = :periodKey
AND CYC_CUST.CYCLE_SEQ_NO = :cycleSeqNo
AND CYC_CUST.CUSTOMER_NO = CUST.CUSTOMER_ID
AND (CYC_CUST.UNDO_REQ_TYPE = 'N' OR CYC_CUST.UNDO_REQ_TYPE IS NULL)
AND EXISTS
(SELECT /*+unnest full(payer) parallel(payer 4)*/1
FROM BL1_CYC_PAYER_POP PAYER, BL1_DOCUMENT DOC
WHERE PAYER.PERIOD_KEY = :periodKey
AND PAYER.CYCLE_SEQ_NO = :cycleSeqNo
AND PAYER.CYCLE_SEQ_RUN = :cycleSeqRun
AND PAYER.CUSTOMER_NO = CYC_CUST.CUSTOMER_NO
AND PAYER.DB_STATUS = 'BL'
AND (PAYER.UNDO_REQ_TYPE = 'N' OR PAYER.UNDO_REQ_TYPE IS NULL)
AND PAYER.FORMAT_EXT_DATE IS NULL
AND DOC.PERIOD_KEY = :periodKey
AND DOC.CYCLE_SEQ_NO = :cycleSeqNo
AND DOC.CYCLE_SEQ_RUN = :cycleSeqRun
AND PAYER.BA_NO = DOC.BA_NO
AND doc.DOC_PRODUCE_IND IN ('Y', 'E'))
优化后的执行计划如下:
Plan hash value: 227985194
----------------------------------------------------------------------------------------------------------------------------------------------------------------------
| Id | Operation | Name | Rows | Bytes | Cost (%CPU)| Time | Pstart| Pstop | TQ |IN-OUT| PQ Distrib |
----------------------------------------------------------------------------------------------------------------------------------------------------------------------
| 0 | SELECT STATEMENT | | 1 | 37 | 13688 (1)| 00:02:45 | | | | | |
| 1 | SORT AGGREGATE | | 1 | 37 | | | | | | | |
| 2 | PX COORDINATOR | | | | | | | | | | |
| 3 | PX SEND QC (RANDOM) | :TQ10001 | 1 | 37 | | | | | Q1,01 | P->S | QC (RAND) |
| 4 | SORT AGGREGATE | | 1 | 37 | | | | | Q1,01 | PCWP | |
| 5 | NESTED LOOPS | | | | | | | | Q1,01 | PCWP | |
| 6 | NESTED LOOPS | | 1 | 37 | 13688 (1)| 00:02:45 | | | Q1,01 | PCWP | |
| 7 | NESTED LOOPS | | 1 | 28 | 13688 (1)| 00:02:45 | | | Q1,01 | PCWP | |
| 8 | VIEW | VW_SQ_1 | 1 | 13 | 13686 (1)| 00:02:45 | | | Q1,01 | PCWP | |
| 9 | HASH UNIQUE | | 1 | 52 | | | | | Q1,01 | PCWP | |
| 10 | PX RECEIVE | | 1 | 52 | | | | | Q1,01 | PCWP | |
| 11 | PX SEND HASH | :TQ10000 | 1 | 52 | | | | | Q1,00 | P->P | HASH |
| 12 | HASH UNIQUE | | 1 | 52 | | | | | Q1,00 | PCWP | |
| 13 | NESTED LOOPS | | | | | | | | Q1,00 | PCWP | |
| 14 | NESTED LOOPS | | 1 | 52 | 13686 (1)| 00:02:45 | | | Q1,00 | PCWP | |
| 15 | PX BLOCK ITERATOR | | 1 | 34 | 13686 (1)| 00:02:45 | 171 | 171 | Q1,00 | PCWC | |
|* 16 | TABLE ACCESS FULL | BL1_CYC_PAYER_POP | 1 | 34 | 13686 (1)| 00:02:45 | 171 | 171 | Q1,00 | PCWP | |
| 17 | PARTITION RANGE SINGLE | | 7 | | 1 (0)| 00:00:01 | 171 | 171 | Q1,00 | PCWP | |
|* 18 | INDEX RANGE SCAN | BL1_DOCUMENT_1IX | 7 | | 1 (0)| 00:00:01 | 171 | 171 | Q1,00 | PCWP | |
|* 19 | TABLE ACCESS BY LOCAL INDEX ROWID| BL1_DOCUMENT | 1 | 18 | 1 (0)| 00:00:01 | 171 | 171 | Q1,00 | PCWP | |
| 20 | PARTITION RANGE SINGLE | | 1 | 15 | 1 (0)| 00:00:01 | 171 | 171 | Q1,01 | PCWP | |
|* 21 | TABLE ACCESS BY LOCAL INDEX ROWID | BL1_CYCLE_CUSTOMERS | 1 | 15 | 1 (0)| 00:00:01 | 171 | 171 | Q1,01 | PCWP | |
|* 22 | INDEX RANGE SCAN | BL1_CYCLE_CUSTOMERS_PK | 1 | | 1 (0)| 00:00:01 | 171 | 171 | Q1,01 | PCWP | |
|* 23 | INDEX UNIQUE SCAN | BL1_CUSTOMER_PK | 1 | | 1 (0)| 00:00:01 | | | Q1,01 | PCWP | |
| 24 | TABLE ACCESS BY INDEX ROWID | BL1_CUSTOMER | 1 | 9 | 1 (0)| 00:00:01 | | | Q1,01 | PCWP | |
----------------------------------------------------------------------------------------------------------------------------------------------------------------------
Predicate Information (identified by operation id):
---------------------------------------------------
16 - filter("PAYER"."FORMAT_EXT_DATE" IS NULL AND "PAYER"."CYCLE_SEQ_NO"=4105 AND "PAYER"."PERIOD_KEY"=61 AND ("PAYER"."UNDO_REQ_TYPE"='N' OR
"PAYER"."UNDO_REQ_TYPE" IS NULL) AND "PAYER"."DB_STATUS"='BL' AND "PAYER"."CYCLE_SEQ_RUN"=0)
18 - access("PAYER"."BA_NO"="DOC"."BA_NO")
19 - filter("DOC"."CYCLE_SEQ_NO"=4105 AND "DOC"."PERIOD_KEY"=61 AND "DOC"."CYCLE_SEQ_RUN"=0 AND ("DOC"."DOC_PRODUCE_IND"='E' OR
"DOC"."DOC_PRODUCE_IND"='Y'))
21 - filter("CYC_CUST"."UNDO_REQ_TYPE"='N' OR "CYC_CUST"."UNDO_REQ_TYPE" IS NULL)
22 - access("ITEM_0"="CYC_CUST"."CUSTOMER_NO" AND "CYC_CUST"."CYCLE_SEQ_NO"=4105 AND "CYC_CUST"."PERIOD_KEY"=61)
23 - access("CYC_CUST"."CUSTOMER_NO"="CUST"."CUSTOMER_ID")
最后得到的反馈是,原本执行近20个小时的查询,在添加这个Hint之后,执行时间缩短到了1个小时以内。性能的提升还是相当的可观的。
- google protobuf学习笔记:编译安装、序列化、反序列化
- 自己在Qt上做的辣鸡计算器
- 【Qt】]Qt5中文乱码
- 防止连接Mysql超时,JDBC探活配置
- 剑指offer——面试题10输入一个十进制整数,统计其中二进制1的个数
- 剑指offer——面试题9计算斐波纳切第n个数
- 剑指 offer——面试题8求旋转数组的最小值
- MYSQL INNODB表压缩
- 剑指offer——年龄排序问题
- Mysql Group Replication介绍
- 剑指offer——快速排序
- 架构高性能网站秘笈(四)——反向代理缓存
- 架构高性能网站秘笈(一)——了解衡量网站性能的指标
- MYSQL5.6&5.7编译安装
- JavaScript 教程
- JavaScript 编辑工具
- JavaScript 与HTML
- JavaScript 与Java
- JavaScript 数据结构
- JavaScript 基本数据类型
- JavaScript 特殊数据类型
- JavaScript 运算符
- JavaScript typeof 运算符
- JavaScript 表达式
- JavaScript 类型转换
- JavaScript 基本语法
- JavaScript 注释
- Javascript 基本处理流程
- Javascript 选择结构
- Javascript if 语句
- Javascript if 语句的嵌套
- Javascript switch 语句
- Javascript 循环结构
- Javascript 循环结构实例
- Javascript 跳转语句
- Javascript 控制语句总结
- Javascript 函数介绍
- Javascript 函数的定义
- Javascript 函数调用
- Javascript 几种特殊的函数
- JavaScript 内置函数简介
- Javascript eval() 函数
- Javascript isFinite() 函数
- Javascript isNaN() 函数
- parseInt() 与 parseFloat()
- escape() 与 unescape()
- Javascript 字符串介绍
- Javascript length属性
- javascript 字符串函数
- Javascript 日期对象简介
- Javascript 日期对象用途
- Date 对象属性和方法
- Javascript 数组是什么
- Javascript 创建数组
- Javascript 数组赋值与取值
- Javascript 数组属性和方法
- Redis过期策略和数据淘汰机制
- 图片由彩色渐变到黑白动画
- Kafka如何保证数据可靠性
- ubuntu中snap包的安装、更新删除与简单使用
- 学习在kernel态下使用NEON对算法进行加速的方法
- Centos7上Mesos和Marathon的安装和配置
- Redis高级数据类型-Bitmap和HyperLogLog
- Redis持久化策略
- 在linux下开启FTP服务方法介绍
- Linux中gpio接口的使用方法示例
- Three.js教程(1):初识three.js
- 解决Electron安装报错问题
- Linux端口映射转发的方法
- Centos7如何备份和还原Redis数据的方法
- linux contos6.8下部署kafka集群的方法