转载：postgresql分区与优化

--对于分区表constraint_exclusion 这个参数需要配置为partition或on

postgres=# show constraint_exclusion ;

 constraint_exclusion

----------------------

 partition

 --创建父子表, 用于存储分区数据

create table t(id int primary key);

create table t1(like t including all) inherits(t);

create table t2(like t including all) inherits(t);

create table t3(like t including all) inherits(t);

create table t4(like t including all) inherits(t);

--PostgreSQL的子表和子表之间的约束是没有任何关系的, 所以也可以有重叠, 即非全局约束.

 alter table t1 add constraint ck_t1_1 check(id<0);

 alter table t2 add constraint ck_t2_1 check(id>=0 and id<100);

 alter table t3 add constraint ck_t3_1 check(id>=100 and id<200);

 alter table t4 add constraint ck_t4_1 check(id>=200);

 --分区字段传入常量, 执行时扫描的是父表和约束对应的子表 :

postgres=#  explain select * from t where id=10;

                                 QUERY PLAN

-----------------------------------------------------------------------------

 Append  (cost=0.00..8.17 rows=2 width=4)

   ->  Seq Scan on t  (cost=0.00..0.00 rows=1 width=4)

         Filter: (id = 10)

   ->  Index Only Scan using t2_pkey on t2  (cost=0.15..8.17 rows=1 width=4)

         Index Cond: (id = 10)

(5 rows)

--分区字段传入常量, 执行时扫描的是父表和约束对应的子表;

postgres=#  prepare p_test as select * from t where id=$1;

PREPARE

postgres=# explain execute p_test(1);

                                 QUERY PLAN

-----------------------------------------------------------------------------

 Append  (cost=0.00..8.17 rows=2 width=4)

   ->  Seq Scan on t  (cost=0.00..0.00 rows=1 width=4)

         Filter: (id = 1)

   ->  Index Only Scan using t2_pkey on t2  (cost=0.15..8.17 rows=1 width=4)

         Index Cond: (id = 1)

(5 rows)

--子句查询, 执行时扫描的是父表和所有子表, 注意这里使用的子查询是子表的查询, 理论上应该是扫描父表和该子表

postgres=#  explain select * from t where id=(select id from t1 limit 1);

                                 QUERY PLAN

-----------------------------------------------------------------------------

 Append  (cost=0.01..32.70 rows=5 width=4)

   InitPlan 1 (returns $0)

     ->  Limit  (cost=0.00..0.01 rows=1 width=4)

           ->  Seq Scan on t1 t1_1  (cost=0.00..34.00 rows=2400 width=4)

   ->  Seq Scan on t  (cost=0.00..0.00 rows=1 width=4)

         Filter: (id = $0)

   ->  Index Only Scan using t1_pkey on t1  (cost=0.15..8.17 rows=1 width=4)

         Index Cond: (id = $0)

   ->  Index Only Scan using t2_pkey on t2  (cost=0.15..8.17 rows=1 width=4)

         Index Cond: (id = $0)

   ->  Index Only Scan using t3_pkey on t3  (cost=0.15..8.17 rows=1 width=4)

         Index Cond: (id = $0)

   ->  Index Only Scan using t4_pkey on t4  (cost=0.15..8.17 rows=1 width=4)

         Index Cond: (id = $0)

(14 rows)

--综上可知在对分区表进行查询时最好使用字面常量，而不要使用子查询之类复杂的sql

--如果子表上约束删除，则pg不得不把删除约束的子表也加入到查询中(即使子表可以忽略)

alter table t4 drop constraint ck_t4_1;

postgres=#  explain select * from t where id=10;

                                 QUERY PLAN

-----------------------------------------------------------------------------

 Append  (cost=0.00..16.34 rows=3 width=4)

   ->  Seq Scan on t  (cost=0.00..0.00 rows=1 width=4)

         Filter: (id = 10)

   ->  Index Only Scan using t2_pkey on t2  (cost=0.15..8.17 rows=1 width=4)

         Index Cond: (id = 10)

   ->  Index Only Scan using t4_pkey on t4  (cost=0.15..8.17 rows=1 width=4)

         Index Cond: (id = 10)

(7 rows)

--如果constraint_exclusion设置为off，pg不得不进行全表扫描

postgres=# set constraint_exclusion=off;

SET

postgres=#  explain select * from t where id=10;

                                 QUERY PLAN

-----------------------------------------------------------------------------

 Append  (cost=0.00..32.69 rows=5 width=4)

   ->  Seq Scan on t  (cost=0.00..0.00 rows=1 width=4)

         Filter: (id = 10)

   ->  Index Only Scan using t1_pkey on t1  (cost=0.15..8.17 rows=1 width=4)

         Index Cond: (id = 10)

   ->  Index Only Scan using t2_pkey on t2  (cost=0.15..8.17 rows=1 width=4)

         Index Cond: (id = 10)

   ->  Index Only Scan using t3_pkey on t3  (cost=0.15..8.17 rows=1 width=4)

         Index Cond: (id = 10)

   ->  Index Only Scan using t4_pkey on t4  (cost=0.15..8.17 rows=1 width=4)

         Index Cond: (id = 10)

(11 rows)

--分区表上一般针对分区建立相对应的分区索引

--建在父表的索引为全局索引，但如果你表没有数据要查询子表时，则分区表要进行全表扫描

--父表建立的全局索引

postgres=# \d+ p

                                       Table "public.p"

  Column   |              Type              | Modifiers | Storage | Stats target | Description

-----------+--------------------------------+-----------+---------+--------------+-------------

 city_id   | integer                        | not null  | plain   |              |

 logtime   | timestamp(0) without time zone | not null  | plain   |              |

 peaktemp  | integer                        |           | plain   |              |

 unitsales | integer                        |           | plain   |              |

Indexes:

    "idx_city_id" btree (city_id)

    "idx_p_logtime" btree (logtime)

Triggers:

    delete_p_trigger BEFORE DELETE ON p FOR EACH ROW EXECUTE PROCEDURE p_delete_trigger()

    insert_p_trigger BEFORE INSERT ON p FOR EACH ROW EXECUTE PROCEDURE p_insert_trigger()

Child tables: p_201201,

              p_201202,

              p_201203,

              p_201204,

              p_201205,

              p_201206,

              p_201207,

              p_201208,

              p_201209,

              p_201210,

              p_201211,

              p_201212,

              p_default

Has OIDs: no

--分区没有索引，不能使用父表索引

postgres=# explain select * from p_201202 where city_id=2 and logtime=timestamp '2012-02-02 12:59:59';

                                          QUERY PLAN

----------------------------------------------------------------------------------------------

 Seq Scan on p_201202  (cost=0.00..214.01 rows=2 width=20)

   Filter: ((city_id = 2) AND (logtime = '2012-02-02 12:59:59'::timestamp without time zone))

(2 rows)

--建立分区索引，可以使用分区索引

postgres=# CREATE INDEX idx_p_201202_city_id ON p_201202 (city_id);

CREATE INDEX

postgres=# explain select * from p_201202 where city_id=2 and logtime=timestamp '2012-02-02 12:59:59';

                                      QUERY PLAN

--------------------------------------------------------------------------------------

 Index Scan using idx_p_201202_city_id on p_201202  (cost=0.29..8.33 rows=2 width=20)

   Index Cond: (city_id = 2)

   Filter: (logtime = '2012-02-02 12:59:59'::timestamp without time zone)

--也可以指定只查询父表的数据

postgres=# select * from only p;

 city_id | logtime | peaktemp | unitsales

---------+---------+----------+-----------

(0 rows)

--如果一个分区表，父子表之间不再有继承关系，则查询父表时不再过滤到子表

postgres=# alter table t3 no inherit t;

ALTER TABLE

postgres=# explain select count(*) from t;

                            QUERY PLAN

------------------------------------------------------------------

 Aggregate  (cost=73.50..73.51 rows=1 width=0)

   ->  Append  (cost=0.00..62.80 rows=4281 width=0)

         ->  Seq Scan on t  (cost=0.00..0.00 rows=1 width=0)

         ->  Seq Scan on t1  (cost=0.00..31.40 rows=2140 width=0)

         ->  Seq Scan on t2  (cost=0.00..31.40 rows=2140 width=0)

(5 rows)

--再次添加继承，查询父表可以过滤到子表

postgres=# alter table t3 inherit t;

ALTER TABLE

postgres=# explain select count(*) from t;

                            QUERY PLAN

------------------------------------------------------------------

 Aggregate  (cost=110.25..110.26 rows=1 width=0)

   ->  Append  (cost=0.00..94.20 rows=6421 width=0)

         ->  Seq Scan on t  (cost=0.00..0.00 rows=1 width=0)

         ->  Seq Scan on t1  (cost=0.00..31.40 rows=2140 width=0)

         ->  Seq Scan on t2  (cost=0.00..31.40 rows=2140 width=0)

         ->  Seq Scan on t3  (cost=0.00..31.40 rows=2140 width=0)

(6 rows)

--以下为p表测试数据代码

CREATE TABLE p (

    city_id         int not null,

    logtime         timestamp(0) not null,

    peaktemp        int,

    unitsales       int

);

CREATE INDEX idx_p_logtime ON p (logtime);

CREATE TABLE p_201201 (LIKE p INCLUDING all) INHERITS (p);

CREATE TABLE p_201202 (LIKE p INCLUDING all) INHERITS (p);

CREATE TABLE p_201203 (LIKE p INCLUDING all) INHERITS (p);

CREATE TABLE p_201204 (LIKE p INCLUDING all) INHERITS (p);

CREATE TABLE p_201205 (LIKE p INCLUDING all) INHERITS (p);

CREATE TABLE p_201206 (LIKE p INCLUDING all) INHERITS (p);

CREATE TABLE p_201207 (LIKE p INCLUDING all) INHERITS (p);

CREATE TABLE p_201208 (LIKE p INCLUDING all) INHERITS (p);

CREATE TABLE p_201209 (LIKE p INCLUDING all) INHERITS (p);

CREATE TABLE p_201210 (LIKE p INCLUDING all) INHERITS (p);

CREATE TABLE p_201211 (LIKE p INCLUDING all) INHERITS (p);

CREATE TABLE p_201212 (LIKE p INCLUDING all) INHERITS (p);

CREATE TABLE p_default (LIKE p INCLUDING all) INHERITS (p);

CREATE OR REPLACE FUNCTION p_insert_trigger()

RETURNS TRIGGER AS $$

BEGIN

    IF    ( NEW.logtime >= DATE '2012-01-01' AND NEW.logtime < DATE '2012-02-01' ) THEN

        INSERT INTO p_201201 VALUES (NEW.*);

    ELSIF ( NEW.logtime >= DATE '2012-02-01' AND NEW.logtime < DATE '2012-03-01' ) THEN

        INSERT INTO p_201202 VALUES (NEW.*);

    ELSIF ( NEW.logtime >= DATE '2012-03-01' AND NEW.logtime < DATE '2012-04-01' ) THEN

        INSERT INTO p_201203 VALUES (NEW.*);

    ELSIF ( NEW.logtime >= DATE '2012-04-01' AND NEW.logtime < DATE '2012-05-01' ) THEN

        INSERT INTO p_201204 VALUES (NEW.*);

    ELSIF ( NEW.logtime >= DATE '2012-05-01' AND NEW.logtime < DATE '2012-06-01' ) THEN

        INSERT INTO p_201205 VALUES (NEW.*);

    ELSIF ( NEW.logtime >= DATE '2012-06-01' AND NEW.logtime < DATE '2012-07-01' ) THEN

        INSERT INTO p_201206 VALUES (NEW.*);

    ELSIF ( NEW.logtime >= DATE '2012-07-01' AND NEW.logtime < DATE '2012-08-01' ) THEN

        INSERT INTO p_201207 VALUES (NEW.*);

    ELSIF ( NEW.logtime >= DATE '2012-08-01' AND NEW.logtime < DATE '2012-09-01' ) THEN

        INSERT INTO p_201208 VALUES (NEW.*);

    ELSIF ( NEW.logtime >= DATE '2012-09-01' AND NEW.logtime < DATE '2012-10-01' ) THEN

        INSERT INTO p_201209 VALUES (NEW.*);

    ELSIF ( NEW.logtime >= DATE '2012-10-01' AND NEW.logtime < DATE '2012-11-01' ) THEN

        INSERT INTO p_201210 VALUES (NEW.*);

    ELSIF ( NEW.logtime >= DATE '2012-11-01' AND NEW.logtime < DATE '2012-12-01' ) THEN

        INSERT INTO p_201211 VALUES (NEW.*);

    ELSIF ( NEW.logtime >= DATE '2012-12-01' AND NEW.logtime < DATE '2013-01-01' ) THEN

        INSERT INTO p_201212 VALUES (NEW.*);

    ELSIF ( NEW.logtime >= DATE '2013-01-01' OR NEW.logtime < DATE '2012-01-01' ) THEN

        INSERT INTO p_default VALUES (NEW.*);

    ELSE

        RAISE EXCEPTION 'Date out of range.  Fix the p_insert_trigger() function!';

    END IF;

    RETURN NULL;

END;

$$ LANGUAGE plpgsql;

CREATE OR REPLACE FUNCTION p_delete_trigger()

RETURNS TRIGGER AS $$

BEGIN

    IF    ( OLD.logtime >= DATE '2012-01-01' AND OLD.logtime < DATE '2012-02-01' ) THEN

        DELETE FROM p_201201 WHERE logtime=OLD.logtime;

    ELSIF ( OLD.logtime >= DATE '2012-02-01' AND OLD.logtime < DATE '2012-03-01' ) THEN

        DELETE FROM p_201202 WHERE logtime=OLD.logtime;

    ELSIF ( OLD.logtime >= DATE '2012-03-01' AND OLD.logtime < DATE '2012-04-01' ) THEN

        DELETE FROM p_201203 WHERE logtime=OLD.logtime;

    ELSIF ( OLD.logtime >= DATE '2012-04-01' AND OLD.logtime < DATE '2012-05-01' ) THEN

        DELETE FROM p_201204 WHERE logtime=OLD.logtime;

    ELSIF ( OLD.logtime >= DATE '2012-05-01' AND OLD.logtime < DATE '2012-06-01' ) THEN

        DELETE FROM p_201205 WHERE logtime=OLD.logtime;

    ELSIF ( OLD.logtime >= DATE '2012-06-01' AND OLD.logtime < DATE '2012-07-01' ) THEN

        DELETE FROM p_201206 WHERE logtime=OLD.logtime;

    ELSIF ( OLD.logtime >= DATE '2012-07-01' AND OLD.logtime < DATE '2012-08-01' ) THEN

        DELETE FROM p_201207 WHERE logtime=OLD.logtime;

    ELSIF ( OLD.logtime >= DATE '2012-08-01' AND OLD.logtime < DATE '2012-09-01' ) THEN

        DELETE FROM p_201208 WHERE logtime=OLD.logtime;

    ELSIF ( OLD.logtime >= DATE '2012-09-01' AND OLD.logtime < DATE '2012-10-01' ) THEN

        DELETE FROM p_201209 WHERE logtime=OLD.logtime;

    ELSIF ( OLD.logtime >= DATE '2012-10-01' AND OLD.logtime < DATE '2012-11-01' ) THEN

        DELETE FROM p_201210 WHERE logtime=OLD.logtime;

    ELSIF ( OLD.logtime >= DATE '2012-11-01' AND OLD.logtime < DATE '2012-12-01' ) THEN

        DELETE FROM p_201211 WHERE logtime=OLD.logtime;

    ELSIF ( OLD.logtime >= DATE '2012-12-01' AND OLD.logtime < DATE '2013-01-01' ) THEN

        DELETE FROM p_201212 WHERE logtime=OLD.logtime;

    ELSIF ( OLD.logtime >= DATE '2013-01-01' OR OLD.logtime < DATE '2012-01-01' ) THEN

        DELETE FROM p_default WHERE logtime=OLD.logtime;

    ELSE

        RAISE EXCEPTION 'Date out of range.  Fix the p_insert_trigger() function!';

    END IF;

    RETURN NULL;

END;

$$ LANGUAGE plpgsql;

CREATE TRIGGER insert_p_trigger

    BEFORE INSERT ON p

    FOR EACH ROW EXECUTE PROCEDURE p_insert_trigger();

CREATE TRIGGER delete_p_trigger

    BEFORE DELETE ON p

    FOR EACH ROW EXECUTE PROCEDURE p_delete_trigger();

INSERT INTO p (city_id, logtime, peaktemp, unitsales) VALUES (1, timestamp '2012-01-02 12:59:59', 20, 10);

INSERT INTO p (city_id, logtime, peaktemp, unitsales) VALUES (2, timestamp '2012-02-02 12:59:59', 20, 10);

INSERT INTO p (city_id, logtime, peaktemp, unitsales) VALUES (3, timestamp '2012-03-02 12:59:59', 20, 10);

INSERT INTO p (city_id, logtime, peaktemp, unitsales) VALUES (4, timestamp '2012-04-02 12:59:59', 20, 10);

INSERT INTO p (city_id, logtime, peaktemp, unitsales) VALUES (5, timestamp '2012-05-02 12:59:59', 20, 10);

INSERT INTO p (city_id, logtime, peaktemp, unitsales) VALUES (6, timestamp '2012-06-02 12:59:59', 20, 10);

INSERT INTO p (city_id, logtime, peaktemp, unitsales) VALUES (7, timestamp '2012-07-02 12:59:59', 20, 10);

INSERT INTO p (city_id, logtime, peaktemp, unitsales) VALUES (8, timestamp '2012-08-02 12:59:59', 20, 10);

INSERT INTO p (city_id, logtime, peaktemp, unitsales) VALUES (9, timestamp '2012-09-02 12:59:59', 20, 10);

INSERT INTO p (city_id, logtime, peaktemp, unitsales) VALUES (10, timestamp '2012-10-02 12:59:59', 20, 10);

INSERT INTO p (city_id, logtime, peaktemp, unitsales) VALUES (11, timestamp '2012-11-02 12:59:59', 20, 10);

INSERT INTO p (city_id, logtime, peaktemp, unitsales) VALUES (12, timestamp '2012-12-02 12:59:59', 20, 10);

INSERT INTO p (city_id, logtime, peaktemp, unitsales) VALUES (13, timestamp '2013-01-02 12:59:59', 20, 10);

INSERT INTO p (city_id, logtime, peaktemp, unitsales) VALUES (14, timestamp '2011-12-02 12:59:59', 20, 10);

INSERT INTO p (city_id, logtime, peaktemp, unitsales) select m, timestamp '2012-02-02 12:59:59', 20, 10 from generate_series(1,10000) m;

explain select * from p_201202 where city_id=2 and logtime=timestamp '2012-02-02 12:59:59';

转载：https://yq.aliyun.com/articles/2637?spm=5176.100240.searchblog.12.59Jibq#

转载：postgresql分区与优化的更多相关文章

PostgreSQL之性能优化（转）
转载自:https://blog.csdn.net/huangwenyi1010/article/details/72853785 解决问题前言 PostgreSQL的配置参数作为性能调优的一部分, ...
PostgreSQL 分区索引演进
PostgreSQL 分区表,操作性相当便捷. 但只能在创建时决定是否为分区表,并决定分区条件字段,普通表创建后,不能在修改为分区表. Note:通过其他方法也可转化为分区表. 和其他数据库一样,分区 ...
【转载】PHP性能优化干货
PHP优化对于PHP的优化主要是对php.ini中的相关主要参数进行合理调整和设置,以下我们就来看看php.ini中的一些对性能影响较大的参数应该如何设置. # vi /etc/php.ini (1) ...
【转载】 Spark性能优化指南——基础篇
转自:http://tech.meituan.com/spark-tuning-basic.html?from=timeline 前言开发调优调优概述原则一:避免创建重复的RDD 原则二:尽可能 ...
【转载】MySQL性能优化的最佳20+条经验
今天,数据库的操作越来越成为整个应用的性能瓶颈了,这点对于Web应用尤其明显.关于数据库的性能,这并不只是DBA才需要担心的事,而这更是我们程序员需要去关注的事情.当我们去设计数据库表结构,对操作数据 ...
[转载]U3d常规性能优化技巧
以下技巧并不是必须的,但是对于想要提升游戏性能的人来说应该还是很不错的. 优化的常规技巧 n 剖析你的游戏. 不要花费时间来优化那些晦涩的代码或者缩减图形文件的大小,除非这是你游戏的瓶颈.第一次剖析你 ...
转载 50种方法优化SQL Server数据库查询
原文地址 http://www.cnblogs.com/zhycyq/articles/2636748.html 50种方法优化SQL Server数据库查询查询速度慢的原因很多,常见如下几种: 1 ...
【转载】 Spark性能优化：资源调优篇
在开发完Spark作业之后,就该为作业配置合适的资源了.Spark的资源参数,基本都可以在spark-submit命令中作为参数设置.很多Spark初学者,通常不知道该设置哪些必要的参数,以及如何设置 ...
PostgreSQL查询优化逻辑优化之其他
上一节我们介绍了PostgreSQL的子查询优化,子查询优化把一部分可以优化的子查询上拉到主查询成为join. preprocess_expression 将表达式(目标列,where,join,ha ...

随机推荐

利用oneproxy部署mysql数据库的读写分离
实验系统:CentOS 6.6_x86_64 实验前提:防火墙和selinux都关闭实验说明:本实验共有4台主机,IP分配如拓扑实验软件:mariadb-10.0.20 oneproxy-rhel ...
Memcached在windows下安装与使用
建议:windows系统下仅为测试所有,生产环境下服务端应使用Linux系统. 本文最后更新于:2014-08-03 18:24 原文:http://www.yaosansi.com/post/mem ...
Java报表工具FineReport导出EXCEL的四种API
在实际的应用中会经常需要将数据导出成excel,导出的方式除原样导出还有分页导出.分页分sheet导出和大数据量导出.对于excel 2003版,由于限制了每个sheet的最大行数和列数,大数据量导出 ...
洛谷CON1041 NOIP模拟赛一试
A T2-power of 2 题目描述是一个十分特殊的式子. 例如: n=0时 =2 然而,太大了所以,我们让对10007 取模输入输出格式输入格式: n 输出格式: % 10007 输入 ...
hibernate缓存
http://www.cnblogs.com/wean/archive/2012/05/16/2502724.html http://www.cnblogs.com/xiaoluo501395377/ ...
嵌入式Linux驱动学习之路(十四)按键驱动-同步、互斥、阻塞
目的:同一个时刻,只能有一个应用程序打开我们的驱动程序. ①原子操作: v = ATOMIC_INIT( i ) 定义原子变量v并初始化为i atomic_read(v) 返回原子变量 ...
iOS关于XML解析请求数据
XML数据的请求: 和json请求几乎一样,只有请求参数修改为xml即可: AFHTTPSessionManager *manager = [AFHTTPSessionManager manager] ...
BZOJ 1227 【SDOI2009】虔诚的墓主人
Description 小W 是一片新造公墓的管理人.公墓可以看成一块 \(N×M\) 的矩形,矩形的每个格点,要么种着一棵常青树,要么是一块还没有归属的墓地.当地的居民都是非常虔诚的基督徒,他们愿意 ...
Linux服务器高并发实践经历
作为一个师父离职早的野生程序员,业务方面还可以达到忽悠别人的水平,但上升到性能层面那就是硬伤. 真实天上掉馅饼,公司分配了一个测试性能的任务,真是感觉我的天空星星都亮了. 高并发主要限制因素:CPU. ...
【三石jQuery视频教程】03.创建垂直时间表(Timeline)
视频地址:http://v.qq.com/page/g/i/o/g0150rvi6io.html 大家好,欢迎来到[三石jQuery视频教程],我是您的老朋友 - 三生石上. 今天,我们要通过基本的H ...

转载：postgresql分区与优化

转载：postgresql分区与优化的更多相关文章

随机推荐

热门专题