iis服务器助手广告广告
返回顶部
首页 > 资讯 > 数据库 >PostgreSQL中不同数据类型对查询性能的影响有哪些
  • 300
分享到

PostgreSQL中不同数据类型对查询性能的影响有哪些

2024-04-02 19:04:59 300人浏览 安东尼
摘要

本篇内容主要讲解“postgresql中不同数据类型对查询性能的影响有哪些”,感兴趣的朋友不妨来看看。本文介绍的方法操作简单快捷,实用性强。下面就让小编来带大家学习“Postgresql中不同数据类型对查询

本篇内容主要讲解“postgresql中不同数据类型对查询性能的影响有哪些”,感兴趣的朋友不妨来看看。本文介绍的方法操作简单快捷,实用性强。下面就让小编来带大家学习“Postgresql中不同数据类型对查询性能的影响有哪些”吧!

容量
数据列占用空间大小

[local]:5432 pg12@testdb=# SELECT pg_column_size(SMALLINT '1'),pg_column_size(INT4 '1'), pg_column_size(NUMERIC(6,0) '1'),pg_column_size(FLOAT '1');
 pg_column_size | pg_column_size | pg_column_size | pg_column_size 
----------------+----------------+----------------+----------------
              2 |              4 |              8 |              8

创建数据表,0和1的数据值各插入100w行,查看数据表的占用空间大小。
numeric

[local]:5432 pg12@testdb=# create table t_numeric(id numeric);
CREATE TABLE
[local]:5432 pg12@testdb=# 
[local]:5432 pg12@testdb=# insert into t_numeric select 0 from generate_series(1,1000000);
INSERT 0 1000000
[local]:5432 pg12@testdb=# insert into t_numeric select 1 from generate_series(1,1000000);
INSERT 0 1000000
[local]:5432 pg12@testdb=# select pg_size_pretty(pg_relation_size('t_numeric'));
 pg_size_pretty 
----------------
 69 MB
(1 row)

float

[local]:5432 pg12@testdb=# create table t_float(id int);
CREATE TABLE
[local]:5432 pg12@testdb=# 
[local]:5432 pg12@testdb=# insert into t_float select 0 from generate_series(1,1000000);
INSERT 0 1000000
[local]:5432 pg12@testdb=# insert into t_float select 1 from generate_series(1,1000000);
INSERT 0 1000000
[local]:5432 pg12@testdb=# 
[local]:5432 pg12@testdb=# select pg_size_pretty(pg_relation_size('t_float'));
 pg_size_pretty 
----------------
 69 MB
(1 row)
[local]:5432 pg12@testdb=#

int

[local]:5432 pg12@testdb=# create table t_int(id int);
CREATE TABLE
[local]:5432 pg12@testdb=# 
[local]:5432 pg12@testdb=# insert into t_int select 0 from generate_series(1,1000000);
INSERT 0 1000000
[local]:5432 pg12@testdb=# insert into t_int select 1 from generate_series(1,1000000);
INSERT 0 1000000
[local]:5432 pg12@testdb=# select pg_size_pretty(pg_relation_size('t_int'));
 pg_size_pretty 
----------------
 69 MB
(1 row)

smallint

[local]:5432 pg12@testdb=# create table t_smallint(id smallint);
CREATE TABLE
[local]:5432 pg12@testdb=# 
[local]:5432 pg12@testdb=# insert into t_smallint select 0 from generate_series(1,1000000);
INSERT 0 1000000
[local]:5432 pg12@testdb=# insert into t_smallint select 1 from generate_series(1,1000000);
INSERT 0 1000000
[local]:5432 pg12@testdb=# 
[local]:5432 pg12@testdb=# select pg_size_pretty(pg_relation_size('t_smallint'));
 pg_size_pretty 
----------------
 69 MB
(1 row)

boolean

[local]:5432 pg12@testdb=# create table t_bool(id boolean);
CREATE TABLE
[local]:5432 pg12@testdb=# insert into t_bool select 0::boolean from generate_series(1,1000000);
INSERT 0 1000000
[local]:5432 pg12@testdb=# insert into t_bool select 1::boolean from generate_series(1,1000000);
INSERT 0 1000000
[local]:5432 pg12@testdb=# 
[local]:5432 pg12@testdb=# select pg_size_pretty(pg_relation_size('t_bool'));
 pg_size_pretty 
----------------
 69 MB
(1 row)

可以看到,四种数据类型占用的空间都是69 MB。

查询性能
不加条件,全表扫描

-- 禁用并行
[local]:5432 pg12@testdb=# SET max_parallel_workers_per_gather = 0;
SET
[local]:5432 pg12@testdb=# explain (analyze,verbose,buffers) select count(*) from t_numeric;
                                                            QUERY PLAN                                                            
----------------------------------------------------------------------------------------------------------------------------------
 Aggregate  (cost=33850.00..33850.01 rows=1 width=8) (actual time=478.196..478.196 rows=1 loops=1)
   Output: count(*)
   Buffers: shared hit=8850
   ->  Seq Scan on public.t_numeric  (cost=0.00..28850.00 rows=2000000 width=0) (actual time=0.053..255.949 rows=2000000 loops=1)
         Output: id
         Buffers: shared hit=8850
 Planning Time: 0.716 ms
 Execution Time: 478.280 ms
(8 rows)
[local]:5432 pg12@testdb=# explain (analyze,verbose,buffers) select count(*) from t_float;
                                                           QUERY PLAN                                                           
--------------------------------------------------------------------------------------------------------------------------------
 Aggregate  (cost=33850.00..33850.01 rows=1 width=8) (actual time=421.919..421.919 rows=1 loops=1)
   Output: count(*)
   Buffers: shared hit=8850
   ->  Seq Scan on public.t_float  (cost=0.00..28850.00 rows=2000000 width=0) (actual time=0.010..222.624 rows=2000000 loops=1)
         Output: id
         Buffers: shared hit=8850
 Planning Time: 0.231 ms
 Execution Time: 421.948 ms
(8 rows)
[local]:5432 pg12@testdb=# explain (analyze,verbose,buffers) select count(*) from t_int;
                                                          QUERY PLAN                                                          
------------------------------------------------------------------------------------------------------------------------------
 Aggregate  (cost=33850.00..33850.01 rows=1 width=8) (actual time=440.328..440.328 rows=1 loops=1)
   Output: count(*)
   Buffers: shared hit=8850
   ->  Seq Scan on public.t_int  (cost=0.00..28850.00 rows=2000000 width=0) (actual time=0.011..236.078 rows=2000000 loops=1)
         Output: id
         Buffers: shared hit=8850
 Planning Time: 0.208 ms
 Execution Time: 440.359 ms
(8 rows)
[local]:5432 pg12@testdb=#  explain (analyze,verbose,buffers) select count(*) from t_smallint;
                                                            QUERY PLAN                                                             
-----------------------------------------------------------------------------------------------------------------------------------
 Aggregate  (cost=33850.00..33850.01 rows=1 width=8) (actual time=439.007..439.007 rows=1 loops=1)
   Output: count(*)
   Buffers: shared hit=8850
   ->  Seq Scan on public.t_smallint  (cost=0.00..28850.00 rows=2000000 width=0) (actual time=0.043..232.069 rows=2000000 loops=1)
         Output: id
         Buffers: shared hit=8850
 Planning Time: 0.553 ms
 Execution Time: 439.081 ms
(8 rows)
[local]:5432 pg12@testdb=# explain (analyze,verbose,buffers) select count(*) from t_bool;
                                                          QUERY PLAN                                                           
-------------------------------------------------------------------------------------------------------------------------------
 Aggregate  (cost=33850.00..33850.01 rows=1 width=8) (actual time=430.800..430.800 rows=1 loops=1)
   Output: count(*)
   Buffers: shared hit=8850
   ->  Seq Scan on public.t_bool  (cost=0.00..28850.00 rows=2000000 width=0) (actual time=0.010..230.333 rows=2000000 loops=1)
         Output: id
         Buffers: shared hit=8850
 Planning Time: 0.224 ms
 Execution Time: 430.831 ms
(8 rows)
[local]:5432 pg12@testdb=#

不带条件全表扫描,时间相差不大,执行时长最大的是numeric类型。

添加查询条件,全表扫描

[local]:5432 pg12@testdb=# explain (analyze,verbose,buffers) select count(*) from t_numeric where id = '0'::numeric;
lain (analyze,verbose,buffers) select count(*) from t_bool where id = 0::boolean;
                                                            QUERY PLAN                                                            
----------------------------------------------------------------------------------------------------------------------------------
 Aggregate  (cost=36358.67..36358.68 rows=1 width=8) (actual time=723.356..723.357 rows=1 loops=1)
   Output: count(*)
   Buffers: shared hit=8850
   ->  Seq Scan on public.t_numeric  (cost=0.00..33850.00 rows=1003467 width=0) (actual time=0.057..610.907 rows=1000000 loops=1)
         Output: id
         Filter: (t_numeric.id = '0'::numeric)
         Rows Removed by Filter: 1000000
         Buffers: shared hit=8850
 Planning Time: 1.901 ms
 Execution Time: 723.449 ms
(10 rows)
[local]:5432 pg12@testdb=# explain (analyze,verbose,buffers) select count(*) from t_float where id = '0'::numeric;
                                                          QUERY PLAN                                                          
------------------------------------------------------------------------------------------------------------------------------
 Aggregate  (cost=38875.00..38875.01 rows=1 width=8) (actual time=827.686..827.687 rows=1 loops=1)
   Output: count(*)
   Buffers: shared hit=8850
   ->  Seq Scan on public.t_float  (cost=0.00..38850.00 rows=10000 width=0) (actual time=0.015..725.737 rows=1000000 loops=1)
         Output: id
         Filter: ((t_float.id)::numeric = '0'::numeric)
         Rows Removed by Filter: 1000000
         Buffers: shared hit=8850
 Planning Time: 0.234 ms
 Execution Time: 827.720 ms
(10 rows)
[local]:5432 pg12@testdb=# explain (analyze,verbose,buffers) select count(*) from t_int where id = 0;
                                                         QUERY PLAN                                                          
-----------------------------------------------------------------------------------------------------------------------------
 Aggregate  (cost=36329.50..36329.51 rows=1 width=8) (actual time=434.067..434.067 rows=1 loops=1)
   Output: count(*)
   Buffers: shared hit=8850
   ->  Seq Scan on public.t_int  (cost=0.00..33850.00 rows=991800 width=0) (actual time=0.014..333.883 rows=1000000 loops=1)
         Output: id
         Filter: (t_int.id = 0)
         Rows Removed by Filter: 1000000
         Buffers: shared hit=8850
 Planning Time: 0.295 ms
 Execution Time: 434.101 ms
(10 rows)
[local]:5432 pg12@testdb=# explain (analyze,verbose,buffers) select count(*) from t_smallint where id = 0;
                                                            QUERY PLAN                                                             
-----------------------------------------------------------------------------------------------------------------------------------
 Aggregate  (cost=36354.50..36354.51 rows=1 width=8) (actual time=486.466..486.466 rows=1 loops=1)
   Output: count(*)
   Buffers: shared hit=8850
   ->  Seq Scan on public.t_smallint  (cost=0.00..33850.00 rows=1001800 width=0) (actual time=0.053..368.184 rows=1000000 loops=1)
         Output: id
         Filter: (t_smallint.id = 0)
         Rows Removed by Filter: 1000000
         Buffers: shared hit=8850
 Planning Time: 1.396 ms
 Execution Time: 486.554 ms
(10 rows)
[local]:5432 pg12@testdb=# explain (analyze,verbose,buffers) select count(*) from t_bool where id = 0::boolean;
                                                          QUERY PLAN                                                           
-------------------------------------------------------------------------------------------------------------------------------
 Aggregate  (cost=31356.67..31356.68 rows=1 width=8) (actual time=416.510..416.510 rows=1 loops=1)
   Output: count(*)
   Buffers: shared hit=8850
   ->  Seq Scan on public.t_bool  (cost=0.00..28850.00 rows=1002667 width=0) (actual time=0.014..316.188 rows=1000000 loops=1)
         Output: id
         Filter: (NOT t_bool.id)
         Rows Removed by Filter: 1000000
         Buffers: shared hit=8850
 Planning Time: 0.261 ms
 Execution Time: 416.551 ms
(10 rows)
[local]:5432 pg12@testdb=#

存在查询条件的情况下,由于解析表达式的代价不同(bool < int < numeric < float),因此时间相差较大,时长最大的是float类型,时间接近bool类型的2倍。

创建索引,全索引扫描
禁用全表扫描,使用全索引扫描

[local]:5432 pg12@testdb=# explain (analyze,verbose,buffers) select count(*) from t_numeric where id = '0'::numeric;
                                                                           QUERY PLAN                                                                           
----------------------------------------------------------------------------------------------------------------------------------------------------------------
 Aggregate  (cost=35541.77..35541.78 rows=1 width=8) (actual time=594.984..594.984 rows=1 loops=1)
   Output: count(*)
   Buffers: shared hit=7160
   ->  Index Only Scan using idx_t_numeric_id on public.t_numeric  (cost=0.43..33033.10 rows=1003467 width=0) (actual time=0.269..482.525 rows=1000000 loops=1)
         Output: id
         Index Cond: (t_numeric.id = '0'::numeric)
         Heap Fetches: 1000000
         Buffers: shared hit=7160
 Planning Time: 1.392 ms
 Execution Time: 595.253 ms
(10 rows)
[local]:5432 pg12@testdb=# explain (analyze,verbose,buffers) select count(*) from t_float where id = '0'::numeric;
                                                                        QUERY PLAN                                                                         
-----------------------------------------------------------------------------------------------------------------------------------------------------------
 Aggregate  (cost=70854.43..70854.44 rows=1 width=8) (actual time=1337.093..1337.094 rows=1 loops=1)
   Output: count(*)
   Buffers: shared hit=14317
   ->  Index Only Scan using idx_t_float_id on public.t_float  (cost=0.43..70829.43 rows=10000 width=0) (actual time=0.037..1233.730 rows=1000000 loops=1)
         Output: id
         Filter: ((t_float.id)::numeric = '0'::numeric)
         Rows Removed by Filter: 1000000
         Heap Fetches: 2000000
         Buffers: shared hit=14317
 Planning Time: 0.293 ms
 Execution Time: 1337.168 ms
(11 rows)
[local]:5432 pg12@testdb=# explain (analyze,verbose,buffers) select count(*) from t_int where id = 0;
                                                                      QUERY PLAN                                                                       
-------------------------------------------------------------------------------------------------------------------------------------------------------
 Aggregate  (cost=35128.43..35128.44 rows=1 width=8) (actual time=526.942..526.943 rows=1 loops=1)
   Output: count(*)
   Buffers: shared hit=7160
   ->  Index Only Scan using idx_t_int_id on public.t_int  (cost=0.43..32648.93 rows=991800 width=0) (actual time=0.035..414.797 rows=1000000 loops=1)
         Output: id
         Index Cond: (t_int.id = 0)
         Heap Fetches: 1000000
         Buffers: shared hit=7160
 Planning Time: 0.245 ms
 Execution Time: 526.979 ms
(10 rows)
[local]:5432 pg12@testdb=# explain (analyze,verbose,buffers) select count(*) from t_smallint where id = 0;
                                                                            QUERY PLAN                                                                            
------------------------------------------------------------------------------------------------------------------------------------------------------------------
 Aggregate  (cost=35480.43..35480.44 rows=1 width=8) (actual time=551.394..551.394 rows=1 loops=1)
   Output: count(*)
   Buffers: shared hit=4428 read=2735
   ->  Index Only Scan using idx_t_smallint_id on public.t_smallint  (cost=0.43..32975.93 rows=1001800 width=0) (actual time=0.459..438.992 rows=1000000 loops=1)
         Output: id
         Index Cond: (t_smallint.id = 0)
         Heap Fetches: 1000000
         Buffers: shared hit=4428 read=2735
 Planning Time: 1.889 ms
 Execution Time: 551.499 ms
(10 rows)
[local]:5432 pg12@testdb=# explain (analyze,verbose,buffers) select count(*) from t_bool where id = 0::boolean;
                                                                        QUERY PLAN                                                                        
----------------------------------------------------------------------------------------------------------------------------------------------------------
 Aggregate  (cost=35513.77..35513.78 rows=1 width=8) (actual time=497.886..497.886 rows=1 loops=1)
   Output: count(*)
   Buffers: shared hit=7160
   ->  Index Only Scan using idx_t_bool_id on public.t_bool  (cost=0.43..33007.10 rows=1002667 width=0) (actual time=0.035..393.653 rows=1000000 loops=1)
         Output: id
         Index Cond: (t_bool.id = false)
         Heap Fetches: 1000000
         Buffers: shared hit=7160
 Planning Time: 0.250 ms
 Execution Time: 497.922 ms
(10 rows)
[local]:5432 pg12@testdb=#

走全索引扫描,执行时长最长的仍是float类型,其他三种类型则相差不大,numeric的性能相较全表扫描有明显提升(595ms vs 723ms)。

压力测试
使用pgbench进行压力测试,numeric/float/int三种类型,各插入100w数据

drop table t_big_numeric;
create table t_big_numeric(id numeric);
insert into t_big_numeric select 0 from generate_series(1,1000000);
drop table t_big_float;
create table t_big_float(id int);
insert into t_big_float select 0 from generate_series(1,1000000);
drop table t_big_int;
create table t_big_int(id int);
insert into t_big_int select 0 from generate_series(1,1000000);

测试结果

[pg12@localhost test]$ pgbench -C -f ./select_numeric.sql --time=120 --client=8 --jobs=2 -d testdb
...
transaction type: ./select_numeric.sql
scaling factor: 1
query mode: simple
number of clients: 8
number of threads: 2
duration: 120 s
number of transactions actually processed: 1254
latency average = 768.659 ms
tps = 10.407739 (including connections establishing)
tps = 10.906626 (excluding connections establishing)
[pg12@localhost test]$ 
[pg12@localhost test]$ pgbench -C -f ./select_float.sql --time=120 --client=8 --jobs=2 -d testdb
...
transaction type: ./select_float.sql
scaling factor: 1
query mode: simple
number of clients: 8
number of threads: 2
duration: 120 s
number of transactions actually processed: 2167
latency average = 444.006 ms
tps = 18.017778 (including connections establishing)
tps = 19.461350 (excluding connections establishing)
[pg12@localhost test]$ cat select_float.sql 
\set id random(1,1000000)
select * from t_big_float where id = :id; 
[pg12@localhost test]$ 
[pg12@localhost test]$ pgbench -C -f ./select_int.sql --time=120 --client=8 --jobs=2 -d testdb
...
transaction type: ./select_int.sql
scaling factor: 1
query mode: simple
number of clients: 8
number of threads: 2
duration: 120 s
number of transactions actually processed: 2184
latency average = 440.271 ms
tps = 18.170626 (including connections establishing)
tps = 19.658996 (excluding connections establishing)
[pg12@localhost test]$

到此,相信大家对“PostgreSQL中不同数据类型对查询性能的影响有哪些”有了更深的了解,不妨来实际操作一番吧!这里是编程网网站,更多相关内容可以进入相关频道进行查询,关注我们,继续学习!

您可能感兴趣的文档:

--结束END--

本文标题: PostgreSQL中不同数据类型对查询性能的影响有哪些

本文链接: https://www.lsjlt.com/news/63567.html(转载时请注明来源链接)

有问题或投稿请发送至: 邮箱/279061341@qq.com    QQ/279061341

本篇文章演示代码以及资料文档资料下载

下载Word文档到电脑,方便收藏和打印~

下载Word文档
猜你喜欢
软考高级职称资格查询
编程网,编程工程师的家园,是目前国内优秀的开源技术社区之一,形成了由开源软件库、代码分享、资讯、协作翻译、讨论区和博客等几大频道内容,为IT开发者提供了一个发现、使用、并交流开源技术的平台。
  • 官方手机版

  • 微信公众号

  • 商务合作