别裁员!别裁员!别裁员! 一招降低IT数据库成本

背景

2020疫情无情，诸多企业因此受挫，特别中小企业，甚至到了要裁员的地步, 但是人才是最宝贵的，裁员一定是下下策，如何渡过这个难关，疫情带给我们什么反思?

开源节流有新方法，通常数据库在企业IT支出中的占比将近一半，降低数据库成本对降低企业IT成本效果明显，但是一般企业没有专业DBA，很难在这方面下手，不过没关系，有了云厂商，一切变得简单。借助阿里云我们找到了可以为企业IT节省至少一倍成本的方法.

到底时什么方法呢? 回顾一下年前做的一系列MySQL+PG联合解决方案的课程.

《阿里云 RDS PostgreSQL+MySQL 联合解决方案课程 - 汇总视频、课件》

在众多数据库中, PG是一个企业级的开源数据库, 各方面的功能与Oracle对齐, 适合范围广, 能处理的数据量庞大. 采用PG的大型企业例如平安，邮储银行，阿里，华为，中兴，人保, 招商, 富士康, 苹果, SAP, saleforce等以及全球财富1000强等众多企业。 《外界对PostgreSQL 的评价》

阿里云RDS PG的优势:

· 支持完整生命周期管理,包括高可用, 容灾, 备份, 安全, 审计, 加密, cloud dba等模块, 大幅降低企业的使用和管理成本.

· 专业内核和DBA团队 7*24小时服务.

· 支持并行计算，LLVM，GPU加速，向量计算，分析能力更强。

· PG的优化器强大，应对复杂SQL处理效率更高，适合复杂业务场景, 更适合新零售、制造业、工业、在线教育、游戏、金融、政府、企业ERP等行业或领域。

· 内核扩展, 根据垂直领域的需求定制化。

o Ganos插件, GIS功能更强更专业，支持平面、球面几何，栅格，时空轨迹，点云，拓扑网络模型。

o pase插件, 支持高维向量搜索, 支持精确的图像搜索, 人脸识别, 相似查询.

o roaringbitmap插件, 支持实时大数据用户画像, 精准营销.

o rdkit插件, 支持化学分析, 分子式的相似搜索, 化学机器学习等.

· 多模能力更强，其表现在索引更丰富，除了btree,hash还支持gin,gist,spgist,brin,bloom,rum等索引接口，适合模糊搜索，全文检索，多维任意搜索，时空搜索，高维向量（广泛应用于图像识别、相似特征扩选，时序搜索，用户画像，化学分析，DNA检索等。

· 类型更加丰富，同时支持扩展类型，除了基本类型以外，支持网络、全文检索、数组、xml、JSON、范围、域、树、多维、分子、GIS等类型。支持更丰富的应用场景。

· 支持oss_fdw, 可以将数据库的归档数据存储在oss中, 降低成本, 并且访问方法不变.

本文将对PG和MySQL进行多方位对比, 在某些方面PG的综合性能比MySQL高出一个数量级, PG+MySQL结合使用, 可以大幅降低企业成本.

疫情无情PG有情, 别裁员了, 建立多元化的技术栈, 强化企业IT能力更重要.

环境

阿里云RDS PG 12实例, 8核32G 1500G ESSD

同硬件配置的MySQL 8.0

用户密码:

user123      

xxxxxxxx!      

库:

db1      

连接串:

PG:

export PGPASSWORD=xxxxxxxx!      
psql -h pgm-bp1z26gbo3gx893a129310.pg.rds.aliyuncs.com -p 1433 -U user123 db1      
MySQL:
mysql -h rm-bp1wv992ym962k85888370.mysql.rds.aliyuncs.com -P 3306 -u user123 --password=xxxxxx! -D db1

测试用的客户端ecs centos 7.x x64安装mysql, pg客户端

yum install -y mysql-*      
yum install https://download.postgresql.org/pub/repos/yum/reporpms/EL-7-x86_64/pgdg-redhat-repo-latest.noarch.rpm      
yum install -y postgresql12

MySQL 8.0测试

测试表

CREATE TABLE employees (      
id INT NOT NULL,      
  fname VARCHAR(30),      
  lname VARCHAR(30),      
  birth TIMESTAMP,      
  hired DATE NOT NULL DEFAULT '1970-01-01',      
  separated DATE NOT NULL DEFAULT '9999-12-31',      
  job_code INT NOT NULL,      
  store_id INT NOT NULL
);

批量写入存储过程

DROP PROCEDURE IF EXISTS BatchInsert;      

delimiter //   -- 把界定符改成双斜杠      
CREATE PROCEDURE BatchInsert(IN init INT, IN loop_time INT)  -- 第一个参数为初始ID号（可自定义），第二个位生成MySQL记录个数      
  BEGIN      
DECLARE Var INT;      
DECLARE ID INT;      
      SET Var = 0;      
      SET ID = init;      
WHILE Var < loop_time DO
          insert into employees      
          (id, fname, lname, birth, hired, separated, job_code, store_id)       
          values       
          (ID, CONCAT('chen', ID), CONCAT('haixiang', ID), Now(), Now(), Now(), 1, ID);      
          SET ID = ID + 1;      
          SET Var = Var + 1;      
      END WHILE;      
  END;      
//      
delimiter ;  -- 界定符改回分号

批量写入20万条

-- 开启事务插入,否则会很慢      

begin;      

CALL BatchInsert(1, 200000);      

commit;      

Query OK, 1 row affected (7.53 sec)      

使用insert into继续批量写入

mysql> insert into employees select * from employees;      

Query OK, 200000 rows affected (1.61 sec)      

Records: 200000  Duplicates: 0  Warnings: 0      

mysql> insert into employees select * from employees;      

Query OK, 400000 rows affected (3.25 sec)      

Records: 400000  Duplicates: 0  Warnings: 0      

mysql> insert into employees select * from employees;      

Query OK, 800000 rows affected (6.51 sec)      

Records: 800000  Duplicates: 0  Warnings: 0      

mysql> insert into employees select * from employees;      

Query OK, 1600000 rows affected (12.93 sec)      

Records: 1600000  Duplicates: 0  Warnings: 0      

mysql> insert into employees select * from employees;      

Query OK, 3200000 rows affected (28.61 sec)      

Records: 3200000  Duplicates: 0  Warnings: 0      

mysql> insert into employees select * from employees;      

Query OK, 6400000 rows affected (56.48 sec)      

Records: 6400000  Duplicates: 0  Warnings: 0      

mysql> insert into employees select * from employees;      

Query OK, 12800000 rows affected (1 min 55.30 sec)      

Records: 12800000  Duplicates: 0  Warnings: 0      

查询性能

mysql> select count(*) from employees;      

+----------+      

| count(*) |      

+----------+      

| 25600000 |      

+----------+      

1 row in set (6.15 sec)      

求distinct性能

mysql> select count(distinct id) from employees ;      
+--------------------+      
| count(distinct id) |      
+--------------------+      
|             200000 |      
+--------------------+      
1 row in set (16.67 sec)

分组求distinct性能

mysql> select count(*) from (select id from employees group by id) t;      
+----------+      
| count(*) |      
+----------+      
|   200000 |      
+----------+      
1 row in set (15.52 sec)

再写入200万

begin;      
CALL BatchInsert(1, 2000000);      
commit;

测试表2, 写入200万.

CREATE TABLE employees1 (      
  id INT NOT NULL,      
  fname VARCHAR(30),      
  lname VARCHAR(30),      
  birth TIMESTAMP,      
  hired DATE NOT NULL DEFAULT '1970-01-01',      
  separated DATE NOT NULL DEFAULT '9999-12-31',      
  job_code INT NOT NULL,      
  store_id INT NOT NULL
);      

DROP PROCEDURE IF EXISTS BatchInser1;      

delimiter //   -- 把界定符改成双斜杠      
CREATE PROCEDURE BatchInsert1(IN init INT, IN loop_time INT)  -- 第一个参数为初始ID号（可自定义），第二个位生成MySQL记录个数      
  BEGIN      
DECLARE Var INT;      
DECLARE ID INT;      
      SET Var = 0;      
      SET ID = init;      
WHILE Var < loop_time DO
          insert into employees1      
          (id, fname, lname, birth, hired, separated, job_code, store_id)       
          values       
          (ID, CONCAT('chen', ID), CONCAT('haixiang', ID), Now(), Now(), Now(), 1, ID);      
          SET ID = ID + 1;      
          SET Var = Var + 1;      
      END WHILE;      
  END;      
//      
delimiter ;  -- 界定符改回分号

使用loop insert写入200万行

-- 开启事务插入,否则会很慢      begin;      CALL BatchInsert1(1, 2000000);      commit;      Query OK, 1 row affected (1 min 7.06 sec)

2560万 多对一JOIN 200万, 分组,排序

select t1.lname,count(*) from employees t1 join employees1 t2 using (id) group by t1.lname order by count(*) desc,lname limit 10;

简单查询性能(因为以上查询几个小时都没有出结果, 不得不新建一个200万的表进行查询测试):

CREATE TABLE employees2 (      
id INT NOT NULL,      
fname VARCHAR(30),      
lname VARCHAR(30),      
birth TIMESTAMP,      
hired DATE NOT NULL DEFAULT '1970-01-01',      
separated DATE NOT NULL DEFAULT '9999-12-31',      
job_code INT NOT NULL,      
store_id INT NOT NULL      
);      

DROP PROCEDURE IF EXISTS BatchInser2;      

delimiter //   -- 把界定符改成双斜杠      
CREATE PROCEDURE BatchInsert2(IN init INT, IN loop_time INT)  -- 第一个参数为初始ID号（可自定义），第二个位生成MySQL记录个数      
BEGIN      
DECLARE Var INT;      
DECLARE ID INT;      
SET Var = 0;      
SET ID = init;      
WHILE Var < loop_time DO      
insert into employees2      
(id, fname, lname, birth, hired, separated, job_code, store_id)       
values       
(ID, CONCAT('chen', ID), CONCAT('haixiang', ID), Now(), Now(), Now(), 1, ID);      
SET ID = ID + 1;      
SET Var = Var + 1;      
END WHILE;      
END;      
//      
delimiter ;  -- 界定符改回分号      

-- 开启事务插入,否则会很慢      

begin;      
CALL BatchInsert2(1, 2000000);      
commit;      

Query OK, 1 row affected (1 min 7.06 sec)

创建索引

create index idx_employees2_1 on employees2(id);

建立查询存储过程, 查询200万次.

DROP PROCEDURE IF EXISTS select1;      

delimiter //   -- 把界定符改成双斜杠      
CREATE PROCEDURE select1(IN init INT, IN loop_time INT)  -- 第一个参数为初始ID号（可自定义），第二个位生成MySQL记录个数      
BEGIN      
DECLARE Var INT;      
DECLARE ID1 INT;      
DECLARE vid INT;      
DECLARE vfname VARCHAR(30);      
DECLARE vlname VARCHAR(30);      
DECLARE vbirth TIMESTAMP;      
DECLARE vhired DATE;      
DECLARE vseparated DATE;      
DECLARE vjob_code INT;      
DECLARE vstore_id INT;      
SET Var = 0;      
SET ID1 = init;      
WHILE Var < loop_time DO      
select t.id,t.fname,t.lname,t.birth,t.hired,t.separated,t.job_code,t.store_id       
into       
vid,vfname,vlname,vbirth,vhired,vseparated,vjob_code,vstore_id      
from employees2 t       
where t.id=id1;      
SET ID1 = ID1 + 1;      
SET Var = Var + 1;      
END WHILE;      
END;      
//      
delimiter ;  -- 界定符改回分号

基于KEY简单查询, 查询200万次的耗时.

-- 开启事务查询      

begin;      
CALL select1(1, 2000000);      
commit;      


Query OK, 1 row affected (1 min 10.23 sec)

MySQL 1亿+:

继续测试到1亿数据量.

mysql> insert into employees select * from employees;    Query OK, 27600000 rows affected (4 min 38.62 sec)    Records: 27600000  Duplicates: 0  Warnings: 0    mysql> insert into employees select * from employees;    Query OK, 55200000 rows affected (11 min 13.40 sec)    Records: 55200000  Duplicates: 0  Warnings: 0    mysql> select count(*) from employees;    +-----------+    | count(*)  |    +-----------+    | 110400000 |    +-----------+    1 row in set (28.00 sec)    mysql> select count(distinct id) from employees ;      +--------------------+    | count(distinct id) |    +--------------------+    |            2000000 |    +--------------------+    1 row in set (1 min 17.73 sec)    mysql> select count(*) from (select id from employees group by id) t;      +----------+    | count(*) |    +----------+    |  2000000 |    +----------+    1 row in set (1 min 24.64 sec)

1.1亿全量数据更新

mysql> update employees set lname=lname||'new';  
Query OK, 110400000 rows affected, 65535 warnings (21 min 30.34 sec)  
Rows matched: 110400000  Changed: 110400000  Warnings: 220800000

1.1亿 多对一JOIN 200万, 分组,排序, 超过3小时没有查询出结果.

select t1.lname,count(*) from employees t1 join employees1 t2 using (id) group by t1.lname order by count(*) desc,lname limit 10;

1.1亿创建索引

mysql> create index idx_employees_1 on employees(id);    
Query OK, 0 rows affected (3 min 49.04 sec)    
Records: 0  Duplicates: 0  Warnings: 0

阿里云RDS PostgreSQL 12测试    

测试表

CREATE TABLE employees (      
id INT NOT NULL,      
  fname VARCHAR(30),      
  lname VARCHAR(30),      
  birth TIMESTAMP,      
  hired DATE NOT NULL DEFAULT '1970-01-01',      
  separated DATE NOT NULL DEFAULT '9999-12-31',      
  job_code INT NOT NULL,      
  store_id INT NOT NULL
);

直接使用srf快速写入20万数据

\timing      

insert into employees      
    (id, fname, lname, birth, hired, separated, job_code, store_id)       
select
ID, CONCAT('chen', ID), CONCAT('haixiang', ID), Now(), Now(), Now(), 1, ID
from generate_series(1,200000) id;      

INSERT 0 200000
Time: 355.652 ms

也可以使用和mysql一样的方法loop insert写入20万

create or replace function BatchInsert(IN init INT, IN loop_time INT)  -- 第一个参数为初始ID号（可自定义），第二个位生成记录个数      
returns void as $$      
DECLARE
Var INT := 0;      
begin
for id in init..init+loop_time-1 loop
insert into employees      
    (id, fname, lname, birth, hired, separated, job_code, store_id)       
values
    (ID, CONCAT('chen', ID), CONCAT('haixiang', ID), Now(), Now(), Now(), 1, ID);      
end loop;      
end;      
$$ language plpgsql strict;      


db1=# select batchinsert(1,200000);      
 batchinsert       
-------------      

(1 row)      
Time: 1292.559 ms (00:01.293)

使用insert into继续批量写入

db1=> insert into employees select * from employees ;      
INSERT 0 400000      
Time: 322.335 ms      
db1=> insert into employees select * from employees ;      
INSERT 0 800000      
Time: 835.365 ms      
db1=> insert into employees select * from employees ;      
INSERT 0 1600000      
Time: 1622.475 ms (00:01.622)      
db1=> insert into employees select * from employees ;      
INSERT 0 3200000      
Time: 3583.787 ms (00:03.584)      
db1=> insert into employees select * from employees ;      
INSERT 0 6400000      
Time: 7277.764 ms (00:07.278)      
db1=> insert into employees select * from employees ;      
INSERT 0 12800000      
Time: 15639.482 ms (00:15.639)      
db1=> \dt+ employees       
List of relations      
Schema |   Name    | Type  |  Owner  |  Size   | Description       
--------+-----------+-------+---------+---------+-------------      
public | employees | table | user123 | 2061 MB |       
(1 row)

查询性能

db1=> select count(*) from employees ;      
count         
----------      
25600000      
(1 row)      

Time: 604.982 ms

求distinct性能

db1=> select count(*) from (select id from employees group by id) t;      
 count        
--------      
200000
(1 row)      

Time: 2982.907 ms (00:02.983)

再写入200万

insert into employees      
    (id, fname, lname, birth, hired, separated, job_code, store_id)       
select
ID, CONCAT('chen', ID), CONCAT('haixiang', ID), Now(), Now(), Now(), 1, ID
from generate_series(1,2000000) id;

测试表2, 写入200万.

CREATE TABLE employees1 (      
id INT NOT NULL,      
  fname VARCHAR(30),      
  lname VARCHAR(30),      
  birth TIMESTAMP,      
  hired DATE NOT NULL DEFAULT '1970-01-01',      
  separated DATE NOT NULL DEFAULT '9999-12-31',      
  job_code INT NOT NULL,      
  store_id INT NOT NULL
);      


insert into employees1      
    (id, fname, lname, birth, hired, separated, job_code, store_id)       
select
ID, CONCAT('chen', ID), CONCAT('haixiang', ID), Now(), Now(), Now(), 1, ID
from generate_series(1,2000000) id;      

INSERT 0 2000000
Time: 3037.777 ms (00:03.038)

2560万 多对一JOIN 200万, 分组,排序

select t1.lname,count(*) from employees t1 join employees1 t2 using (id) group by t1.lname order by count(*) desc,lname limit 10;      
     lname      | count       
----------------+-------      
 haixiang1      |   129      
 haixiang10     |   129      
 haixiang100    |   129      
 haixiang1000   |   129      
 haixiang10000  |   129      
 haixiang100000 |   129      
 haixiang100001 |   129      
 haixiang100002 |   129      
 haixiang100003 |   129      
 haixiang100004 |   129      
(10 rows)      

Time: 8897.907 ms (00:08.898)

简单查询性能:

创建索引

create index idx_employees1_1 on employees1(id);      
CREATE INDEX
Time: 1436.346 ms (00:01.436)

基于KEY简单查询, 查询200万次的耗时.

do language plpgsql $$       
declare
begin
for i in 1..2000000 loop
    perform * from employees1 where id=i;      
end loop;      
end;      
$$;      

DO
Time: 9515.728 ms (00:09.516)      
db1=> select 9515.728/2000000;      
        ?column?              
------------------------      
 0.00475786400000000000      
(1 row)

PG 1亿+:    

db1=> INSERT INTO employees select * from employees;      INSERT 0 27600000      Time: 25050.665 ms (00:25.051)      db1=> INSERT INTO employees select * from employees;      INSERT 0 55200000      Time: 64726.430 ms (01:04.726)      db1=> select count(*) from employees;      count         -----------      110400000      (1 row)      Time: 7286.152 ms (00:07.286)      db1=> select count(distinct id) from employees;      count        ---------      2000000      (1 row)      Time: 39783.068 ms (00:39.783)      db1=> select count(*) from (select id from employees group by id) t;      count        ---------      2000000      (1 row)      Time: 14668.305 ms (00:14.668)      db1=> select t1.lname,count(*) from employees t1 join employees1 t2 using (id) group by t1.lname order by count(*) desc,lname limit 10;      lname      | count       ----------------+-------      haixiang1      |   516      haixiang10     |   516      haixiang100    |   516      haixiang1000   |   516      haixiang10000  |   516      haixiang100000 |   516      haixiang100001 |   516      haixiang100002 |   516      haixiang100003 |   516      haixiang100004 |   516      (10 rows)      Time: 33731.431 ms (00:33.731)

更新1.1亿

db1=> update employees set lname=lname||'new';  
UPDATE 110400000  
Time: 385372.063 ms (06:25.372)

创建索引:

db1=> create index idx_employees_1 on employees(id);    
CREATE INDEX    
Time: 70450.491 ms (01:10.450)

MySQL vs PG 性能报表

8核32G 1500G essd云盘， MySQL 8.0 vs PG 12

数据量	sql	MySQL耗时	PG耗时
20万	{写入}存储过程loop insert	7.53 s	1.29 s
20万	{写入}SRF insert	不支持	0.36 s
40万	{写入}INSERT INTO employees select * from employees;	3.25 s	0.32 s
80万	{写入}INSERT INTO employees select * from employees;	6.51 s	0.84 s
160万	{写入}INSERT INTO employees select * from employees;	12.93 s	1.62 s
320万	{写入}INSERT INTO employees select * from employees;	28.61 s	3.58 s
640万	{写入}INSERT INTO employees select * from employees;	56.48 s	7.28 s
1280万	{写入}INSERT INTO employees select * from employees;	115.30 s	15.64 s
2760万	{写入}INSERT INTO employees select * from employees;	278.62 s	25.05 s
5520万	{写入}INSERT INTO employees select * from employees;	673.40 s	64.73 s
200万	{普通查询}KV查询200万次.PS: 进程模型,建议实际应用时使用连接池，总连接控制在1000以内绝佳,未来支持内置线程池,几万连接完全没问题.	70.23 s	9.52 s
2560万	{复杂查询}select count(*) from employees;	6.15 s	0.60 s
2560万	{复杂查询}select count(distinct id) from employees;	16.67 s	7.85 s
2560万	{复杂查询}select count(*) from (select id from employees group by id) t;	15.52 s	2.98 s
1.1亿	{复杂查询}select count(*) from employees;	28 s	7.29 s
1.1亿	{复杂查询}select count(distinct id) from employees;	77.73 s	39.78 s
1.1亿	{复杂查询}select count(*) from (select id from employees group by id) t;	84.64 s	14.67 s
2760万 多对一JOIN 200万	{JOIN + 运算}select t1.lname,count(*) from employees t1 join employees1 t2 using (id) group by t1.lname order by count(*) desc,lname limit 10;	超过3小时未出结果	8.90 s
1.1亿 多对一JOIN 200万	{JOIN + 运算}select t1.lname,count(*) from employees t1 join employees1 t2 using (id) group by t1.lname order by count(*) desc,lname limit 10;	超过3小时未出结果	33.73 s
1.1亿	{更新}update employees set lname=concat(lname,'new');	1290.34 s	70.45 s
1.1亿	{创建索引}create index idx_employees_1 on employees(id);	229.04 s	70.45 s

通过以上测试, 在大多数场景中, 阿里云RDS PG相比MySQL的综合性能提升了1个数量级, PG+MySQL结合使用可以大幅降低企业成本. 疫情无情PG有情, 别裁员了, 建立多元化的技术栈, 强化企业IT能力更重要.

更多应用场景和使用方法请参考回顾视频, 包括如何将mysql数据同步到pg(dts):

《阿里云 RDS PostgreSQL+MySQL 联合解决方案课程 - 汇总视频、课件》

课程内容: 

· 2019.12.30 19:30 RDS PG产品概览，如何与mysql结合使用

· 2019.12.31 19:30 如何连接PG，GUI(pgadmin, navicat, dms)，cli的使用

· 2020.1.3 19:30 如何压测PG数据库、如何瞬间构造海量测试数据

· 2020.1.6 19:30 mysql与pg类型、语法、函数等对应关系

· 2020.1.7 19:30 如何将mysql数据同步到pg（dts）

· 2020.1.8 19:30 PG外部表妙用 - mysql_fdw, oss_fdw（直接读写mysql、冷热分离）

· 2020.1.9 19:30 PG应用场景介绍 - 并行计算，实时分析

· 2020.1.10 19:30 PG应用场景介绍 - GIS

· 2020.1.13 19:30 PG应用场景介绍 - 用户画像、实时营销系统

· 2020.1.14 19:30 PG应用场景介绍 - 多维搜索

· 2020.1.15 19:30 PG应用场景介绍 - 向量计算、图像搜索

· 2020.1.16 19:30 PG应用场景介绍 - 全文检索、模糊查询

· 2020.1.17 19:30 pg 数据分析语法介绍

· 2020.1.18 19:30 pg 更多功能了解：扩展语法、索引、类型、存储过程与函数。如何加入PG技术社群

阿里云PG免费试用活动进行中, 请钉钉扫码加入咨询:

文中的相关链接:

1、《阿里云 RDS PostgreSQL+MySQL 联合解决方案课程 - 汇总视频、课件》

https://github.com/digoal/blog/blob/master/202001/20200118_02.md

2、https://github.com/digoal/blog/blob/master/202002/20200206_02.md

https://github.com/digoal/blog/blob/master/202002/20200206_02.md

3、《PG buildin pool(内置连接池)版本 原理与测试》

https://github.com/digoal/blog/blob/master/201805/20180521_03.md

PostgreSQL中文社区欢迎广大技术人员投稿

投稿邮箱：[e-mail]