Record a PG master-slave setup and data synchronization performance test process

Catalog

background

PG install

Master slave configuration

Main library configuration

Configuration from library

Verify master-slave setup  

Data synchronization

---

background

With the development of relational databases in the financial field ORACLE turn , More and more customer sites begin to use domestic cloud databases or open source databases . Domestic databases include OceanBase, Reach a dream , And our eternal LightDB( Super easy to use , Extreme push ), They all have strong performance and customer-oriented operation and maintenance , There will be no more talk here . The open source used in our products RDBMS At present MySQL Mainly , Its advantages are many , Small volume 、 Fast 、 Low total cost of ownership 、 Open source code and so on , This is also the product most used by customers . however MySQL One drawback is that master-slave synchronization is extremely slow under large transactions , A certain site in Beijing is synchronizing some extra large data tables , Such as fund evaluation data ( Tens of millions ) It will cause master-slave synchronization stagnation , It has seriously affected the use of the product . Of course, the subsequent process is emptied instead of deleted , Reducing the amount of each submission is also a reasonable and effective solution to this problem . Because the back will go up PG edition , To prevent this from happening again during data synchronization , The leader asked me to start testing PG Master slave synchronization , It's time to record , It is a little reference for subsequent field applications and new library testing .

PG install

Default environment Linux centos7 Distribution version

• Create directory

# mkdir -p /opt/database/

• Download decompression

It is recommended to download the installation package on the official website ： After selecting the version ( Here 12.2), choice tar.gz Format https://www.postgresql.org/ftp/source/https://www.postgresql.org/ftp/source/

 

Put the installation package in /opt/database/ Under the table of contents
# tar -zxvf postgresql-12.2.tar.gz

•   Installation dependency

# yum install -y bison
# yum install -y flex
# yum install -y readline-devel
# yum install -y zlib-devel

• Compilation and installation  

# ./configure --prefix=/opt/database/postgresql-12.2
# make
# make install

• Create user  

# groupadd postgres                             
# useradd -g postgres postgres    
# chown -R postgres:postgres /opt/database/postgresql-12.2
# mkdir -p /opt/database/postgresql-12.2/data

• Installation initialization

# su postgres 
$ /opt/database/postgresql-12.2/bin/initdb -D /opt/database/postgresql-12.2/data/  

$ /opt/database/postgresql-12.2/bin/pg_ctl -D /opt/database/postgresql-12.2/data/ -l logfile start　 -- Start database
$ /opt/database/postgresql-12.2/bin/pg_ctl -D /opt/database/postgresql-12.2/data/ stop               -- Stop database
$ /opt/database/postgresql-12.2/bin/pg_ctl restart -D /opt/database/postgresql-12.2/data/ -m fast    -- Restart the database

• environment variable

$ su root
# cd /home/postgres
# vim .bash_profile

add to :
        export PGHOME=/opt/database/postgresql-12.2
        export PGDATA=/opt/database/postgresql-12.2/data
        PATH=$PATH:$HOME/bin:$PGHOME/bin

# source .bash_profile

• Boot from boot

# cp /opt/database/postgresql-12.2/contrib/start-scripts/linux /etc/init.d/postgresql
# chmod a+x /etc/init.d/postgresql
# vim /etc/init.d/postgresql

change prefix and PGDATA , See the above for the changes

# chkconfig --add postgresql （ Add boot entry ）
# chkconfig   ( Check whether the addition is accurate )

• Set the password

PostgreSQL A database user will be automatically created after installation , be known as postgres  

$ service postgresql start
$ psql -U postgres  
postgres=#   ALTER USER postgres with encrypted password '${PASSWD}'; （ Password is freely set ）postgres=#   \q （ You can quit ）

notes ： The above steps should be implemented once for both the main database server and the standby database server , That is, install them all PG database

Master slave configuration

Through the above installation configuration , The master and slave servers have been installed PostgreSQL database

Main library configuration

• Create a copy permission account

CREATE ROLE replica login replication encrypted password '${PASSWD}';

• Modify role authorization

# vim /opt/database/postgresql-12.2/date/pg_hba.conf

Add all IP You can visit ：host            all             all               0.0.0.0/0           trust

Add copy from library ：    host    replication     replica         From the machine IP/24         md5

• Modify the configuration file  

# vim /opt/database/postgresql-12.2/date/postgresql.conf

#  Add or modify the following attribute settings 
#  To monitor all IP
listen_addresses = '*'  
#  Open archive 
archive_mode = on 
# Filing order pg Database data Folder  
archive_command = 'test ! -f /opt/database/postgresql-12.2/data/pg_archive/%f && cp %p /opt/database/postgresql-12.2/data/pg_archive/%f' 
# Hot standby mode 
wal_level = replica 
# At most 2 Stream replication connection  
max_wal_senders = 2 
wal_keep_segments = 16 
# Stream copy timeout  
wal_sender_timeout = 60s 
# maximum connection , The slave needs to be greater than or equal to this value  
max_connections = 100

• Restart the database

$ pg_ctl -D /opt/database/postgresql-12.2/data -l /opt/database/postgresql-12.2/data/logfile restart

Configuration from library

The configuration of slave database is simpler than that of master database

1. Out of Service
$ pg_ctl -D /opt/database/postgresql-12.2/data -l /opt/database/postgresql-12.2/data/logfile stop

2. Empty files
$ rm -rf /opt/database/postgresql-12.2/data/*

3.copy Data on the master server
$ pg_basebackup -h Master node IP -p 5432 -U replica -Fp -Xs -Pv -R -D /opt/database/postgresql-12.2/data/

4. edit data It will appear in the directory standby.signal file , Join in standby_mode = 'on'
$ vim /opt/database/postgresql-12.2/data/standby.signal

5. edit  postgresql.conf file , See below for details
$ vim /opt/database/postgresql-12.2/data/postgresql.conf

# Slave information and connected users 
primary_conninfo = 'host= Master node IP port=5432 user=replica password=replica User's password '
# Description: restore to the latest state 
recovery_target_timeline = latest 
# Greater than the primary node , The formal environment should reconsider the size of this value 
max_connections = 120 
# It shows that this machine is not only used for data archiving , It can also be used for data query 
hot_standby = on
# Maximum latency of streaming backup 
max_standby_streaming_delay = 30s 
# The interval between reporting local status to the host 
wal_receiver_status_interval = 10s 
#r An error occurred copying , Feedback to the host 
hot_standby_feedback = on 

  Finally, start the database service

$ pg_ctl -D /opt/database/postgresql-12.2/data -l /opt/database/postgresql-12.2/data/logfile start

Verify master-slave setup  

Connect the main library to run sql

select client_addr,sync_state from pg_stat_replication;

Return the following values to indicate success , You can also create tables in the main database , Insert , Check whether the slave library is synchronized by operations such as emptying

Data synchronization

Now perform data synchronization verification , Look at the configuration of master-slave PG Whether the synchronization of a large amount of data in the database can run quickly , Will there be any other problems ？

What we use here is kettle This ETL Tools to write text data , Data synchronization , Data calculation and other operations

Verified by a large amount of data , The test results are as follows ：

1. truncate these DDL The operation can be master-slave synchronization , Seconds to complete
2. In the process of batch writing , Master slave synchronization always maintains data consistency
3. The batch write rate is too slow ,320w Data time 960 second (3333r/s)
4. There is a problem with the process cleaning mechanism , After each run failure , You need to delete the sql process , Otherwise, it will stop running again （ Lock table ）
5. It's a strong type of , When fields are compared with each other or assigned values, they must be of the same type
6. update use update table1  from table2 Is the fastest performance
7. Refer to other synchronization tools such as datax, It is found that the increase rate is nothing more than removing the index , Read / write separation , Close and archive these operations that the bank will not do on site ,PG There's nothing like MySQL Of rewriteBatchedStatements Force batch submission