Redis - understand the master-slave replication mechanism

brief introduction

Redis Master-slave replication of , In fact, it is a Redis Server data , Copy to other Redis Server for . The former is called the master node of the server （master/leader）, The latter is the slave node （slave/follower）, Data can only be copied from master node to slave node ,Master Write as master ,slave Mainly reading . In fact, each server is a master node , A master node can have zero or more slave nodes , And each slave node can only have one master node . Master slave copy , Read / write separation , In most cases , Many people browse the web more for reading , This is a pattern often used in architecture , To relieve reading pressure ：

The role of master-slave replication

1、 data redundancy ： Master-slave replication realizes hot backup of data , It's a way of data redundancy other than persistence .
2、 Fault recovery ： When the primary node fails , The slave node can temporarily replace the master node to provide services , It's a way of service redundancy
3、 Load balancing ： On the basis of master-slave replication , Cooperate with the separation of reading and writing , Write by the master node , Read from node , Share the load on the server ; Especially in the context of more reading and less writing , Load sharing through multiple slave nodes , Increase concurrency .
4、 High availability cornerstone ： Master slave replication is also the foundation for sentinels and clusters to implement .
Be careful ：
take Redis Apply to the project , It will not be built with only one server , Because if it's a single Redis, Then a single server will complete the read and write operations from the client alone , Larger load , And there is only one Redis The server , It's too easy to hang up , It's troublesome if it goes down accidentally
Single server , There is an upper limit to the memory capacity , No matter this one Redis No matter how large the memory of the server , It is also impossible to completely use it to store memory , And single Redis Generally, the memory consumption of the server will not exceed 20 individual G.

Why cluster

1、 It is difficult for a single server to load a large number of requests .
2、 Single server failure rate is high , The probability of system collapse is high .
3、 The memory capacity of a single server is limited .

Cluster environment construction

View the current library information ：info replication

Build configuration , Copy out three configuration files , modify 4 Messages
1、 Build master-slave replication , Different servers need different port numbers
2、pidfile File name for
3、logfile Log file name
4、Redis Default RDB Persistence , therefore dump.rdb The name of also needs to be modified

edit vim redis79.conf The configuration file , Use as a host ：


edit vim redis80.conf The configuration file , Used as a slave ：




edit vim redis81.conf The configuration file , Used as a slave ：



Start after modification , If the process exists , It means that the construction is completed .

One master and two subordinates

be-all Redis The server is a master node by default , When configuring , Generally, you only need to configure the slave .
To put it bluntly, it means to recognize the boss ！ A master （79） Two from （80,81）
Use SLAVEOF host port You can configure the host for the slave .


Operate in the same way 6381 Also recognize the boss , Then the host will display ：

Be careful ：
actually , So the master-slave configuration in the command line , It's just temporary , If you really want to achieve permanent master-slave configuration , You need to configure... In the configuration file , The location of the configuration file is as follows , Remove the comment , Then change to the corresponding configuration ：

Usage rule

The slave can only read , Can't write , The host is readable and writable, but mostly used for writing .

The setting value of the host , It will also be obtained by two slaves .

But as a slave, it cannot be written by default .

1、 If one master and two slaves are realized , The host accidentally hung up , Then the remaining slaves stand by before the master arrives , At this time, the main engine is down , It means that the write operation is lost . And after the host goes online again , Still maintain the original identity , The slave can still get the information written by the host . How can we create a new host when the host is down ？ Mode one 、 The slave executes the command manually slaveof no one, After this execution, it will be independent from the opportunity and become a host . Mode two 、 Use sentry mode （ Automatic election ）

2、 At this time, if you hang up from the computer , It means that the slave has been disconnected , When waiting to go online again , Will not connect to the original host , It becomes the host itself , But the original data still exists , If it reconnects , The newly added data in the host will also be shared to .

The principle of replication

If one Slave Specified a master Be the boss , that Master The node will send a message called sync The synchronization command of , In the use of the SLAVEOF After the command ,Master received , Will start the background save process , Start collecting commands for modifying datasets , After the execution of the save process , Data files will be sent to Slave, Achieve a full synchronization .
There are two ways to copy ：
1、 Copy in full ,Slave The document is receiving Master After the data file sent , It will be saved and loaded into memory
2、 Incremental replication ,Master and Slave When synchronization has been achieved , If Master Continue modifying dataset , Will collect commands again , Send it to Slave
Be careful ：
As long as it's reconnecting Master Server for , Full replication will be performed automatically after connection . We can see all the data of the host from the slave ！

Link model

In fact, there are two models for this kind of replication , The above is a model （ Master slave copy model ）, There is also a link model similar to , Linear relation ：

Link model ：
In the actual application scenario, there can be multiple nodes , Such nodes have Master and Slave The double identity of （ But in fact, it is still a Slave node , It's just a write operation Slave node ）, Can share Redis The pressure of writing , And in real Master After the node hangs up , These people with dual identities can work on top , It effectively solves the unwritten problem of one master and two slaves .