About redis master-slave replication

Master slave copy

 Master slave copy , It means to put one Redis Server data , Copy to other Redis The server . The former is called the main node (master), The latter is called the slave node (slave); Data replication is one-way , From master to slave only .

Master slave replication is divided into three stages

Connect from node , Request full replication ,

1. Master node execution bgsave The instruction generates the current data rdb snapshot , Send data to the slave node , And create a buffer , This buffer records all write commands from the current time

2. Clear your original data from the node , receive rdb And load , Synchronize the database to the master node to execute bgsave The state of

3. The master node sends the instructions of the buffer mentioned above to the slave node , Execute these instructions from the node , So that the database is up to date

Incremental master-slave replication

Both master and slave nodes maintain an offset offset, This offset Represents the number of bytes sent by the master node to the slave node . When the master-slave node offset In unison , We can think that the two databases are in the same state .

The slave node is disconnected for some reason during the connection process , Restore connection , At this point, incremental replication is requested .

The master node maintains a replication backlog buffer , He is a first in, first out queue , Each time a write command is sent to the slave node, the command will also be sent to the buffer , The buffer also stores offset.

After the connection is restored from the node , Go to the copy backlog buffer to view your own offset Whether in buffer in , If in , Just start from the recording position , Do incremental replication . If not here buffer in , Then perform a full copy

How to handle expired data

There are two types of expired data processing

• Lazy deletion ： The server will not actively delete , When the client query data is , If the data is out of date , Then the server determines whether to delete
• Delete periodically ： Delete expired data regularly , But operate release frequently , Loading memory will affect server performance , Therefore, it is necessary to limit the frequency and execution time of deletion .
  Do not delete data from the node , Therefore, the expired data will not be deleted when the slave node reads the expired data , Only null will be returned .

Sentinel mode

brief introduction

Generalize , The sentry is divided into three questions ： monitor , Elector , notice .
monitor ： Monitor the health of the node , And make timely adjustments
Elector ： When the master node is down , The sentinel chooses the master
notice ： Notify the client of the current master node address

The sentry cluster

In order to prevent network fluctuations from causing a single sentinel to misjudge , Therefore, the concept of sentinel cluster mode is introduced

How did the Sentinels find each other ？

Through the master node , Subscribe to news , And find each other , Resume the connection between sentinels

How to know from the library ？

To be improved

Judge the offline rule

Divided into two

1. Subjective offline ： Just need a single sentry to think offline , This node is considered offline , Used to judge from the offline of a node .
2. Objective offline ： The sentinel cluster needs to vote to determine whether the node is offline , When the number of tickets exceeds the set value , To judge the offline , It is used to judge the downline of the master node .

Elector

It's based on

1. Select the node with high server configuration as the primary node
2. Select the node with the longest running time , The longer it takes to run , The more stable the node is
3. Select according to the configuration file

The process

• The selected node is the master node
• Other nodes connect to this node
• The original master node becomes a slave node