Briefly describe the implementation principle of redis cluster

1.Redis Cluster working principle

（ Multimaster , The contents stored in each master node are different , Each master node has its own slave node , And MySQL（ Every node is the same ） Dissimilarity ）

At the sentry sentinel In mechanism , Can solve redis High availability problem , When master After failure, it can automatically change slave Upgrade to master, So as to guarantee redis The normal use of the service , But it can't be solved redis The bottleneck of single machine write , Stand alone redis Write performance is limited by the memory size of a single machine 、 Concurrent number 、 Network card speed and other factors .

In order to solve the bottleneck of single machine performance , Improve Redis performance , You can use distributed cluster solutions

1.1 In the early Redis Distributed cluster deployment scheme ：

1. Client partition ： It's up to the client program to decide key Write allocated and written redis node, But the client needs to handle the write allocation itself 、 High availability management and fail over, etc

2. Agency scheme ： Based on three party software implementation redis proxy, The proxy layer where clients connect first , Implemented by the agent layer key Write allocation for , It is relatively simple for the client , However, it is relatively troublesome to increase or decrease the number of cluster management nodes , And the agent itself is a single point and performance bottleneck .
   redis 3.0 After the release, we introduced the non central architecture redis cluster Mechanism , In the centreless redis In the cluster , Each node holds the current node data and the entire cluster state , Each node is connected to all other nodes

1.2Redis Cluster The characteristics are as follows

1. all Redis Node usage (PING Mechanism ) interconnection

2. Whether a node in the cluster fails , More than half of the nodes in the whole cluster fail to monitor （ Equivalent to a sentry ）, It's a real failure

3. Client does not need proxy You can connect directly to redis, The application needs to be configured with all redis The server IP, Direct sum redis Server communication .

4. redis cluster Put all the redis node Average maps to 0-16383 Slots (slot) On , Read and write to the specified redisnode Operation on top , So how many redis node amount to redis How many times has concurrency been extended , Every redis node To undertake 16384/N Slots

5. Redis cluster Pre allocation 16384 individual (slot) Slot position , When need is in redis Write a key -value When , Will use CRC16(key) mod 16384 The value after , It was decided that key Which slot of the value is written to determine which one to write Redis Node , So as to effectively solve the single machine bottleneck .

2.Redis cluster Basic framework

Suppose the three primary nodes are ：A, B, C Three nodes , Using the Hashimoto trough (hash slot) How to allocate 16384 individual slot If so, their three nodes undertake slot The interval can be ：

 node A Cover  0－5460
 node B Cover  5461－10922
 node C Cover  10923－16383

The client sends a request , Will be able to key Conduct hash（crc16 Algorithm ） operation , The resulting array pair 16384 Take the mold , Remainder is 0-16383 Number between , Three nodes , Just allocate , Each node is assigned a node number , Then store the data according to the node number .

3.Redis cluster Master slave architecture

Redis cluster Although the architecture solves the problem of concurrency , But it introduces a new problem , Every Redis master How to solve the problem of high availability ？ That's for everyone master All nodes implement master-slave replication , So as to achieve redis High availability

4. summary

redis cluster Each host has its own ip, The client can communicate directly with redis To communicate , There is no need for an agent , Each host communicates with each other , Judge the viability of the other party and , And switch between master and slave freely , Under normal operation , Multiple master nodes can be used for visitors key -value Make allocation records ,（ Through slot allocation , Make a difference key -value Write to different master nodes ）, Thus, high concurrency can be achieved .
The following figure shows client access redis The process , The core is redirection （ There is no need for an agent ）.