Structure training camp module II operation

Wechat circle of friends has high performance and complexity

Business indicators

【 Business complexity 】

The business complexity of the circle of friends is low , Only content publishing 、 see 、 Comments and likes .

【 Quality complexity 】

There are many users in the circle of friends , According to zhangxiaolong in “2021 WeChat public class PRO” Speech in ：

Every day 10.9 100 million users open wechat ,3.3 Hundreds of millions of users make video calls ,7.8 100 million users enter the circle of friends ,1.2 100 million users publish circle of friends ( One of the photos 6.7 One hundred million , Short video 1 Billion bars ),3.6 Billion users read official account articles ,4 Hundreds of millions of users use small programs .

You know , Wechat circle of friends UV about 7.8 Billion , Considering that most people are concentrated in 06:00-24:00 Time period for access , Then we get the average qps by 1.2w, Then consider some time periods, such as eating 、 The use of a circle of friends on the way to and from work will be relatively concentrated , Assume that the peak is of the average 5 times , So the rush hour qps about 6w.

Suppose that half of the people who browse the circle of friends will like , The like function tps about 3w.

Suppose the people who like the circle of friends have 1/3 Will comment , Commented tps about 1w.

Suppose the number of people who send a circle of friends is equal to the number of people who comment , It's published tps about 1w.

therefore , For the architectural design of the circle of friends , The key lies in ：

• qps/tps Sustained high

• There are a lot of users

therefore , Need to ensure high availability 、 High performance .

Here we only discuss its high-performance architecture design .

High performance is divided into computing high performance and storage high performance .

Computing high performance is divided into cluster high performance and single machine high performance . Cluster high performance is generally achieved through business splitting + The load balancing of each split service cluster is realized ,( Similar to sharding in middleware cluster + Replica mechanism ); The high performance of a single machine is generally realized by various optimization methods , Caching 、 asynchronous 、 Line up, etc .

Storage high performance is also divided into cluster high performance and single machine high performance . The high performance of the cluster is generally achieved by dividing relational data into databases and tables ( Such as ShardingJdbc)、NoSQL( Such as HBase/ES) Or distributed storage (FastDFS/ Cloud storage ( Such as OSS)) Realization . The high performance of a single machine is generally optimized SQL To achieve .

Its overall architecture can be as follows ：

Hair circle of friends 、 Browse the circle of friends 、 The like circle of friends and comment circle of friends can be used as a separate service respectively , Each service deploys multiple replicas to achieve the requested load balancing and high availability .

Recent pictures 、 Video is suitable for caching in CDN.

Storage tier content publishing 、 There is a separate library for compliments and comments . You can use a relational database or NoSQL： A relational database is like MySQL, It can be used as dual active and multiple standby 、 Read / write separation 、 Separate database and table and separate hot and cold data ;NoSQL Such as HBase colony .

Which friends each user has can be maintained in Redis in . Writing data ( Hair circle of friends 、 Like the circle of friends 、 Comment on the circle of friends ) You can write first Redis Cluster and MQ Message queue , Then asynchronously fetch data from the queue MySQL Persistence , In this way, you can read directly when browsing the circle of friends Redis.

picture 、 Video storage FastDFS Or cloud storage ( Such as OSS), Recent pictures / The video can be pushed to CDN.

It may also need to be supplemented by fusing 、 Current limiting 、 Degradation and other service governance measures , Such as Sentinel.