Architecture evolution of high-performance servers -- Suggestions

Catalog

Write in the front

The way of architecture evolution

summary

---

Write in the front

For back-end development , High concurrency is an old topic , Recently, the author encountered a very thorny problem , It's about high concurrency , At present, it is in support of 30W Terminals , But it will continue to grow in the later period , When I hear to support 500W When , I feel like I will have a bald head soon , But don't panic , There is always a way , I checked various materials and consulted several big men , Solved the problem step by step , Next, I will share how to build a high-performance , Highly concurrent servers .

The way of architecture evolution

1： The simplest --- Direct access tomcat Single application , Basically no problem

2： in consideration of tomcat Handle task Upper limit （ Plus tuning ） I still can't bear it , use nginx Load balancing , There should be no problem compared with this

3： Find this time DB The pressure will be greater , So I made a separation of reading and writing , No problem

4： Even if it is read-write separation , But now a large number of streams are generally queries ,DB There is still a bottleneck , You can consider DB colony （ at present mysql Has supported ）, Skip here , Caching is considered in this article （ Now the projects are basically distributed , Then consider distributed caching ）, So the architecture diagram becomes a dead end

5：4 It just basically solves the problem of reading , But what if it's written ？ I'm not going straight to DB Write , Every time I write ？ It doesn't exist well , Consider using caching , When the data is accumulated to a certain amount , Write the data once again DB, Or is it ES（ After my test ,ES Write performance is really low , But reading is really powerful , Basically, it can be counted G Data in 100ms Back inside .）, Use here JVM The cache of , You may need a global clock . Then the architecture diagram looks like this

6： Think it's over here ？ As a person who is about to have a bald head , We must not stop here , So consider whether you can not let the request return results without passing through the background server ？ So there was a great roar OpenResty（nginx Enhanced version of , The original author is Russian , The Chinese will strengthen and upgrade it , Add a series of plug-ins , And open source , at present JD Use this to cache product details ）, Currently used in the project OpenResty Of Cache( operation nginx The cache of , Expiration time can be set , Cache size , Strategy and so on ),mysql（ Can be operated directly mysql）,redis（ Can operate redis） modular , Cooperate with Lua Script together , The result is really fragrant , Then the architecture diagram evolves into the following ghost , It looks a little complicated anyway .

 

summary

There are many high concurrency problems in the project , Many solutions can be considered , For example, a single body cannot , I just distributed + colony , If not, consider caching , Caching can be divided into two types , Server cache and web Layer cache , Then you can optimize the code , Unnecessary synchronization can be changed to asynchronism , Use thread pool , The parameters of the thread pool can be tuned , Reference resources https://blog.csdn.net/weixin_45191798/article/details/100153524 , When writing for a large amount of data traffic , We can take into account the so-called tens of millions of KAFKA To cut the flow peak , Finally, it's really no good , With a bald head , Growing cabbages and raising pigs at home is also an option （ Pork is still more expensive , The way to get rich ）, like JAVA And architecture students can discuss , AC duck .