[high concurrency] how to realize distributed flow restriction under 100 million level traffic? You must master these theories!!

Hello everyone , I'm glacier ~~

In Internet applications , High concurrency systems face a major challenge , That is a large number of streams and high concurrent access , such as ： Tmall's double eleven 、 JD.COM 618、 seckill 、 Panic buying, promotion, etc , These are typical high traffic and high concurrency scenarios . About second kill , Friends can see another article of mine 《

【 High concurrency 】 Decryption of high concurrency seckill system architecture , Not every second kill is a second kill ！

》 Be careful ： Due to the length of the original , So it is split into ：

theory 、 Algorithm 、 actual combat （HTTP Interface practice + Distributed current limiting practice ）

  The three most .

High concurrency system current limiting

Huge traffic in a short time , How can we make the system handle high concurrency while ensuring the stability of its own system ？ Someone will say , Just add machines , Because my system is distributed , So you can just add machines to solve the problem . however , What if you can't solve this problem by adding more machines ？ And in this case, we can't add machines without limit , The hardware resources of the server are always limited , With limited resources , We need to deal with this kind of access with large traffic and high concurrency , We have to take some other measures to protect our back-end service system , such as ： cache 、 asynchronous 、 Downgrade 、 Current limiting 、 Static, etc .

here , Let's talk about how to achieve current limiting first .

What is current restriction ？

In high concurrency systems , Current limiting usually refers to ： Speed limit high concurrent access or requests, or speed limit requests within a period of time to protect our system , Once the system's speed limit is reached （ For example, the system limits the speed of requests ）, You can handle these requests in the following way .

• Denial of service （ Friendly tips or jump to the error page ）.

• To line up or wait for （ For example, seckill system ）.

• service degradation （ Return the default profile data ）.

Actually , It's a speed limit on requests , such as 10r/s, That is, only 10 A request , This limits the speed of requests . In a sense , Current limiting , In fact, it is to limit the quantity at a certain frequency .

Current limiting is generally used to control the rate of system service requests , such as ： Tmall double 11's current limiting , JD.COM 618 Current limiting of ,12306 Tickets, etc .

What are the use scenarios of current limiting ？

here , Let's take an example , Suppose you build a mall system , On a holiday , I suddenly found that the number of interface requests for order submission has increased by nearly 50 times , It wasn't long before the interface that submitted the order timed out and threw an exception , It's almost unavailable . and , Because the order interface is not available due to timeout , It also caused other services of the system to fail .

How should we deal with this kind of big traffic scenario ？ A typical solution is current limiting . Yes, of course , In addition to current limiting , There are other solutions , Our article is mainly about current limiting .

• Second kill of scarce resources 、 Rush purchase ;

• High concurrent read and write operations on the database , Like submitting an order , Instantly insert a large amount of data into the database ;

Current limiting is a powerful tool to deal with high concurrency , With current limiting, you don't have to worry about the instant peak traffic crushing system services or service avalanche , In the end, it will damage the service rather than not serve it .

The same thing to note when using current limiting is ： Current limiting should be evaluated , The test is good , Otherwise, normal access will be limited .

Okay , That's all for today , I'm glacier , See you next time ~~