How to design a seckill system?

Seckill system mainly has three characteristics High performance 、 High concurrency 、 High availability .

Starting from the process of a second kill , Consider the three characteristics of seckill system , Then you can design a seckill system .

1. Seckill page access

Optimization plan ：

1. Dynamic and static separation . Deploy the static resources of the page to Nginx perhaps CDN, This can speed up the seckill page acquisition .
2. Static resource merge acquisition . By merging multiple requests into a single request , Get multiple static resources at a time , This can speed up the seckill page acquisition .
3. service degradation . The seckill page is degraded , List the recommended products 、 Downgrade comments, etc , Show less or no . You need to log in to view the seckill page , Return to the login interface directly for users who are not logged in .
4. Service monitoring . Monitor the flow , Use the token bucket algorithm and other flow limiting algorithms to control the flow . Fuse some tasks if necessary .
5. Page data cache . Cache page data to Redis in , Reduce database operations .
6. Second kill connection adding salt . send URL Dynamic , Illegal user operations can be reduced .

2. Order goods

Optimization plan ：

1. front end / Back end current limiting . front end / Client anti shake . Limit the number of orders placed within the time interval .
2. Anti robot brushing . Add the step of filling in the verification code for the order placing operation , Such as ：55+44=？、“ Hello ” Lowercase Pinyin of 、 Select all the aircraft and other questions , Filter out illegal requests .
3. Withholding inventory when placing an order . When designing database tables, you need to set the lock inventory field . During the second kill , Inventory reduction will be Redis Using distributed locks in . Other subsequent operations can use RabbitMQ To operate .
4. Withholding inventory when placing an order （ Stock preheating ） You can add delay queues . Forward the timeout goods to the dead letter route , And then operate .
5. Asynchronous operation can be performed when placing an order , Multiple threads can be used for operations such as double price verification .

3. payment

Optimization plan ：

1. Divide payments into a separate system , Only the corresponding payment interface is opened . Because the payment system is financial sensitive , Therefore, the high availability of the payment system should be guaranteed .
2. Rollback mechanism . Distributed transactions are recommended , For the payment business TCC Business , Because the payment system is financial sensitive .

---

therefore , Seckill system usually introduces MQ、Redis、MySQL、Nginx Such as middleware , Each middleware needs to be High performance 、 High concurrency 、 High availability Analysis of .

MQ

Optimization plan ：

1. Cluster deployment .MQ Systems are generally deployed in clusters , Deploy the image cluster , Can improve the availability of the system .
2. Turn on persistence . Yes MQ The information in the system turns on persistence , Brush it into the hard disk , Prevent downtime .
3. Turn off automatic consumption ACK, Manual operation is required ACK. Prevent abnormal information consumption .

Redis

Optimization plan ：

1. Redis Read and write separation ,Master Node to write , Read from other nodes .
2. Redis Deploy sentinels , After a node goes down, it can be replaced by a machine .
3. Redis Deploy the partitioned cluster , Distribute the requests to each server Redis On the machine .
4. Enable persistent logging .AOF and RDB Adjust according to business conditions .
5. A system can have multiple Redis colony , For example, page data and product order Redis You can use multiple clusters Redis.

MySQL

Optimization plan ：

1. Index based on business . unique index 、 General index 、 Joint index, etc .
2. See if the business is optimized , Reduce the return operation .
3. Sub database and sub table .MySQL Cluster deployment should be done , A single Redis Usually only 2000QPS about .
   1. sub-treasury . Use MyCat perhaps ShardingSphere And so on , Assign the operation to the corresponding machine by algorithm .
   2. table . There are two types of sub tables: vertical and horizontal . Vertical partitioning is to divide some fields into other tables . Horizontal partitioning is to divide data horizontally into different tables in the same database , Avoid too large data on a table .
   3. Generally speaking , Suggestion points 32 Databases , Each Treasury point 32 A watch , This can fully meet the needs of most enterprises .
4. MySQL The bottleneck is disk IO, Solid state drives can be replaced .

Nginx

Optimization plan ：

1. Dynamic and static separation . Deploy static resources to Nginx in , No need to query in other middleware .
2. Nginx The current limiting operation can be started . Both token bucket and exposed copper algorithm support .
3. Nginx Turn on load balancing , Send the service request to different servers , Reduce the pressure on a single server .

---

Except for the ones listed above , There are many optimization operations .

Hot data separation

The systems used by hot goods and ordinary goods can be isolated , In this way, even if the secsha system goes down , There won't be any problem in ordering ordinary goods .

1. Second kill products are put into the hot data system .
2. Live broadcast products can also be put into the hot data system .
3. Traffic monitoring . Products with more orders can be put into the hot data system .
4. Merchant Report . Businesses can report commodities that may sell more in the future , Put it in the hot data system .
5. Data analysis . Analyze past data , Some products that may sell more in the future , Put it in the hot data system .

---

performance optimization

Finally, you can optimize the performance of the machine .

1. Replace CPU
2. Replace memory
3. Replace a faster hard drive
4. to update Linux system kernel
5. Update the stable version of the software system
6. close Linux Some useless services above

---

Seckill system mainly has three characteristics High performance 、 High concurrency 、 High availability . Just think about these three points , Then we will gradually come up with a second kill system .