Go High Concurrency Flash Sale

Project Structure

1. model
2. repositories
3. services
4. controllers
5. views

Sample Page:

Data Flow

Improvements

Object-Relationl Mapping

• Built mapping between relational databases and objects using GORM.

Bidirectional Encryption

• Use Cookie instead of Session to reduce request consumption under heavy traffic.
• Encrypt Cookie with Advanced Encryption Standard(AES-128).

Distributed System

• Server Load Balancer(SLB) - Consistent Hashing
• Add virtual server node on the hash ring to ensure fairness.
• Use Binary Search to find the IP address of the corresponding server according to the hash value of key to realize SLB.

RabbitMQ

• Use RabbitMQ instead of Redis to improve the Queries Per Second(QPS) of each product.

Traffic Control

1. Front-end

• Users must log in before shopping.
• Users can only click the purchase button once every 10 seconds.
  

  

• Captcha: verification is required when click buy-now-button and refresh captcha frequently.
  

  

  
  If verification failed, users cannot buy the product and has to re-try captcha.
  

  

  
  If verification succeed and meet other flash sale restriction, it will show
  

  

2. Back-end

• Server responds to requests every 10 seconds.
• Add blacklist.
• Add Token Bucket:
  "/check" : time/rate, token is generated every millisecond and token bucket size is 1000.
  "/product" : middleware, token is generated every millisecond and token bucket size is 10000.
• Limited access time: Access to the sale page not within a specific time period will be denied.
  

  

Test

Servers

1. Server Zone: Malaysia (Kuala Lumpur)

2. Server Image: CentOS 7.6 64 bit, 1vCPU(s) 512 MB memory

3. Maximum Bandwidth: 100Mbps

4. IP Addresses

   Server Name	Intranet IP	Public IP
   Validate1 + SLB	172.26.194.42	47.250.49.25
   Validate2 + SLB	172.26.194.41	47.250.147.115
   Stress Test	172.17.169.89	47.250.131.235
   Quantity Control	172.26.194.43	47.250.144.234
   RabbitMQ Simple	172.17.169.88	47.254.251.25

5. Cross Compile

GOOS=linux GOARCH=amd64 go build xxx.go

WRK

Set product number 10000, WRK result:

RabbitMQ queue information:

The 10,000 flash sale products were sold within 10 seconds. Quantity control is functioning properly. The bottleneck of running is the publishing rate of RabbitMQ. When the product number is large, we can set the Rabbitmq queue for every 100 requests.

Environment

Go 1.18.1 arm64
Iris 11.1.1
Gorm 1.25.0
Mysql 8.0.31
RabbitMQ 3.11.11