Module 9 - Designing an e-commerce seckill system

Requirements

Business performance analysis

Business Scenario

• To be loaded into the Redis cache in advance of the seckill product

• Before the seckill10 minutes user centralized login App

• Users frequently refresh the corresponding product page before the spike

• Seckill requests are usually concentrated within 10 seconds after the event starts

• Successful orders will be paid within 5 minutes
  div>

Performance Analysis

• The normal daily activity is 1 million. It is assumed that due to the warm-up of the 618 activity, the daily activity doubled to 2 million, of which30% (600,000) will participate in the spike activity.

• 600,000 login requests 10 minutes before the spike, QPS = 600,000 / 10 / 60 = 1000

• There are 3 types of return results for the seckill request: success, sold out, and failure.

• In the case of success or sold out, users cannot participate in the flash sale again.

• Failure is due to the fact that the system fails to process the request under high concurrency, and the user may click the seckill to try again.Assume on average two retries per user.

• According to this estimate, TPS = 600,000 * 2 / 10 = 120,000

StorageArchitecture

• Seckill activities are mainly handled by Redis cluster, including product preview, real-time inventory refresh, etc.;

• MySQL is used for data persistence.

Load Balancing

• Because the company is still in the start-up stage, F5 is not considered.Adopt three-level load balancing architecture

Cache Design

High Availability Architecture

• The normal daily life of the system is at the level of one million, and it is not considered to live in different places for the time being

• The boss asked to be foolproof, use the same city dual center