Module 9 (E-commerce spike system)

Business background

• You choose the best-selling and well-received products on major e-commerce platforms for sale, each category does not exceed 20 products, and currently has 10 categories;

• This 6.18 seckill selected 1000 power banks and 10 iPhone 12s as seckill commodities;

• normal daily activitiesAbout 1 million users;

• The boss asks for foolproof

Technical background

• The technical team is mainly Java and has already implemented the microservice architecture

• The main channel is self-serviceFor some apps (including iOS and Android) and WeChat mini-programs, in order to promote the conversion of users into app users, only the app can be downloaded to participate in the seckill activity;

• Currently only a single computer room

Performance analysis

• The normal daily activity is 1 million, and the number of people is assumed to be the usual number during the spike2 to 3 times the number of users, then the architecture design needs to be designed at the millions level, and the number of users is expected to be 3 million.

• Users need to advance in advanceLogin, assuming all logins within 30 minutes before the seckill, then QPS=3 million/180=1666/s

• Seckill starts, assuming the seckill requestIt is completed within 30 seconds, and each user clicks and requests 5 times on average, then TPS=3 million*5/30=500,000/s

Storage architecture design

Data analysis:

• Inventory data, frequent read and write, corresponding to spike scenarios, when a single database cannot support write requests, it can be shared by inventory splittingWrite pressure

• Commodity information is relatively fixed, read more and write less, you can use read and write separation to improve performance

• Order and payment data, because the number of products in the seckill is small, it will not increase the number of orders, so do not particularly consider it

• Commodity information, orders, bill payment, MySQL master-slave architecture can be used first

• The dynamic and static resources of the product details page are separated, and the static resources are accelerated by introducing CDN storage

• User information and inventory information are cached in Redis, the TPS of a single redis machine is 50,000 to 100,000, and there are high concurrent writes during the spike, the TPS is 500,000/s, and the multi-machine Cluster architecture is adopted

Computing Architecture Design

Load Balancing Design

cache design

• adoptedapp cache+web container cache+distributed cache three-level cache

• APP cache+web container cache, which can cache most static resources,Reduce back-end pressure

• Distributed cache, cache inventory, user, product information

High Availability Design

For the seckill request, adopt the current limiting strategy:

• App access terminal current limiting: reduce the access request through the clicker to prevent users from making high-frequency requests

• Gateway entry current limit: filter by business rules, check inventory

• Seckill task queue:The leaky bucket is used to limit the current, and the last legitimate seckill request is placed in the current-limiting bucket queue, and the queue is discarded when the queue is full.

The current situation is that there is only one computer room, and the bossIt is required to ensure that the spike is foolproof. You can purchase cloud services first and upload data to off-site cloud services.If you don’t purchase cloud services, you can build dual-active in the same city, and you don’t need to live in different places. If the distance is too far, it will affect the delay of spikes.

YesExtended Architecture Design (Microservice Split)

原网站

版权声明
本文为[InfoQ]所创，转载请带上原文链接，感谢
https://yzsam.com/2022/219/202208071728161989.html