When new retail meets serverless

Introduction to Serverless brings more opportunities to the digital transformation of traditional enterprises.

A leading retailer in the super-industry. Its main business covers shopping malls, hypermarkets, general supermarkets, standard supermarkets, boutique supermarkets, convenience stores and unattended smart stores, and other retail formats, involving omni-channel retail, warehousing and logistics, catering, Consumer services, data services, financial services, cross-border trade and other fields. In order to continue to support the rapid and stable development of the business, after the rapid of the cloud, the core business was transformed into a middle-office model with a full serverless architecture. 1611cb70041bff adopts function computing + API gateway + table storage OTS as the computing network storage core, elastically supporting daily , it can easily support the 618/Double 11/Double 12 promotion 1611cb70041c01.

Why traditional enterprises need to pay more attention to serverless

In order to reduce technology R&D costs and improve operation and maintenance efficiency, more and more companies choose to use Serverless as their basic R&D base to develop their business vigorously. The CNCF Serverless research report shows that a large number of domestic developers are migrating the traditional architecture to Serverless. The emergence of Serverless brings more opportunities to the digital transformation of traditional enterprises.

Nowadays, a large number of cutting-edge technical talents are more inclined to work in Internet companies, and traditional enterprises are facing a large number of technical upgrades and the rigid need for restructuring technical architecture. Between the talent gap and technical upgrades, there is a demand for cloud-native technology. The emergence of Serverless smoothed out the lack of R&D personnel in budget and operation and maintenance experience. In the case of helping enterprises fight the peak of business, R&D personnel can easily control the processing, which not only greatly reduces the threshold of R&D technology, but also greatly improves R&D efficiency. For developers, online early warning, traffic observation and other tools are all available, the key is to eliminate the burden of operation and maintenance, and effectively provide developers with inclusive technology dividends. For traditional enterprises, Serverless shortens the technological competitiveness distance between Internet companies and traditional enterprises.

From cloud to cloud native

After 2016, with the rapid development of domestic public cloud, the comprehensive cloud access is unstoppable. A well-known large shopping mall gradually migrated the various system modules in its self-built computer room to the public cloud during 2018~2019, and the overall architecture did not change much, so the migration work went smoothly.

Some improvements and shortcomings after the system is fully migrated to the cloud

No longer need to care about the hardware details of the network and operating system

For example, Alibaba Cloud's ECS will perform scheduling and early warning in advance, transfer user data and perform disaster preparedness for multiple copies of data to prevent disk failure.

Quick and easy upgrade

For example, if the user is using a 4-core machine, when he finds that the business is growing rapidly and requires a hardware upgrade, he only needs to make a mirror image. For example, take a disk snapshot at night, reapply for a new machine, and then restore the snapshot to complete the one-click migration. This is a very quick way for users and a better experience for developers.

Machine expansion time is greatly shortened

The above mentioned is the expansion of single machine, such as 4 cores to 8 cores, 16G to 32G of memory. In addition, there are horizontal expansion, such as the API interface of the user transaction system. With the development of the business, the original 2 machines need to be expanded to 8 machines. In this case, the user only needs to apply for the machine, and then mirror It can be extended to different machines.

Difficulties in resource budgeting

It is impossible to estimate the volume that the business can reach when it encounters a big promotion event, so it is impossible to accurately calculate the amount of hardware required.

Horizontal expansion

Horizontal expansion has higher requirements for R&D. For example, whether the data needs to be stateless, and if it is stateless, horizontal expansion will be easier. If it is stateful, the data may need to be cached, which will involve database-related issues, such as data expiration and consistency. If you do not have a thorough understanding of these, it will be more difficult to do horizontal expansion.

Water level monitoring

Many developers have not done well in water level monitoring. If you mix various business systems on one machine, when the water level of the machine is high, you want to quickly troubleshoot problems and perform flow control, splitting, and temporary repairs in time. It seems particularly difficult.

Financial budget difficulties

Similar to resource budget difficulties.

High hardware upgrade costs

To achieve user-insensitive and non-destructive upgrades, it may involve the consistency of the connection processing and the database. If multiple modules need to be upgraded at the same time, also pay attention to the compatibility of data structures.

Single point of failure

Many manufacturers put all the data in a database, if not handled properly, it may cause a single point of failure. This requires data splitting. For rough splitting, you need to pay attention to issues related to transactions and locks, and efficiency will be greatly reduced; for detailed splitting, it will be more difficult to query and sort, which will cause certain troubles to business implementation.

Business challenge

During a big promotion every year, due to the uncontrollable access of online business users and the excessive amount of data, the MySQL single-machine access was broken, causing problems with the storage database, affecting multiple systems, and causing certain losses. Therefore, in the subsequent service transformation process, the database selection is changed from MySQL to Table Storage OTS. The biggest advantage of Table Storage is that users do not need to care about the proportional relationship between the amount of access and the number of machines. As long as the number of visits increases, the background will automatically expand and expand the machine to meet the high concurrency of data reading; when the data concurrency request is reduced at a low peak period, the background will recycle the machine, and users no longer need to care about the number of machines and how to mobilize .

Serverless transformation

In response to the problem of uncontrollable user traffic, the customer introduced Alibaba Cloud’s product "API Gateway", which can control release and traffic control for different channel providers. For example, if the WeChat channel traffic is found to be abnormal, the API gateway can be used to limit the traffic.

In addition, calculation is also a very important issue. After exploration, customers found that Alibaba Cloud Function Computing FC fits their business scenarios very well. For example, regular snap-ups, coupons and other activities have caused a huge burst impact. When it is found that the computing resources are not enough, it is definitely too late to buy a machine, and the function of function computing to expand in time can solve this problem. In addition, its data observation and abnormal alarm functions have also attracted customers.

In March of this year, the authoritative consulting agency Forrester released the FaaS platform evaluation report for the first quarter of 2021. Alibaba Cloud Function Computing stood out by virtue of its advantages in product capabilities, security, strategic vision, and market size, and its product capabilities ranked first in the world. This is also the first time that a Chinese cloud vendor has entered the FaaS leader quadrant.

After intense testing and verification, the technicians found that the excellent performance of function calculation fits well with the highly flexible membership query system of their own business. Starting in July 2019, the customer’s technical team migrated all copies of the original membership data to Table Store in less than 3 months, and fully migrated the APIs of all channel providers to the Alibaba Cloud API gateway. The computing business of the distribution and membership query business has also been fully migrated to Alibaba Cloud Function Computing.

On Double 11 in 2019, function computing is used as the computing module, and table storage is used as the storage module to smoothly help customers get through the big promotions, carry peak traffic, and ensure business flexibility. Businesses that do not use Serverless have some exceptions due to insufficient estimates. It is precisely because of the performance of functional computing in Double 11 that customer and technical people are very excited. After successfully passing the big promotion event, customers will fully use function calculation and table storage in all businesses!

New retailer's overall architecture diagram

• Full serverless architecture: function computing + API gateway + table storage;
• Flexible high availability: millisecond-level elastic expansion, sufficient resource pool water level, high availability across availability zones;
• Agile development free of operation and maintenance: functional minimalist programming can focus on business innovation, without procurement and deployment costs, and provide complete observability such as monitoring and alarming.

In the second half of 2019, Alibaba Cloud Function Computing announced the launch of 2.0, which supports the reservation mode, and comprehensively solves the problem of large cold start delay; launched the single-instance multi-request problem, and fewer instances support heavy IO and high-concurrency request calls; support custom operation At the time, it supports one-click migration of traditional Web architecture servers. With the emergence of 2.0, functional computing has achieved a huge upgrade in business and scale.

After experiencing the test of the offline scenarios in the past, customers have migrated the businesses of various channel providers and their mobile apps, as well as online transactions, timed coupon grabbing, and seckill services from ECS to Function Computing 2.0. After adjusting the single-instance multi-concurrency mode, the peak traffic request that was dozens of times more than usual was successfully passed.

Comparing the above-mentioned "time-flow graph" and "time-delay" two graphs, it can be seen that the sudden increase in burst traffic has very little impact on the delay change caused by users. From the actual user feedback, it does confirm that the user experience is very good. Smooth.

All data and business going to the cloud not only relieves the psychological pressure of R&D personnel, but also reduces the burden of R&D personnel a lot, so that everyone can do things that are more focused on business logic. Function computing allows R&D personnel to not need to manage the infrastructure of servers. As long as they write code and upload, the system will prepare computing resources. It also provides functions such as log query, performance monitoring, and alarms. If it were to follow the previous model and the supermarket was engaged in the Double 11 promotion, the relevant technical team could not sleep, and it would be impossible for anyone to rely on the expansion of the machine to support a large amount of traffic and business. Now that the problem of capacity expansion is left to Alibaba Cloud, the water level is far higher than the original limit of the customer's original reserve capacity.

This year, Serverless ushered in a major upgrade. Function Computing has launched a heavy release of container image acceleration technology, which shortens container startup delay by 50%-80%, and transfers the burden of image optimization that originally belonged to developers to function computing, which further helps developers improve production efficiency and focus on business innovation. This technology originated from the ultra-large scale and highly complex container environment of Alibaba Group. It has deep accumulation of image storage and acceleration technology, and has outstandingly undertaken the strict seckill scenarios of double eleven, double twelve, and Spring Festival in three years. Harsh challenge.

At the same time, the Serverless Application Engine (SAE) has released the Java application startup acceleration function. For the first time, Alibaba Dragonwell (Alibaba Cloud's open source Open JDK long-term support version)'s cold start acceleration technology, multi-threaded operation acceleration technology and SAE's own in-situ The combination of the upgrade strategy and the image warm-up strategy has achieved a 45% increase in the end-to-end startup speed of Java applications, which only takes 15 seconds at the fastest, and a 30% increase in multi-threading performance.

Due to the rapid changes in business scenarios and user habits, digital services in many industries have experienced rapid growth, and accelerating the development of digital services has become an inevitable choice for traditional enterprises. Cloud native is the shortest path for enterprise digitalization . More and more traditional enterprises are embracing cloud native, with faster and more flexible development and delivery models to meet the rapidly changing needs of the market, and accelerate business innovation. Traditional retail companies rely on Serverless to ensure the success of major promotions again and again, which is the best proof of this trend.

Copyright Statement: content of this article is contributed spontaneously by Alibaba Cloud real-name registered users. The copyright belongs to the original author. The Alibaba Cloud Developer Community does not own its copyright and does not assume corresponding legal responsibilities. For specific rules, please refer to the "Alibaba Cloud Developer Community User Service Agreement" and the "Alibaba Cloud Developer Community Intellectual Property Protection Guidelines". If you find suspected plagiarism in this community, fill in the infringement complaint form to report it. Once verified, the community will immediately delete the suspected infringing content.