A retailer is a leading enterprise in the super industry , Its main business covers shopping centers 、 The hypermarket 、 The supermarket 、 Standard supermarket 、 Boutique supermarket 、 Convenience stores, unattended smart stores and other retail formats , Involving Omni channel retail 、 Warehousing logistics 、 Restaurant 、 Consumer service 、 Data services 、 Financial business 、 Cross border trade, etc . In order to continuously support the rapid and stable development of business , After it's fast up in the cloud , Transform the core business into the whole Serverless The middle platform mode of Architecture , Using function calculation + API gateway + Form storage OTS As the core of computing network storage , Flexible support for daily and big peak valley resources , Easy support 618/ double 11/ double 12 Great promotion .
Why do traditional enterprises need more attention Serverless
In order to reduce the cost of technology development 、 Improve operation and maintenance efficiency , More and more enterprises choose to use Serverless As a base for R & D , Develop your business . stay CNCF Serverless According to the Research Report , A large number of domestic developers are moving the traditional architecture to Serverless On the migration .Serverless The emergence of the digital transformation of traditional enterprises has brought more opportunities .
Today, , A large number of cutting-edge technical talents prefer to work in Internet companies , Traditional enterprises are faced with a large number of rigid needs for technology upgrading and restructuring , There is a demand for cloud native technology between talent gap and technology upgrading .Serverless The emergence of the R & D personnel in the budget 、 Lack of operation and maintenance experience . In the case of helping enterprises to cope with the business peak , R & D personnel can easily handle , It not only greatly reduces the threshold of R & D technology , And it has greatly improved R & D efficiency . For developers , Online alert 、 Flow observation and other tools are available , The key is to eliminate the burden of operation and maintenance , It has provided the technology dividend for the vast number of developers . For traditional enterprises ,Serverless It shortens the distance between Internet companies and traditional enterprises in terms of technological competitiveness .
From the upper cloud to the primary cloud
2016 Years later , With the rapid development of domestic public cloud , It's irresistible to go up to the cloud in an all-round way . A famous shopping mall is in 2018~2019 During the year , The system modules in the self built computer room are gradually migrated to the public cloud , The overall architecture hasn't changed much , Therefore, the migration work is relatively smooth .
There are some improvements and deficiencies after the system is fully migrated to the cloud
No longer need to care about the Internet 、 Hardware details of the operating system
For example, Alibaba cloud's ECS We will do scheduling and early warning in advance , Transfer user data and prepare multiple data for disaster , Prevent disk failure .
It's fast and easy to upgrade
For example, users use 4 Nuclear machines , When you find that your business is growing rapidly and you need to upgrade your hardware , Just make a mirror image . Like taking a disk snapshot at night , Re apply for a new machine , Then restore the snapshot , You can complete one click migration . It's a very quick way for users , It's also a good experience for developers .
The machine expansion time is greatly shortened
The above mentioned is single machine expansion , such as 4 Nuclear rise to 8 nucleus 、16G Rise to 32G Of memory . In addition, there are horizontal expansion , For example, the user trading system API Interface , With the development of business, we need to change from the original 2 The number of machines expanded to 8 Taiwan machine , In this case, the user only needs to apply for the machine , Then extend the image to different machines .
It's difficult to budget resources
It's impossible to predict the volume that a business can achieve when it encounters a big promotion activity , Therefore, it is impossible to accurately calculate the number of required hardware .
Horizontal expansion has higher requirements for R & D . For example, whether the data should be stateless , If there is no state, horizontal expansion will be easier , And if it's a state , Data may need to be cached , This will involve database related issues , For example, data is out of date 、 Consistency, etc . If you don't know enough about this , It will be more difficult to do horizontal expansion .
Water level monitoring
Many developers are not perfect in water level monitoring , If the various business systems are mixed on one machine , When the water level of the machine is high , Want to quickly check the problem and timely flow control 、 Split 、 Temporary repair is particularly difficult .
Financial budget difficulties
Similar to resource budget difficulties .
High cost of hardware upgrade
To achieve the user no sense of lossless upgrade , It may involve the consistency between the processing on the connection and the database . If multiple modules need to be upgraded at the same time , We should also pay attention to the compatibility of data structures .
Database single point of failure
Many manufacturers put all the data in one database , If it is not handled properly, it may cause a single point of failure . This requires data splitting , If it's rough , You need to pay attention to transaction and lock related issues , Efficiency will be greatly reduced ; In detail , It will be difficult to do query and sort , Cause some trouble to the business implementation .
In the middle of a year , Due to the uncontrollable access of online business users , Too much data ,MySQL Stand alone access is broken , This caused problems with the storage database , It affects multiple systems , Caused a certain loss . Therefore, in the follow-up service transformation process , Database selection is made by MySQL Change to table storage OTS, The biggest advantage of table storage is that users don't need to care about the proportion between the number of visits and the number of machines . As long as the number of visits increases , The background will automatically expand the capacity of machines , Meet the high concurrency of data reading ; When the data concurrent request reduction is at a low peak period , The background will recycle the machine , Users no longer need to care about the number of machines and how to transfer them .
For the problem of uncontrollable user traffic , Customers have introduced alicloud products “API gateway ” ,API Gateway can control and release and control traffic for different channel providers . For example, it is found that the wechat channel traffic is abnormal , With the help of API Gateway for current limiting .
In addition, calculation is also a very important problem , After exploration, customers found that alicloud function computing FC It fits its business scenario very well . For example, time to buy 、 Coupons and other activities caused a huge burst impact , When we find that the computing resources are not enough, it's definitely too late to buy a machine , The function calculation and timely expansion function can solve this problem . In addition, its data observation and abnormal alarm function , And attracted customers .
This year, 3 month , An authoritative advisory body Forrester Release 2021 First quarter FaaS Platform evaluation report , Alicloud's function computing relies on its in product capabilities 、 Security 、 Strategic vision, market size and other advantages stand out , The product capacity ranks first in the world , This is also the first time that a Chinese cloud manufacturer has entered FaaS The leadership quadrant .
After intense testing verification , Technical personnel found that the excellent performance of function computing is very suitable for their own business highly flexible member query system . from 2019 year 7 Month begins , The customer's technical team is less than 3 In months , Transfer all the original member data copies to form storage , And will all channel business's API Full migration to alicloud API The gateway does distribution , The computing business of member query service is also fully migrated to alicloud function computing .
2019 Years of double 11, Function calculation as a calculation module , Tables are stored as storage modules , Successfully help customers through the big promotion , While carrying the peak traffic, it ensures the flexibility to deal with the business . Instead of using Serverless Because of the underestimation of our business , There are some anomalies . It is precisely because the function is calculated in double 11 It's exciting for customers and technicians . After a successful promotion campaign , Customers use function calculation and table storage in all businesses ！
Super overall architecture of new retailers
- whole Serverless framework ： Function calculation + API gateway + Form storage ;
- High flexibility ： Millisecond elastic expansion 、 Sufficient resource pool water level 、 High availability across zones ;
- Agile development is free of operation and maintenance ： Functional minimalist programming can focus on business innovation , No procurement and deployment costs 、 Provide complete observable ability such as monitoring and alarming .
2019 In the second half of , Alibaba cloud announced the launch of functional computing 2.0, Support reservation mode , Solve the problem of large cold start delay ; This paper introduces the single instance multi request problem , Less instances support heavy IO High concurrency type request calls ; Support custom runtime , Support one click migration tradition Web Architecture server .2.0 The emergence of functional computing in business and scale to achieve a huge upgrade .
After experiencing the test of offline scenes in the past , Customers will be the channel business and its mobile App, And online trading 、 Grab coupons regularly 、 Second kill business also all from ECS Migration to function computing 2.0, After the reservation mode is enabled and the multiple concurrent mode of single instance is adjusted , Successfully carried the flood peak discharge request which is dozens of times as usual .
Compare the above “ Time - Flow chart ” And “ Time - Delay ” Two pictures show , The impact of the sharp rise of burst traffic on the delay change caused by users is very small , From the actual user feedback, it also confirmed that the user experience is very smooth .
All the data and business goes to the cloud , It's not just the psychological pressure of R & D personnel that can be relieved , And a lot of R & D personnel to reduce the burden , So that we can do things that focus more on business logic . Functional computing allows developers to avoid managing the server infrastructure , Just write code and upload , The system will be ready to compute resources , It also provides log queries 、 Performance monitoring 、 Alarm and other functions . If it's the old model , Supermarket double 11 Great promotion , The relevant technical team couldn't sleep , Only rely on the expansion machine to support large volume of traffic and business , No one's in his mind . Now the problem of capacity expansion is left to alicloud , The water level is much higher than the limit of the customer's original reserve capacity .
This year, ,Serverless Ushering in a major upgrade . Image acceleration technology of function computing heavyweight publishing container , Container startup delay is shortened 50%-80%, Transfer the burden of image optimization originally belonging to developers to function calculation , Further help developers improve productivity , Focus on business innovation . This technology comes from the container environment with super large scale and highly complex scenes of Alibaba group , For mirror storage 、 Acceleration technology has a deep accumulation , And took on 3 Double eleven 、 12-12 、 The Spring Festival and other scenes are challenging .
meanwhile ,Serverless Application engine (SAE) Blockbuster released Java Application startup acceleration function , For the first time Alibaba Dragonwell ( Alibaba cloud open source Open JDK Long term support version ) Cold start acceleration technology 、 Multithreading acceleration technology and SAE Our own in situ upgrade strategy 、 Image preheating strategy , Realized Java The end-to-end startup speed of the application is improved 45%, The fastest way is just 15s, Multithreading performance improvement 30%.
Because of the business scenario 、 User habits change rapidly , Digital business is growing rapidly in many industries , Accelerating the development of digital business has become the inevitable choice of traditional enterprises . Cloud native is the shortest path for enterprise digitalization , More and more traditional enterprises are embracing cloud nativity , With the help of faster 、 Flexible development and delivery model , Meet the rapidly changing needs of the market , And accelerate business innovation . With the help of Serverless Ensure the success of big promotion again and again , It's the best proof of this trend .