Technology selection and decision-making of large-scale Internet applications, 10 records of success and failure

As a decoupled version based on the old version , This large Internet application product is from 2009 It began to land in the middle of the year . And I am also one of the main creators of this version , To this day , The team has grown to the scale of a large Internet product team with hundreds of people to develop and test , Release 、 Many commercial versions have been cut and launched .

An examination of the architecture , Reflection on model selection and design , It's not just at the beginning of the product , It should be carried out in the whole process of product development , This is how the team's ability to make similar products matures in continuous summary and self-criticism . Here is my personal opinion , Inevitably, it is not to the appetite of many people , But I still hope these words have enough value to make people think . in any case , For such a product , Interpret its historical story from today's perspective , What's more, it has a special flavor .

—————————————————————————————

5 Successful records ：

1、Portlet Technology is the core of the whole architecture .

This is a record of success , It is also a record of failure .

Portlet It brings considerable page customization flexibility to different customized versions of each office , Don't understand, jsp Administrators of can deploy pages according to their own requirements , Through simple selection and dragging , Show the rich channels .

On the other hand ,Portlet It retains considerable flexibility for column expansion and customization , Especially for potential Internet applications, keep the scalability according to the column dimension , Leave enough space .

2、 The persistence layer chooses a more lightweight iBatis, no choice Hibernate.

The fact proved that ,iBatis transparent 、 Simple , Easy to configure and use , Most developers can read through the code of the persistence layer , Developers who are good at database can easily complete Oracle Next SQL Statement tuning , Applied to the iBatis In the configuration file , It doesn't take much ORM Technology accumulation , There is no need for deep Hibernate Tuning techniques .

3、 The business core logic layer can be abstracted , It can be independently released API modular .

Facing a variety of access channels , Pass the core business API Behavior extracted , The biggest benefit to the project team is ： Clear team division .

although API The module is not mature , however API The birth and development of means that the development and customization teams of various access portals can focus more on the work centered on presentation , Give the sorting of business code to a special API The team does .

In the long run , pure Java Of API It's clean 、 Simple and easy UT Of , In this natural way, the persistent layer is isolated , It also protects the code quality of the core business .

4、 Extract the interface display part of the function into reusable business tags .

Some people will disagree with this , in fact , except FreeMarker Its performance is really beyond flattery , The benefits of componentizing the display part of the interface in the form of labels are great .

Ideal situation , The customization team can insert through simple tags 、 Deletion and modification , Complete the customization of the page , This is more complicated than understanding grandeur jsp page , It's a lot easier to copy and paste .

5、 The infrastructure is stable and guaranteed .

Infrastructure includes logs 、 Protocol stack 、License wait , Most are stable and easy to use .

For example, abnormal system , The whole exception system brings a natural and easy exception handling process to the development , Developers only need to pay more attention to the core process , Leave the error process to the runtime exception ; Different levels of exception capture bring convenience to the error display and induction of the final page .

There are also regrets , For example, the error code is tangled , The error code is the result of fierce struggle and compromise in the internal discussion of the team , Some are too large and complicated , This seems to validate ： Software engineering is not an election .

—————————————————————————————

5 Failed records ：

1、Portlet Technology is the core of the whole architecture .

This is a record of success , It is also a record of failure .

Portlet Many of its features are far from being properly developed , for example Portlet The maintenance of state 、 The ability of remote aggregation and so on , But it brings many troubles to developers , For example, page decomposition is difficult ,Portlet Session and Portal Session The mutual differences between them , Aggregation process performance problems, etc , It puts forward higher requirements for developers to customize their hands .

2、 Independence is based on Portlet The core is responsible for portal operation Portal platform .

Portlet Specification is an abstraction of aggregate presentation behavior , Through the form of such an independent platform as componentization , Separate the page control aggregation process from business page presentation and business process processing , Developers can focus more on business development .

I think this is the original intention of its birth , But actually , But it brings complexity to the aggregation process , The method call stack is too deep , And the developers of portal customization , It also requires considerable training to get started .

3、 adopt ajax Way aggregation Portal In a different place war The management console page in the package .

The original intention is to make different management console pages follow different Portal Access channel separation release , Simple integration on the page during presentation .

However, due to the security mechanism of the browser and the mechanism of independent session management for different domains , It was introduced like a demon , It brings more difficulties than customization , Difficulties for developers to understand , There are also some problems that are difficult to solve when transferring information between pages in different domains because the session cannot be unified .

4、 Retain XDIME、WML and XHTML Three sets WAP Template page for .

Of course, there may be some factors beyond human control .

XDIME The purpose of the page is to convert it into a mobile phone through the terminal adaptation component WML and XHTML Page , And because of local or some historical reasons ,WML and XHTML It can't be simply abandoned , Whether it is an active decision or forced by helplessness , The problem is that the number of page templates has tripled , It brings a lot of workload to development and testing .

Final ,WML and XHTML The template was abandoned , Just keep XDIME A set of templates .

5、 Lack of a simple and manageable UI frame .

Front end development is the bottleneck of the whole product , Although the page is not very complex , The front-end chaos has brought many problems , These problems are mainly exposed in the product customization and the final user detail experience . Whether Internet products are professional , It is largely decided by the front-end team of the product .

According to different team levels 、 Different front-end display requirements , Need to customize different UI frame . Simplicity shown by , And you need a customized baseline version , That determines our UI The framework should be simple ; And due to the poor front-end ability of development members , It's our decision UI The framework should be easy to control and manage , You should not expose overly complex interface behavior to ordinary developers .

The articles are all original without special indication , It shall not be used for any commercial purpose without permission , Please keep the integrity of reprint and indicate the source link 《 Four fire's nagging 》

×Scan to share with WeChat