[untitled] how to realize pluggable configuration?

Source code Gitee link ：http://gitee.com/austin

01、 Pluggable

After my project gradually took shape , A lot of friends spit about my project Too heavy 了 , There are a few middleware dependencies .

In the first edition, you need Strong dependence MySQL/Redis/Kafka/Apollo（ You need to deploy these middleware when the project starts ）, Weak dependence
prometheus/graylog/flink/xxl-job（ Want to have a complete project experience , We need to deploy these ）.

• MySQL No one roast , Databases , It can be said that it is necessary for the back end .
• Redis No one has roast yet , After all, I still use too much , There are no strong competitors .
• Apollo Often roast whether it can be replaced by Nacos.
• Kafka Sometimes someone roast , Want to support RabbitMQ、RocketMQ.

I used to have an idea ： Middleware is not easy to replace in the company , Now my code has implemented a posture , I don't think it's necessary to support multiple middleware implementations , If you want to change, just change it yourself , It's not hard .

「“Apollo Too heavy ,Apollo It fails to work well ！ Support quickly Nacos！”」「“ Support RocketMQ How about ”」「“ Can you support RabbitMQ？”」

It's not a big reason for me , I still think Apollo Good use , Mature enough to be stable , Empathy Kafka The same is true . But when I was roast too much , Always doubt whether you are not doing well enough , I will also discuss with the bosses around me , Is it necessary to support some functions .

reasoning , I have changed , I get it again

In order to make the message push platform Austin Easy to use , I put it first Apollo Make weak dependence , Can be selected by configuration Read local files Or read the configuration center （Apollo）. In fact, when we use Apollo when , Even if Apollo Hang up ,Apollo It has a strong disaster tolerance ability （ Bring your own local file ）

secondly , I put Kafka Make weak dependence , Can be selected by configuration use Guava Of eventbus Or take the distributed message queue （Kafka）, It may be supported later RocketMQ/RabbitMQ, Those who are interested can also implement one based on my code , Rub pull request Also very fragrant .

On the one hand is Lower the use threshold And do , On the other hand, the specific implementation can be Pluggable , This is a Open source project The required . I think if it is a company level production environment online project , More consideration is given to this Heterogeneous disaster recovery （ Not pluggable ）.

So , Now the message push platform Austin The default strong dependency is only left MySQL and Redis, Other middleware are weak dependencies , To be pluggable, I am With the help of configuration To instantiate different middleware .

When my configuration austin-mq-pipeline=eventbus when , I won't instantiate Kafka Relevant producers and consumers , Instead, initialize eventBus The producers and consumers of , That's natural Implementation class under interface Is to eventbus Of

02、（ Colored eggs ）KAFKA Support TAG Filter

My shareholders can use my remote service directly ：Kafka Of Topic Is Shared ,Group Consumers are also shared , Without modification , Direct use will bring a problem .

When two or more shareholders start up locally at the same time Austin, It will compete for consumption Topic（ It is equivalent to having multiple consumers in a consumer group ）, As a result, you may not receive your own debugging message when the test is issued （ Was robbed by other shareholders ）.

My first thought to solve this problem is very simple ： Different shareholders use different group（ Equivalent to each shareholder will have an independent consumer group ）, Then it's over ？ Just my groupId Generation depends on channels code, Get rid of code It's over .

But there is still a problem ： Each shareholder has an independent consumer group , It means that each shareholder can consume the whole topic All the news of , This in turn means that shareholders will receive the test information from other shareholders （ Obviously, I only want to test the news myself , But there was a message sent by someone else ）.

To solve this problem , In addition to giving each shareholder an independent topic, That's it according to tag Filter La .

stay Kafka To achieve this effect , Pretty simple ： At the time of sending , hold tag Write in Kafka The head of , Put yourself before consumption tag Filter out the message and it's over .

03、 summary

I've been iterating since I started writing this project , This process received a lot of roast . This kind of roast is mostly positive , After all, some people roast, that shows that my project is really used , Someone is watching .

I have an idea recently ： Make this system online , Major developers can call my interface when pushing messages , It must be interesting to do this , There will also be more challenges and needs . Then I can always iterate , In this process, I will also learn a lot of things I didn't know before .

This time I use @ConditionAlOnProperties This annotation is used to realize pluggable configuration , But in fact, if it is in the form of a two-party Library , Use SPI The posture of Will be more elegant .

That's it today , Welcome to like 、 Leaving a message. .

Source code GitHub link ：http://github.com/austin