It doesn't make sense without a distributed gateway

From the perspective of routing agent 、 Interpret the common routing proxy methods in centralized

Common way

nginx+lua

• nginx It's a high performance HTTP Reverse proxy . In our previous projects, we used nginx There are two main uses ：  Direction agent + Static resource server management

• Click me to download

• Once the download is complete nginx It is also very convenient to start , What the author demonstrates here is windows Version of . You can either click on the application directly or start it in command line mode
  ./nginx
  perhaps
  nginx
  . See the specific system

The service broker

• Under normal circumstances, agents use hair HTTP Under the Server To configure the . For example, we need a simple proxy for one of our services
  http://192.168.44.131:9200
   .. The effect we want is through local localhost You can access the interface provided by this service .

server {
 listen 9200;
 server_name localhost;
 location / {
 proxy_pass http://192.168.44.131:9200;
 #root html;
 #index index.html index.htm;
 }
 }


• Of course, after the configuration, we need to
  nginx -s reload
  To restart nginx To reload the configuration . At this time we visit
  localhost:9200/zxhtom/start
  In fact, it is represented to
  http://192.168.44.131:9200/zxhtom/start
  On this interface .

Static resource agents

• As we mentioned above, in addition to service agents .nginx It can also be used as a static resource server .

server{
 listen 90;
 server_name localhost;
 location / {
 root D:\study\PageOffice_4.6.0.3_Java;
 index index.html;
 }
 }


• It also uses HTTP Under the SERVER .  This code means that it will be local
  D:\study\PageOffice_4.6.0.3_Java
  This folder acts as a local 90 Port entry . We go through
  localhost:90/index.html
  You can access local
  D:\study\PageOffice_4.6.0.3_Java\index.html
  The file .

Load balancing

upstream redis {
 hash $remote_addr consistent;
 # $binary_remote_addr;
 server 192.168.44.131:6379 weight=5 max_fails=3 fail_timeout=30s;
 server 192.168.44.132:6379 weight=1 max_fails=3 fail_timeout=30s;
 }
 server {
 listen 6379;#redis Server listening port
 proxy_connect_timeout 10s;
 proxy_timeout 300s;# Set the timeout between the client and the proxy service , If 5 If there is no operation within minutes, it will be disconnected automatically .
 proxy_pass redis;
 }


• Sometimes our services are not just single node Services . If it is distributed, we need to carry out load balancing strategy for services . At this time, we can't just agent services . We need to add load policy . We go through
  upstream
    To define services , And define the service policy . Finally, we server Used directly in
  upstream
  The service name is ok . Among them weight It is to set the weight of service polling in the cluster .

Common commands

restart ： nginx.exe -s reload close ：nginx.exe -s stop Check configuration legitimacy ：nginx.exe -t

zuul

• zuul yes netflix Open source project , About zuul In simple terms, what I analyzed in the previous article should be very comprehensive .
  Let me see zuul chapter

• adopt zuul We can realize dynamic routing 、 Certificate authority 、 Dynamic filters 、 The final thing is that it can be customized or combined hystix Carry out service

• zuul It can also realize gray-scale publishing, which can solve our downtime publishing problem . Think about if your system is 24 It can't stop for hours zuul To achieve gray Publishing

Grayscale expansion

• Above, we realize the forwarding of the route through the parameters specified in the request. The implementation principle is with the help of eureka Of metadata Parameter attribute of the route .

• But we should have encountered some software that treat different regions differently .

• For example, Alipay ant forest has different strategies in different cities

• For example, a software invites you to participate in the use of the inner version

• All of the above are collectively referred to as grayscale publishing , The principle is also very simple. In fact, there are many examples ,zuul Forward to different instances according to the characteristics of the request . Different cities route by Region , Inside the invitation is to route through personal user information . Their implementation is inseparable from what we described above
  Ribbon.Predicate
   . If we want to expand grayscale publishing, we cannot do without
  Predicate

Predicate

• The function of this class is to make assertions , yes google The idea put forward . Specific about google
  Tap me gently

• stay ribbon Special topic , We simply read the source code to understand ribbon How to carry out load balancing and internal load balancing strategies . If you are interested, you can click the homepage to find .

• Today we'll look at it ribbon How to filter the list of services before load balancing .

// Return assertions based on input  true or false
@GwtCompatible
public interface Predicate<T> {
 // Assert against the input , This method has and not only has the following requirements ： 1、 Will not cause any data pollution  2、 stay T Of equals Equal in apply Is the same effect
 boolean apply(@Nullable T input);
 // Return to two Predicate Are they the same? . General situation Predicate The implementation does not need to be rewritten equals Of  .  If the implementation can show according to their own needs predicate Are they the same? . What is the same is two predicate object apply The same result means the same object
 @Override
 boolean equals(@Nullable Object object);
}


• So let's go through Predicate To achieve simple data filtering . Of course, some people will say why not java8 stream What about filtering . This is just for Ribbon Medium service filtration paving . as for ribbon Why not use stream operation ？ Personal role google Of Predicate It is more convenient in decoupling .

 @Test
 public void pt() {
 List<User> userList = new ArrayList<>();
 for (int i = 0; i < 100; i++) {
 userList.add(new User(Long.valueOf(i+1), &quot; Zhang San &quot;+(i+1)));
 }

 Predicate<User> predicate = new Predicate<User>() {
 @Override
 public boolean apply(User user) {
 return user.getId() % 2 == 0;
 }
 };
 ArrayList<User> users = Lists.newArrayList(Iterables.filter(userList, predicate));
 System.out.println(users);
 }


AbstractServerPredicate

Premise recall

• As can be seen from our class structure diagram above
  AbstractServerPredicate
    yes
  Predicate
    Implementation class of . This class is also Ribbon The key role in getting the list of services . Because the function expansion is based on this class .

jmnarloch First time to know

• This is my in RIbbon The content in . We can know Ribbon In the end
  BaseLoadBalancer
    Load balancing in . Inside rule The default is
  new RoundRobinRule()
   , Because we introduced
  io-jmnarloch
   . First look at the internal class structure

• io-jmnarloch
    The interior is not very complicated , At least Ribbon、feign Compared with these, he is really simple . There are four inside package

Register load Rule

• We can see in the support In bag
  RibbonDiscoveryRuleAutoConfiguration
    Configuration of the rule Package defined Ribbon Load balancing class Rule.

• The power failure hit
  BaseLoadBalancer
    We can see that rule It's us
  rule.MetadataAwareRule
    This class . Here and ribbon There seems to be a discrepancy in the chapters , We are ribbon The chapter says you need to customize rule I need to be in
  @RibbonClient(name = "CLOUD-PAYMENT-SERVICE",configuration = MySelfRule.class)
    This way, .

• It's actually configuring
  DiscoveryEnabledRule
    When registering, there is  
  @Scope(ConfigurableBeanFactory.SCOPE_PROTOTYPE)
    Represents the scope

MetadataAwareRule How to filter Services

• adopt
  MetadataAwareRule
    Combined with the code, we can know that the final result is PredicateBaseRule#choose  Select the service list

• This predicate  It's us choose in getPredicate() Method acquired . So in ribbon Before selecting services, you will pass
  MetadataAwarePredicate
    Filter Services .

• After getting the filter object , We will execute
  chooseRoundRibbinAfterFiltering
   .

go back to AbstractServerPredicate

• As mentioned above, it will finally pass Predicate To carry out chooseRoundRobinAfterFiltering.  Remember the beginning predicate Have you got the structure diagram of .
  MetadataAwarePredicate
  Finally inherit
  AbstractServerPredicate
   .  and
  AbstractServerPredicate
   # 
  chooseRoundRobinAfterFiltering
    It's dependence getEligibleServers` To get the appropriate list of services .

• stay
  AbstractServerPredicate
    Achieved a lot chooseXXX Methods . because ribbon The default is polling, so in BaseLoadBalance Middle is choice Round Corresponding method . We can modify these methods by ourselves . I won't repeat it here

• Eligible
    Translate into appropriate .getEligibleServers  Get the right list of services .

• We can obviously see that the final filtering logic falls in apply On the way .

• This is what we passed above
  metadata-map:lancher
    Configure our service information .

• Here is
  AbstractServerPredicate
    After simplification . Mainly
  getEligibleServers
    This method .

Subclass

• on top
  AbstractServerPredicate
    In the structure diagram, we can see that except
  DiscoveryEnabledPredicate
    Outside this subclass , There are four subclasses .