单体架构#

将业务的所有功能集中在一个项目中开发，打成一个包部署。

优点：

• 架构简单
• 部署成本低

缺点：

• 耦合度高（维护困难、升级困难）

分布式架构#

根据业务功能对系统做拆分，每个业务功能模块作为独立项目开发，称为一个服务。

优点：

• 降低服务耦合
• 有利于服务升级和拓展

缺点：

• 服务调用关系错综复杂

分布式架构虽然降低了服务耦合，但是服务拆分时也有很多问题需要思考：

• 服务拆分的粒度如何界定？
• 服务之间如何调用？
• 服务的调用关系如何管理？

人们需要制定一套行之有效的标准来约束分布式架构。

微服务#

微服务的架构特征：

• 单一职责：微服务拆分粒度更小，每一个服务都对应唯一的业务能力，做到单一职责
• 自治：团队独立、技术独立、数据独立，独立部署和交付
• 面向服务：服务提供统一标准的接口，与语言和技术无关
• 隔离性强：服务调用做好隔离、容错、降级，避免出现级联问题

微服务的上述特性其实是在给分布式架构制定一个标准，进一步降低服务之间的耦合度，提供服务的独立性和灵活性。做到高内聚，低耦合。

因此，可以认为微服务是一种经过良好架构设计的分布式架构方案 。

SpringCloud#

SpringCloud是目前国内使用最广泛的微服务框架。官网地址：https://spring.io/projects/spring-cloud。

SpringCloud集成了各种微服务功能组件，并基于SpringBoot实现了这些组件的自动装配，从而提供了良好的开箱即用体验。

组件包括：

• 服务注册发现：Eureka、Nacos、Consul
• 服务远程调用：OpenFeign、Dubbo
• 服务链路监控：Zipkin、Sleuth
• 统一配置管理：SpringCloudConfig、Nacos
• 统一网关路由：SpringCloudGateway、Zuul
• 流控、降级、保护：Hystix、Sentinel

服务间远程调用#

通过调用RestTemplate来实现

1. 在启动项中注册RestTamplate实例

   1
   @Bean
   2
   public RestTemplate restTemplate() {
   3
       return new RestTemplate();
   4
   }

2. 服务中调用

   1
   @Autowired
   2
   private RestTemplate restTemplate;
   3
   
   4
   public Order queryOrderById(Long orderId) {
   5
       // 1.查询订单
   6
       Order order = orderMapper.findById(orderId);
   7
       // 2. 远程查询User
   8
       String url = "<http://localhost:8081/user/>" + order.getUserId();
   9
       User user = restTemplate.getForObject(url, User.class);
   10
       // 3. 存入order
   11
       order.setUser(user);
   12
       // 4.返回
   13
       return order;
   14
   }

提供者与消费者#

在服务调用关系中，会有两个不同的角色：

服务提供者：一次业务中，被其它微服务调用的服务。（提供接口给其它微服务）

服务消费者：一次业务中，调用其它微服务的服务。（调用其它微服务提供的接口）

Eureka结构和作用#

注册中心用来解决消费者远端调用存在部署了多个实例的服务提供者。

• 问题1：order-service如何得知user-service实例地址？
  • user-service服务实例启动后，将自己的信息注册到eureka-server（Eureka服务端）。这个叫服务注册
  • eureka-server保存服务名称到服务实例地址列表的映射关系
  • order-service根据服务名称，拉取实例地址列表。这个叫服务发现或服务拉取
• 问题2：order-service如何从多个user-service实例中选择具体的实例？
  • order-service从实例列表中利用负载均衡算法选中一个实例地址
  • 向该实例地址发起远程调用
• 问题3：order-service如何得知某个user-service实例是否依然健康，是不是已经宕机？
  • user-service会每隔一段时间（默认30秒）向eureka-server发起请求，报告自己状态，称为心跳
  • 当超过一定时间没有发送心跳时，eureka-server会认为微服务实例故障，将该实例从服务列表中剔除
  • order-service拉取服务时，就能将故障实例排除了

搭建Eureka-Server#

1. 创建euraka-server服务

在父工程下创建子模块，maven项目

1. 导入eureka依赖 - server

1
<dependency>
2
    <groupId>org.springframework.cloud</groupId>
3
    <artifactId>spring-cloud-starter-netflix-eureka-server</artifactId>
4
</dependency>

1. 编写启动类

添加@EnableEurekaServer注解，开启Eureka注册中心功能

1
@SpringBootApplication
2
@EnableEurekaServer
3
public class EurekaApplication {
4
    public static void main(String[] args) {
5
        SpringApplication.run(EurekaApplication.class, args);
6
    }
7
}

1. 编写配置文件

1
server:
2
  port: 10086
3
spring:
4
  application:
5
    name: eureka-server
6
eureka:
7
  client:
8
    service-url:
9
      defaultZone: <http://127.0.0.1:10086/eureka>

1. 启动服务

访问设置的Eureka服务路径

服务注册#

1. 依赖导入 - client

1
<dependency>
2
    <groupId>org.springframework.cloud</groupId>
3
    <artifactId>spring-cloud-starter-netflix-eureka-client</artifactId>
4
</dependency>

1. 配置文件

1
spring:
2
  application:
3
    name: userservice
4
eureka:
5
  client:
6
    service-url:
7
      defaultZone: <http://127.0.0.1:10086/eureka>

1. 启动多个服务实例

SpringBoot中复制启动配置，VM options中配置-Dserver.port=8082

服务发现#

作为消费者的order-service前两步与提供者一致。

1. 服务拉取与负载均衡

负载均衡：在restTemplate的Bean中添加@LoadBalanced注解即可

1
@Bean
2
@LoadBalanced
3
public RestTemplate restTemplate(){
4
    return new RestTemplate();
5
}

获取服务：

1
public Order queryOrderById(Long orderId) {
2
    // 1.查询订单
3
    Order order = orderMapper.findById(orderId);
4
    // 2. 远程查询User
5
//        String url = "<http://localhost:8081/user/>" + order.getUserId();
6
    String url = "<http://userservice/user/>" + order.getUserId();
7
    User user = restTemplate.getForObject(url, User.class);
8
    // 3. 存入order
9
    order.setUser(user);
10
    // 4.返回
11
    return order;
12
}

spring会自动帮助我们从eureka-server端，根据userservice这个服务名称，获取实例列表，而后完成负载均衡。

Ribbon负载均衡#

1
@Bean
2
public IRule randomRule(){
3
    return new RandomRule();
4
}

1
userservice: # 给某个微服务配置负载均衡规则，这里是userservice服务
2
  ribbon:
3
    NFLoadBalancerRuleClassName: com.netflix.loadbalancer.RandomRule # 负载均衡规则

1
ribbon:
2
  eager-load:
3
    enabled: true
4
    clients: userservice

服务注册到Nacos#

1
<dependency>
2
    <groupId>com.alibaba.cloud</groupId>
3
    <artifactId>spring-cloud-alibaba-dependencies</artifactId>
4
    <version>2.2.6.RELEASE</version>
5
    <type>pom</type>
6
    <scope>import</scope>
7
</dependency>

1
<dependency>
2
    <groupId>com.alibaba.cloud</groupId>
3
    <artifactId>spring-cloud-starter-alibaba-nacos-discovery</artifactId>
4
</dependency>

1
spring:
2
  cloud:
3
    nacos:
4
      server-addr: localhost:8848

服务分级存储模型#

一个服务可以有多个实例，假如这些实例分布于全国各地的不同机房，Nacos就将同一机房内的实例 划分为一个集群。

微服务互相访问时，应该尽可能访问同集群实例，因为本地访问速度更快。当本集群内不可用时，才访问其它集群。

• 配置

修改application.yml文件，添加集群配置：

1
spring:
2
  cloud:
3
    nacos:
4
      server-addr: localhost:8848
5
      discovery:
6
        cluster-name: HZ # 集群名称

Nacos中提供了一个NacosRule的实现，可以优先从同集群中挑选实例。

1
userservice:
2
  ribbon:
3
    NFLoadBalancerRuleClassName: com.alibaba.cloud.nacos.ribbon.NacosRule # 负载均衡规则

• 权重配置

实际部署中会出现这样的场景：

服务器设备性能有差异，部分实例所在机器性能较好，另一些较差，我们希望性能好的机器承担更多的用户请求。

但默认情况下NacosRule是同集群内随机挑选，不会考虑机器的性能问题。

因此，Nacos提供了权重配置来控制访问频率，权重越大则访问频率越高。

• 环境隔离

  Nacos提供了namespace来实现环境隔离功能。

  • nacos中可以有多个namespace
  • namespace下可以有group、service等
  • 不同namespace之间相互隔离，例如不同namespace的服务互相不可见

• 服务实例

  Nacos的服务实例分为两种l类型：

  • 临时实例：如果实例宕机超过一定时间，会从服务列表剔除，默认的类型。
  • 非临时实例：如果实例宕机，不会从服务列表剔除，也可以叫永久实例。

  配置一个服务实例为永久实例：

  1
  spring:
  2
    cloud:
  3
      nacos:
  4
        discovery:
  5
          ephemeral: false # 设置为非临时实例

Nacos和Eureka区别#

Nacos和Eureka整体结构类似，服务注册、服务拉取、心跳等待，但是也存在一些差异

• Nacos与eureka的共同点
  • 都支持服务注册和服务拉取
  • 都支持服务提供者心跳方式做健康检测
• Nacos与Eureka的区别
  • Nacos支持服务端主动检测提供者状态：临时实例采用心跳模式，非临时实例采用主动检测模式
  • 临时实例心跳不正常会被剔除，非临时实例则不会被剔除
  • Nacos支持服务列表变更的消息推送模式，服务列表更新更及时
  • Nacos集群默认采用AP方式，当集群中存在非临时实例时，采用CP模式；Eureka采用AP方式

从微服务拉取配置#

微服务要拉取nacos中管理的配置，并且与本地的application.yml配置合并，才能完成项目启动。

但如果尚未读取application.yml，又如何得知nacos地址呢？

因此spring引入了一种新的配置文件：bootstrap.yaml文件，会在application.yml之前被读取

1. 导入nacos-config依赖

1
<!--nacos配置管理依赖-->
2
<dependency>
3
    <groupId>com.alibaba.cloud</groupId>
4
    <artifactId>spring-cloud-starter-alibaba-nacos-config</artifactId>
5
</dependency>

1. 添加bootstrap.yml

1
spring:
2
  application:
3
    name: userservice # 服务名称
4
  profiles:
5
    active: dev #开发环境，这里是dev
6
  cloud:
7
    nacos:
8
      server-addr: localhost:8848 # Nacos地址
9
      config:
10
        file-extension: yaml # 文件后缀名

这里会根据spring.cloud.nacos.server-addr获取nacos地址，再根据

${spring.application.name}-${spring.profiles.active}.${spring.cloud.nacos.config.file-extension}作为文件id，来读取配置。

1. 读取nacos配置

1
@Value("${pattern.dateformat}")
2
private String dateformat;
3

4
@GetMapping("now")
5
public String now(){
6
    return LocalDateTime.now().format(DateTimeFormatter.ofPattern(dateformat));
7
}

配置热更新#

修改nacos中的配置后，微服务中无需重启即可让配置生效，也就是配置热更新。

• 方法一：添加注解@RefreshScope

• 方法二

  使用@ConfigurationProperties注解代替@Value注解。

  1. 在user-service服务中，添加一个类，读取patterrn.dateformat属性：

  1
  package cn.itcast.user.config;
  2
  
  3
  import lombok.Data;
  4
  import org.springframework.boot.context.properties.ConfigurationProperties;
  5
  import org.springframework.stereotype.Component;
  6
  
  7
  @Component
  8
  @Data
  9
  @ConfigurationProperties(prefix = "pattern")
  10
  public class PatternProperties {
  11
      private String dateformat;
  12
  }

  1. 在UserController中使用这个类代替@Value：

  1
  @Autowired
  2
  private PatternProperties patternProperties;
  3
  
  4
  @GetMapping("now")
  5
  public String now(){
  6
      return LocalDateTime.now().format(DateTimeFormatter.ofPattern(patternProperties.getDateformat()));
  7
  }

配置共享#

其实微服务启动时，会去nacos读取多个配置文件，例如：

• [spring.application.name]-[spring.profiles.active].yaml，例如：userservice-dev.yaml
• [spring.application.name].yaml，例如：userservice.yaml

而[spring.application.name].yaml不包含环境，因此可以被多个环境共享。

配置共享优先级：

当nacos、服务本地同时出现相同属性时，优先级有高低之分：

Nacos集群#

通过Nginx代理多个Nacos服务器

nacos的conf中配置cluster.conf，其中添加Nacos集群地址。

Feign代替RestTemplate#

1. 引入依赖

   1
   <dependency>
   2
       <groupId>org.springframework.cloud</groupId>
   3
       <artifactId>spring-cloud-starter-openfeign</artifactId>
   4
   </dependency>

2. 添加注解

   在启动项添加@EnableFeignClients注解

3. 编写Feign客户端

   在服务中新建一个接口：

   1
   @FeignClient("userservice")
   2
   public interface UserClient {
   3
       @GetMapping("/user/{id}")
   4
       User findById(@PathVariable("id") Long id);
   5
   }

   这个客户端主要是基于SpringMVC的注解来声明远程调用的信息，比如：

   • 服务名称：userservice
   • 请求方式：GET
   • 请求路径：/user/{id}
   • 请求参数：Long id
   • 返回值类型：User

   这样，Feign就可以帮助我们发送http请求，无需自己使用RestTemplate来发送了。

4. 调用

   在服务类中调用方法：

   1
   @Autowired
   2
   private UserClient userClient;
   3
   
   4
   public Order queryOrderById(Long orderId) {
   5
       // 1.查询订单
   6
       Order order = orderMapper.findById(orderId);
   7
       // 2. 远程查询User
   8
       User user = userClient.findById(order.getUserId());
   9
       // 3. 存入order
   10
       order.setUser(user);
   11
       // 4.返回
   12
       return order;
   13
   }

自定义配置#

Feign可以支持很多的自定义配置，如下表所示：

一般情况下，默认值就能满足我们使用，如果要自定义时，只需要创建自定义的@Bean覆盖默认Bean即可。

1
feign:
2
  client:
3
    config:
4
      userservice: # 针对某个微服务的配置
5
        loggerLevel: FULL #  日志级别

1
feign:
2
  client:
3
    config:
4
      default: # 这里用default就是全局配置，如果是写服务名称，则是针对某个微服务的配置
5
        loggerLevel: FULL #  日志级别

1
public class DefaultFeignConfiguration  {
2
    @Bean
3
    public Logger.Level feignLogLevel(){
4
        return Logger.Level.BASIC; // 日志级别为BASIC
5
    }
6
}

1
@EnableFeignClients(defaultConfiguration = DefaultFeignConfiguration .class)

1
@FeignClient(value = "userservice", configuration = DefaultFeignConfiguration .class)

Feign使用优化#

Feign底层发起http请求，依赖于其它的框架。其底层客户端实现包括：

• URLConnection：默认实现，不支持连接池
• Apache HttpClient ：支持连接池
• OKHttp：支持连接池

因此提高Feign的性能主要手段就是使用连接池代替默认的URLConnection。

这里我们用Apache的HttpClient来演示。

1. 引入依赖

在order-service的pom文件中引入Apache的HttpClient依赖：

1
<!--httpClient的依赖 -->
2
<dependency>
3
    <groupId>io.github.openfeign</groupId>
4
    <artifactId>feign-httpclient</artifactId>
5
</dependency>

1. 配置连接池

在order-service的application.yml中添加配置：

1
feign:
2
  client:
3
    config:
4
      default: # default全局的配置
5
        loggerLevel: BASIC # 日志级别，BASIC就是基本的请求和响应信息
6
  httpclient:
7
    enabled: true # 开启feign对HttpClient的支持
8
    max-connections: 200 # 最大的连接数
9
    max-connections-per-route: 50 # 每个路径的最大连接数

总结，Feign的优化：

1. 日志级别 尽量用basic
2. 使用HttpClient或OKHttp代替URLConnection
   • 引入feign-httpClient依赖
   • 配置文件开启httpClient功能，设置连接池参数

最佳实现#

Feign的客户端与服务提供者的controller代码非常相似，简化这种重复的代码编写方法如下

为什么需要网关#

Gateway网关是我们服务的守门神，所有微服务的统一入口。

网关的核心功能特性：

• 权限控制：网关作为微服务入口，需要校验用户是是否有请求资格，如果没有则进行拦截。
• 路由和负载均衡：一切请求都必须先经过gateway，但网关不处理业务，而是根据某种规则，把请求转发到某个微服务，这个过程叫做路由。当然路由的目标服务有多个时，还需要做负载均衡。
• 限流：当请求流量过高时，在网关中按照下流的微服务能够接受的速度来放行请求，避免服务压力过大。

在SpringCloud中网关的实现包括两种：

• gateway
• zuul

Zuul是基于Servlet的实现，属于阻塞式编程。而SpringCloudGateway则是基于Spring5中提供的WebFlux，属于响应式编程的实现，具备更好的性能。

快速入门#

1. 创建SpringBoot工程gateway，引入网关依赖

   1
   <!--网关-->
   2
   <dependency>
   3
       <groupId>org.springframework.cloud</groupId>
   4
       <artifactId>spring-cloud-starter-gateway</artifactId>
   5
   </dependency>
   6
   <!--nacos服务发现依赖-->
   7
   <dependency>
   8
       <groupId>com.alibaba.cloud</groupId>
   9
       <artifactId>spring-cloud-starter-alibaba-nacos-discovery</artifactId>
   10
   </dependency>

2. 编写启动类

   1
   @SpringBootApplication
   2
   public class GatewayApplication {
   3
   
   4
     public static void main(String[] args) {
   5
       SpringApplication.run(GatewayApplication.class, args);
   6
     }
   7
   }

3. 编写基础配置和路由规则

   路由配置包括：

   1. 路由id：路由的唯一标示
   2. 路由目标（uri）：路由的目标地址，http代表固定地址，lb代表根据服务名负载均衡
   3. 路由断言（predicates）：判断路由的规则，
   4. 路由过滤器（filters）：对请求或响应做处理

   1
   server:
   2
     port: 10010 # 网关端口
   3
   spring:
   4
     application:
   5
       name: gateway # 服务名称
   6
     cloud:
   7
       nacos:
   8
         server-addr: localhost:8848 # nacos地址
   9
       gateway:
   10
         routes: # 网关路由配置
   11
           - id: user-service # 路由id，自定义，只要唯一即可
   12
             # uri: <http://127.0.0.1:8081> # 路由的目标地址 http就是固定地址
   13
             uri: lb://userservice # 路由的目标地址 lb就是负载均衡，后面跟服务名称
   14
             predicates: # 路由断言，也就是判断请求是否符合路由规则的条件
   15
               - Path=/user/** # 这个是按照路径匹配，只要以/user/开头就符合要求

4. 启动网关服务进行测试

   重启网关，访问http://localhost:10010/user/1时，符合`/user/**`规则，请求转发到uri：[http://userservice/user/1，得到了结果：](http://userservice/user/1%EF%BC%8C%E5%BE%97%E5%88%B0%E4%BA%86%E7%BB%93%E6%9E%9C%EF%BC%9A)

断言工厂#

我们在配置文件中写的断言规则只是字符串，这些字符串会被Predicate Factory读取并处理，转变为路由判断的条件

例如Path=/user/**是按照路径匹配，这个规则是由org.springframework.cloud.gateway.handler.predicate.PathRoutePredicateFactory类来处理的，像这样的断言工厂在SpringCloudGateway还有十几个:

过滤器工厂#

GatewayFilter是网关中提供的一种过滤器，可以对进入网关的请求和微服务返回的响应做处理

• 对路由的请求或响应做加工处理，比如添加请求头
• 配置在路由下的过滤器只对当前路由的请求生效

1
spring:
2
  cloud:
3
    gateway:
4
      routes:
5
      - id: user-service
6
        uri: lb://userservice
7
        predicates:
8
        - Path=/user/**
9
        filters: # 过滤器
10
        - AddRequestHeader=Truth, dreaife is freaking awesome! # 添加请求头

1
spring:
2
  cloud:
3
    gateway:
4
      routes:
5
      - id: user-service
6
        uri: lb://userservice
7
        predicates:
8
        - Path=/user/**
9
      default-filters: # 默认过滤项
10
      - AddRequestHeader=Truth, dreaife is freaking awesome!

全局过滤器#

1
public interface GlobalFilter {
2
    /**
3
     *  处理当前请求，有必要的话通过{@link GatewayFilterChain}将请求交给下一个过滤器处理
4
     *
5
     * @param exchange 请求上下文，里面可以获取Request、Response等信息
6
     * @param chain 用来把请求委托给下一个过滤器
7
     * @return {@code Mono<Void>} 返回标示当前过滤器业务结束
8
     */
9
    Mono<Void> filter(ServerWebExchange exchange, GatewayFilterChain chain);
10
}

1
package cn.itcast.gateway.filters;
2

3
import org.springframework.cloud.gateway.filter.GatewayFilterChain;
4
import org.springframework.cloud.gateway.filter.GlobalFilter;
5
import org.springframework.core.annotation.Order;
6
import org.springframework.http.HttpStatus;
7
import org.springframework.stereotype.Component;
8
import org.springframework.web.server.ServerWebExchange;
9
import reactor.core.publisher.Mono;
10

11
@Order(-1)
12
@Component
13
public class AuthorizeFilter implements GlobalFilter {
14
    @Override
15
    public Mono<Void> filter(ServerWebExchange exchange, GatewayFilterChain chain) {
16
        // 1.获取请求参数
17
        MultiValueMap<String, String> params = exchange.getRequest().getQueryParams();
18
        // 2.获取authorization参数
19
        String auth = params.getFirst("authorization");
20
        // 3.校验
21
        if ("admin".equals(auth)) {
22
            // 放行
23
            return chain.filter(exchange);
24
        }
25
        // 4.拦截
26
        // 4.1.禁止访问，设置状态码
27
        exchange.getResponse().setStatusCode(HttpStatus.FORBIDDEN);
28
        // 4.2.结束处理
29
        return exchange.getResponse().setComplete();
30
    }
31
}

跨域问题#

跨域：域名不一致就是跨域，主要包括：

• 域名不同： www.taobao.com 和 www.taobao.org 和 www.jd.com 和 miaosha.jd.com
• 域名相同，端口不同：localhost:8080和localhost8081

跨域问题：浏览器禁止请求的发起者与服务端发生跨域ajax请求，请求被浏览器拦截的问题

解决方案：CORS

在gateway服务的application.yml文件中，添加下面的配置：

1
spring:
2
  cloud:
3
    gateway:
4
      globalcors: # 全局的跨域处理
5
        add-to-simple-url-handler-mapping: true # 解决options请求被拦截问题
6
        corsConfigurations:
7
          '[/**]':
8
            allowedOrigins: # 允许哪些网站的跨域请求
9
              - "<http://localhost:8090>"
10
            allowedMethods: # 允许的跨域ajax的请求方式
11
              - "GET"
12
              - "POST"
13
              - "DELETE"
14
              - "PUT"
15
              - "OPTIONS"
16
            allowedHeaders: "*" # 允许在请求中携带的头信息
17
            allowCredentials: true # 是否允许携带cookie
18
            maxAge: 360000 # 这次跨域检测的有效期

Monolithic Architecture#

Develop all business functionality in a single project and deploy as a single package.

Advantages:

• Simple architecture
• Low deployment cost

Disadvantages:

• High coupling (maintenance and upgrade are difficult)

Distributed Architecture#

Split the system by business function, with each business function module developed as an independent project, called a service.

Advantages:

• Reduces service coupling
• Facilitates service upgrades and scalability

Disadvantages:

• The calling relationships between services become intricate

Although distributed architecture reduces service coupling, there are many questions to consider when partitioning services:

• How to define the granularity of service decomposition?
• How do services call each other?
• How to manage the calling relationships between services?

People need to establish a practical and effective set of standards to constrain distributed architectures.

Microservices#

Characteristics of a microservice architecture:

• Single Responsibility: Microservice decomposition is smaller; each service corresponds to a unique business capability, achieving single responsibility
• Autonomy: Teams, technologies, and data are independent, with independent deployment and delivery
• Service-Oriented: Services provide standardized interfaces, independent of language and technology
• Strong isolation: Service calls are isolated, fault-tolerant, and degradable to avoid cascading issues

The above characteristics of microservices are essentially setting a standard for distributed architectures, further reducing coupling between services and providing independence and flexibility. Achieve high cohesion, low coupling.

Therefore, one can regard microservices as a well-architected distributed architecture solution.

SpringCloud#

SpringCloud is currently the most widely used microservice framework in China. Official website: https://spring.io/projects/spring-cloud。

SpringCloud integrates various microservice functionality components and, based on SpringBoot, provides auto-configuration for these components, delivering a good out-of-the-box experience.

Components include:

• Service registration and discovery: Eureka, Nacos, Consul
• Remote service invocation: OpenFeign, Dubbo
• Service tracing/monitoring: Zipkin, Sleuth
• Unified configuration management: SpringCloudConfig, Nacos
• Unified gateway routing: SpringCloudGateway, Zuul
• Traffic control, degradation, protection: Hystix, Sentinel

Inter-service Remote Calls#

Implemented by calling RestTemplate

1. Register a RestTemplate instance in the startup item

   1
   @Bean
   2
   public RestTemplate restTemplate() {
   3
       return new RestTemplate();
   4
   }

2. Calls within services

   1
   @Autowired
   2
   private RestTemplate restTemplate;
   3
   
   4
   public Order queryOrderById(Long orderId) {
   5
       // 1. Query the order
   6
       Order order = orderMapper.findById(orderId);
   7
       // 2. Remote query User
   8
       String url = "<http://localhost:8081/user/>" + order.getUserId();
   9
       User user = restTemplate.getForObject(url, User.class);
   10
       // 3. Put into order
   11
       order.setUser(user);
   12
       // 4. Return
   13
       return order;
   14
   }

Providers and Consumers#

In the service invocation relationship, there are two distinct roles:

Service Provider: In a business flow, the service that is called by other microservices. (Provides interfaces to other microservices)

Service Consumer: In a business flow, the service that calls other microservices. (Calls interfaces provided by other microservices)

Eureka Structure and Role#

The registry center is used to solve the problem of consumers calling providers that have multiple instances deployed.

• Question 1: How does order-service know the instance addresses of user-service?
  • After the user-service instance starts, it registers its information to the eureka-server (Eureka server). This is service registration.
  • The eureka-server maintains a mapping from service names to the list of instance addresses.
  • order-service pulls the instance address list by service name. This is service discovery or service pulling.
• Question 2: How does order-service pick a specific instance among multiple user-service instances?
  • order-service uses a load-balancing algorithm on the instance list to select an address.
  • It makes a remote call to that address.
• Question 3: How does order-service know whether a user-service instance is still healthy or down?
  • The user-service sends heartbeats to the eureka-server at regular intervals (default every 30 seconds) to report its status.
  • If heartbeats are not received for a certain period, the eureka-server marks the microservice instance as failed and removes it from the service list.
  • When order-service pulls the service list, faulty instances are excluded.

Building Eureka-Server#

1. Create the Eureka-Server service

Create a submodule under the parent project, a Maven project

1. Import Eureka dependency - server

1
<dependency>
2
    <groupId>org.springframework.cloud</groupId>
3
    <artifactId>spring-cloud-starter-netflix-eureka-server</artifactId>
4
</dependency>

1. Write the startup class

Add @EnableEurekaServer to enable Eureka registry center functionality

1
@SpringBootApplication
2
@EnableEurekaServer
3
public class EurekaApplication {
4
    public static void main(String[] args) {
5
        SpringApplication.run(EurekaApplication.class, args);
6
    }
7
}

1. Write the configuration file

1
server:
2
  port: 10086
3
spring:
4
  application:
5
    name: eureka-server
6
eureka:
7
  client:
8
    service-url:
9
      defaultZone: <http://127.0.0.1:10086/eureka>

1. Start the service

Access the configured Eureka service path

Service Registration#

1. Dependency import - client

1
<dependency>
2
    <groupId>org.springframework.cloud</groupId>
3
    <artifactId>spring-cloud-starter-netflix-eureka-client</artifactId>
4
</dependency>

1. Configuration file

1
spring:
2
  application:
3
    name: userservice
4
eureka:
5
  client:
6
    service-url:
7
      defaultZone: <http://127.0.0.1:10086/eureka>

1. Start multiple service instances

In SpringBoot, duplicate the startup configuration; configure -Dserver.port=8082 in VM options

Service Discovery#

As a consumer, order-service follows the same first two steps as the provider.

1. Service pulling and load balancing

Load balancing: simply add @LoadBalanced to the RestTemplate bean

1
@Bean
2
@LoadBalanced
3
public RestTemplate restTemplate(){
4
    return new RestTemplate();
5
}

Getting the service:

1
public Order queryOrderById(Long orderId) {
2
    // 1. Query the order
3
    Order order = orderMapper.findById(orderId);
4
    // 2. Remote query User
5
//        String url = "<http://localhost:8081/user/>" + order.getUserId();
6
    String url = "<http://userservice/user/>" + order.getUserId();
7
    User user = restTemplate.getForObject(url, User.class);
8
    // 3. Put into order
9
    order.setUser(user);
10
    // 4. Return
11
    return order;
12
}

Spring will automatically help us fetch the instance list from the eureka-server side based on the service name userservice, and then perform load balancing.

Ribbon Load Balancing#

1
@Bean
2
public IRule randomRule(){
3
    return new RandomRule();
4
}

1
userservice: # Configure load balancing rule for a specific microservice
2
  ribbon:
3
    NFLoadBalancerRuleClassName: com.netflix.loadbalancer.RandomRule # Load balancing rule

1
ribbon:
2
  eager-load:
3
    enabled: true
4
    clients: userservice

Services registration to Nacos#

1
<dependency>
2
    <groupId>com.alibaba.cloud</groupId>
3
    <artifactId>spring-cloud-alibaba-dependencies</artifactId>
4
    <version>2.2.6.RELEASE</version>
5
    <type>pom</type>
6
    <scope>import</scope>
7
</dependency>

1
<dependency>
2
    <groupId>com.alibaba.cloud</groupId>
3
    <artifactId>spring-cloud-starter-alibaba-nacos-discovery</artifactId>
4
</dependency>

1
spring:
2
  cloud:
3
    nacos:
4
      server-addr: localhost:8848

Service Hierarchical Storage Model#

A service can have multiple instances. If these instances are distributed across different data centers nationwide, Nacos groups the instances within the same data center into a cluster.

When microservices communicate with each other, they should try to access the same cluster’s instances as much as possible because local access is faster. If the current cluster is unavailable, access other clusters.

• Configuration

Modify application.yml to add cluster configuration:

1
spring:
2
  cloud:
3
    nacos:
4
      server-addr: localhost:8848
5
      discovery:
6
        cluster-name: HZ # cluster name

Nacos provides a NacosRule implementation that can prefer selecting instances from the same cluster.

1
userservice:
2
  ribbon:
3
    NFLoadBalancerRuleClassName: com.alibaba.cloud.nacos.ribbon.NacosRule # 负载均衡规则

• Weight configuration

In actual deployments, you may encounter situations like:

Servers have different hardware performance; some machines are more capable, others less so. You want the higher-performing machines to handle more user requests.

But by default, NacosRule selects randomly within the same cluster and does not consider machine performance.

Therefore, Nacos provides weight configuration to control access frequency; higher weight means higher access frequency.

• Environment isolation

Nacos provides namespaces to achieve environment isolation.

• nacos can have multiple namespaces

• within a namespace there can be groups, services, etc.

• different namespaces are isolated from each other; services in different namespaces are not visible to each other

• Service instances

Nacos service instances come in two types:

• Ephemeral (temporary): If an instance goes down for a period, it will be removed from the service list; this is the default type.
• Non-ephemeral: If an instance goes down, it will not be removed from the service list and can be considered permanent.

Configure a service instance as non-ephemeral:

1
spring:
2
  cloud:
3
    nacos:
4
      discovery:
5
        ephemeral: false # set to non-ephemeral

Differences Between Nacos and Eureka#

Nacos and Eureka have similar overall structures—service registration, service pulling, and heartbeat waiting—but there are some differences:

• Common points
  • Both support service registration and service pulling
  • Both support heartbeat-based health checks for service providers
• Differences
  • Nacos supports server-side active monitoring of provider status: ephemeral instances use heartbeat mode; non-ephemeral instances use active checks
  • Ephemeral instances with abnormal heartbeats will be removed; non-ephemeral instances will not be removed
  • Nacos supports push-based notifications for service list changes, making updates more timely
  • Nacos clusters default to AP; when non-ephemeral instances exist, CP mode is used; Eureka uses AP mode

Pulling Configuration from Microservices#

Microservices pull configurations managed in Nacos and merge them with local application.yml configurations to complete startup.

But if application.yml has not yet been read, how to determine the Nacos address?

Therefore Spring introduces a new configuration file: bootstrap.yaml, which is read before application.yml.

1. Importing nacos-config dependency

1
<!--nacos配置管理依赖-->
2
<dependency>
3
    <groupId>com.alibaba.cloud</groupId>
4
    <artifactId>spring-cloud-starter-alibaba-nacos-config</artifactId>
5
</dependency>

1. Add bootstrap.yml

1
spring:
2
  application:
3
    name: userservice # service name
4
  profiles:
5
    active: dev # development environment; here it's dev
6
  cloud:
7
    nacos:
8
      server-addr: localhost:8848 # Nacos address
9
      config:
10
        file-extension: yaml # file extension

Here the Nacos address will be obtained from spring.cloud.nacos.server-addr, and then

${spring.application.name}-${spring.profiles.active}.${spring.cloud.nacos.config.file-extension} is used as the file id to read configurations.

1. Reading Nacos configuration

1
@Value("${pattern.dateformat}")
2
private String dateformat;
3

4
@GetMapping("now")
5
public String now(){
6
    return LocalDateTime.now().format(DateTimeFormatter.ofPattern(dateformat));
7
}

Configuration Hot Update#

After changing configurations in Nacos, microservices can apply the changes without restarting, i.e., configuration hot updating.

• Method 1: add @RefreshScope annotation

• Method 2

  Use @ConfigurationProperties instead of @Value annotations.

  1. In the user-service, add a class to read the pattern.dateformat property:

  1
  package cn.itcast.user.config;
  2
  
  3
  import lombok.Data;
  4
  import org.springframework.boot.context.properties.ConfigurationProperties;
  5
  import org.springframework.stereotype.Component;
  6
  
  7
  @Component
  8
  @Data
  9
  @ConfigurationProperties(prefix = "pattern")
  10
  public class PatternProperties {
  11
      private String dateformat;
  12
  }

  1. In UserController, use this class instead of @Value:

  1
  @Autowired
  2
  private PatternProperties patternProperties;
  3
  
  4
  @GetMapping("now")
  5
  public String now(){
  6
      return LocalDateTime.now().format(DateTimeFormatter.ofPattern(patternProperties.getDateformat()));
  7
  }

Configuration Sharing#

In fact, when the microservice starts, it will read multiple configuration files from Nacos, for example:

• [spring.application.name]-[spring.profiles.active].yaml, e.g., userservice-dev.yaml
• [spring.application.name].yaml, e.g., userservice.yaml

And [spring.application.name].yaml does not include an environment, so it can be shared across environments.

Configuration sharing priority:

When the same property appears in both Nacos and the service’s local configuration, the priority differs:

Nacos Clusters#

Proxy multiple Nacos servers via Nginx

Configure cluster.conf in Nacos’s conf, adding Nacos cluster addresses.

Feign Replaces RestTemplate#

1. Import dependencies

   1
   <dependency>
   2
       <groupId>org.springframework.cloud</groupId>
   3
       <artifactId>spring-cloud-starter-openfeign</artifactId>
   4
   </dependency>

2. Add annotation

   Add @EnableFeignClients in the startup class

3. Create Feign client

   In the service, create an interface:

   1
   @FeignClient("userservice")
   2
   public interface UserClient {
   3
       @GetMapping("/user/{id}")
   4
       User findById(@PathVariable("id") Long id);
   5
   }

   This client is mainly based on Spring MVC annotations to declare remote call details, such as:

   • Service name: userservice
   • HTTP method: GET
   • Path: /user/{id}
   • Parameter: Long id
   • Return type: User

   In this way, Feign can help us send HTTP requests without manually using RestTemplate.

4. Usage

   In the service class, call the method:

   1
   @Autowired
   2
   private UserClient userClient;
   3
   
   4
   public Order queryOrderById(Long orderId) {
   5
       // 1. Query the order
   6
       Order order = orderMapper.findById(orderId);
   7
       // 2. Remote query User
   8
       User user = userClient.findById(order.getUserId());
   9
       // 3. Put into order
   10
       order.setUser(user);
   11
       // 4. Return
   12
       return order;
   13
   }

Custom Configuration#

Feign supports a lot of custom configurations, as shown in the table below:

Generally, the defaults are sufficient; if you want to customize, simply create your own @Bean to override the default beans.

1
feign:
2
  client:
3
    config:
4
      userservice: # configuration for a specific microservice
5
        loggerLevel: FULL #  log level

1
feign:
2
  client:
3
    config:
4
      default: # default is global; if you specify a service name, it's for a single microservice
5
        loggerLevel: FULL #  log level

1
public class DefaultFeignConfiguration  {
2
    @Bean
3
    public Logger.Level feignLogLevel(){
4
        return Logger.Level.BASIC; // log level is BASIC
5
    }
6
}

1
@EnableFeignClients(defaultConfiguration = DefaultFeignConfiguration .class)

1
@FeignClient(value = "userservice", configuration = DefaultFeignConfiguration .class)

Feign Usage Optimization#

Feign’s underlying HTTP requests rely on other frameworks. Their underlying client implementations include:

• URLConnection: default implementation, does not support connection pooling
• Apache HttpClient: supports connection pooling
• OKHttp: supports connection pooling

Thus, the main way to improve Feign performance is to use a connection pool instead of the default URLConnection.

Here we demonstrate using Apache HttpClient.

1. Import dependencies

In the order-service pom, include Apache HttpClient dependency:

1
<!--httpClient dependency -->
2
<dependency>
3
    <groupId>io.github.openfeign</groupId>
4
    <artifactId>feign-httpclient</artifactId>
5
</dependency>

1. Configure the connection pool

In the order-service application.yml, add configuration:

1
feign:
2
  client:
3
    config:
4
      default: # global default
5
        loggerLevel: BASIC # log level; BASIC is the basic request/response info
6
  httpclient:
7
    enabled: true # enable Feign's support for HttpClient
8
    max-connections: 200 # maximum connections
9
    max-connections-per-route: 50 # maximum connections per route

In short, Feign optimization:

1. Prefer BASIC log level when possible
2. Use HttpClient or OKHttp instead of URLConnection
   • Include feign-httpClient dependency
   • Enable HttpClient in the configuration and set pool parameters

Best Practice#

The Feign client and the service provider’s controller code are very similar. Here is a method to reduce repetitive coding:

Why a Gateway is Needed#

The Gateway is the gatekeeper of our services, the single entry point for all microservices.

Core capabilities of the gateway:

• Access control: The gateway, as the entry to microservices, should verify whether the user has permission to request; if not, intercept.
• Routing and load balancing: All requests must pass through the gateway, but the gateway does not process business logic. It forwards requests to a microservice based on certain rules; when there are multiple target services, load balancing is required.
• Rate limiting: When request traffic is high, the gateway releases requests at a rate that downstream microservices can handle, preventing service pressure.

In SpringCloud, there are two gateway implementations:

• gateway
• zuul

Zuul is servlet-based (blocking). SpringCloudGateway is built on WebFlux provided by Spring 5, a reactive programming implementation with better performance.

Quick Start#

1. Create a SpringBoot project gateway and include gateway dependencies

   1
   <!--网关-->
   2
   <dependency>
   3
       <groupId>org.springframework.cloud</groupId>
   4
       <artifactId>spring-cloud-starter-gateway</artifactId>
   5
   </dependency>
   6
   <!--nacos服务发现依赖-->
   7
   <dependency>
   8
       <groupId>com.alibaba.cloud</groupId>
   9
       <artifactId>spring-cloud-starter-alibaba-nacos-discovery</artifactId>
   10
   </dependency>

2. Write the startup class

   1
   @SpringBootApplication
   2
   public class GatewayApplication {
   3
   
   4
     public static void main(String[] args) {
   5
       SpringApplication.run(GatewayApplication.class, args);
   6
     }
   7
   }

3. Write basic configuration and routing rules

   Routing configuration includes:

   1. Route id: the unique identifier of the route
   2. Route target (uri): the route’s target address; http represents a fixed address, lb represents load balancing by service name
   3. Route predicates: the rules to determine routing
   4. Route filters: process requests or responses

   1
   server:
   2
     port: 10010 # gateway port
   3
   spring:
   4
     application:
   5
       name: gateway # service name
   6
     cloud:
   7
       nacos:
   8
         server-addr: localhost:8848 # nacos address
   9
       gateway:
   10
         routes: # gateway route configuration
   11
           - id: user-service # route id, unique
   12
             # uri: <http://127.0.0.1:8081> # route target address; http is fixed
   13
             uri: lb://userservice # route target address; lb indicates load balancing, followed by service name
   14
             predicates: # route predicates, i.e., conditions that determine routing
   15
               - Path=/user/** # path-based matching; matches any path that starts with /user/

4. Start the gateway service for testing

Restart the gateway; visiting http://localhost:10010/user/1 matches the /user/** rule and the request is forwarded to uri: http://userservice/user/1, yielding the result:

Predicate Factory#

What we write in the configuration as assertion rules are strings; these strings are read and processed by Predicate Factory to become route decision conditions.

For example, Path=/user/** matches by path, and this rule is handled by the class org.springframework.cloud.gateway.handler.predicate.PathRoutePredicateFactory. There are a dozen or so such assertion factories in SpringCloudGateway:

Filter Factory#

GatewayFilter is a filter provided by the gateway that can process incoming requests and responses from microservices

• Process requests or responses for routes, such as adding headers
• Filters defined under a route only apply to the current route’s requests

1
spring:
2
  cloud:
3
    gateway:
4
      routes:
5
      - id: user-service
6
        uri: lb://userservice
7
        predicates:
8
        - Path=/user/**
9
        filters: # filters
10
        - AddRequestHeader=Truth, dreaife is freaking awesome! # add header

1
spring:
2
  cloud:
3
    gateway:
4
      routes:
5
      - id: user-service
6
        uri: lb://userservice
7
        predicates:
8
        - Path=/user/**
9
      default-filters: # default filters
10
      - AddRequestHeader=Truth, dreaife is freaking awesome!

Global Filters#

1
public interface GlobalFilter {
2
    /**
3
     *  Process the current request; if necessary, pass the request to the next filter
4
     *
5
     * @param exchange the request context, where you can access Request, Response, etc.
6
     * @param chain used to delegate the request to the next filter
7
     * @return {@code Mono<Void>} indicates the end of this filter’s processing
8
     */
9
    Mono<Void> filter(ServerWebExchange exchange, GatewayFilterChain chain);
10
}

1
package cn.itcast.gateway.filters;
2

3
import org.springframework.cloud.gateway.filter.GatewayFilterChain;
4
import org.springframework.cloud.gateway.filter.GlobalFilter;
5
import org.springframework.core.annotation.Order;
6
import org.springframework.http.HttpStatus;
7
import org.springframework.stereotype.Component;
8
import org.springframework.web.server.ServerWebExchange;
9
import reactor.core.publisher.Mono;
10

11
@Order(-1)
12
@Component
13
public class AuthorizeFilter implements GlobalFilter {
14
    @Override
15
    public Mono<Void> filter(ServerWebExchange exchange, GatewayFilterChain chain) {
16
        // 1. Get request parameters
17
        MultiValueMap<String, String> params = exchange.getRequest().getQueryParams();
18
        // 2. Get authorization parameter
19
        String auth = params.getFirst("authorization");
20
        // 3. Validate
21
        if ("admin".equals(auth)) {
22
            // Let through
23
            return chain.filter(exchange);
24
        }
25
        // 4. Block
26
        // 4.1. Deny access, set status code
27
        exchange.getResponse().setStatusCode(HttpStatus.FORBIDDEN);
28
        // 4.2. End handling
29
        return exchange.getResponse().setComplete();
30
    }
31
}

Cross-Origin (CORS) Issues#

Cross-origin means different domains; examples include:

• Different domains: http://www.taobao.com/ vs http://www.taobao.org/ vs http://www.jd.com/ vs http://miaosha.jd.com/
• Same domain, different ports: localhost:8080 vs localhost:8081

Cross-origin issues occur when the browser blocks cross-origin AJAX requests initiated by the client.

Solution: CORS

In the gateway service’s application.yml, add the following configuration:

1
spring:
2
  cloud:
3
    gateway:
4
      globalcors: # Global cross-origin handling
5
        add-to-simple-url-handler-mapping: true # Solve options requests being blocked
6
        corsConfigurations:
7
          '[/**]':
8
            allowedOrigins: # Which websites are allowed to make cross-origin requests
9
              - "<http://localhost:8090>"
10
            allowedMethods: # Allowed cross-origin AJAX request methods
11
              - "GET"
12
              - "POST"
13
              - "DELETE"
14
              - "PUT"
15
              - "OPTIONS"
16
            allowedHeaders: "*" # Headers that are allowed in requests
17
            allowCredentials: true # Whether to allow credentials (cookies)
18
            maxAge: 360000 # This cross-origin check's validity period

モノリシックアーキテクチャ#

ビジネスのすべての機能を1つのプロジェクトに集中して開発し、1つのパッケージとしてデプロイする。

利点：

• アーキテクチャが単純
• デプロイコストが低い

欠点：

• 結合度が高い（保守が難しい、アップグレードが難しい）

分散アーキテクチャ#

ビジネス機能に基づいてシステムを分割し、各ビジネス機能モジュールを独立したプロジェクトとして開発します。これを1つのサービスと呼ぶ。

利点：

• サービス間の結合を低減
• サービスのアップグレードと拡張に有利

欠点：

• サービス呼び出し関係が複雑になる

分散アーキテクチャは確かにサービスの結合を低減しますが、サービスを分割する際には考えるべき問題が多くあります：

• サービス分割の粒度をどう定義するか？
• サービス間はどう呼び出すのか？
• サービスの呼び出し関係をどう管理するのか？

人々は分散アーキテクチャを制約するための実効的な標準を制定する必要があります。

マイクロサービス#

マイクロサービスのアーキテクチャの特徴：

• 単一責務：マイクロサービスは粒度がより小さく、各サービスが唯一のビジネス能力を持ち、単一責務を実現します
• 自治：チーム独立、技術独立、データ独立、独立デプロイとデリバリー
• サービス指向：サービスは統一された標準インターフェースを提供し、言語や技術に依存しない
• 隔離性が高い：サービス呼び出しは隔離・フォールトトレランス・降格を適切に実施し、連鎖的な問題を回避

上述の特性は、分散アーキテクチャに対して標準を設定し、サービス間の結合をさらに低減し、サービスの独立性と柔軟性を提供することを目的としています。高い内聚、低い結合を実現します。

したがって、マイクロサービスは、良好なアーキテクチャ設計を経た、分散アーキテクチャのソリューションとみなすことができます。

SpringCloud#

SpringCloudは現在国内で最も広く使われているマイクロサービスフレームワークです。公式サイト：https://spring.io/projects/spring-cloud。

SpringCloudはさまざまなマイクロサービス機能コンポーネントを統合し、SpringBootを基盤としてこれらのコンポーネントの自動設定を実現することで、良好な「箱から出してすぐ使える」体験を提供します。

コンポーネントには：

• サービス登録・発見：Eureka、Nacos、Consul
• サービスリモート呼び出し：OpenFeign、Dubbo
• サービストレースモニタリング：Zipkin、Sleuth
• 統一設定管理：SpringCloudConfig、Nacos
• 統一ゲートウェイルーティング：SpringCloudGateway、Zuul
• レート制御、降格、保護：Hystix、Sentinel

サービス間の遠隔呼び出し#

RestTemplateを呼び出して実現します

1. 起動時にRestTemplateのインスタンスを登録する

   1
   @Bean
   2
   public RestTemplate restTemplate() {
   3
       return new RestTemplate();
   4
   }

2. サービス側の呼び出し

   1
   @Autowired
   2
   private RestTemplate restTemplate;
   3
   
   4
   public Order queryOrderById(Long orderId) {
   5
       // 1.注文を取得
   6
       Order order = orderMapper.findById(orderId);
   7
       // 2. ユーザーをリモート照会
   8
       String url = "<http://localhost:8081/user/>" + order.getUserId();
   9
       User user = restTemplate.getForObject(url, User.class);
   10
       // 3. orderに格納
   11
       order.setUser(user);
   12
       // 4. 返却
   13
       return order;
   14
   }

提供者と消費者#