/gs-client-side-load-balancing

Client Side Load Balancing with Ribbon and Spring Cloud :: Dynamically support services coming up and going down without interrupting the client

Primary LanguageShell

tags projects
spring-cloud
ribbon
spring-cloud
spring-cloud-netflix

This guide walks you through the process of providing client-side load balancing for a microservice application using Netflix Ribbon.

What you’ll build

You’ll build a microservice application that uses Netflix Ribbon and Spring Cloud Netflix to provide client-side load balancing in calls to another microservice.

What you’ll need

Build with Gradle

Build with Gradle

say-hello/build.gradle

link:initial/say-hello/build.gradle[role=include]

user/build.gradle

link:initial/user/build.gradle[role=include]

Build with Maven

Build with Maven

say-hello/pom.xml

link:initial/say-hello/pom.xml[role=include]

user/pom.xml

link:initial/user/pom.xml[role=include]

Write a server service

Our “server” service is called Say Hello. It will return a random greeting (picked out of a static list of three) from an endpoint accessible at /greeting.

In src/main/java/hello, create the file SayHelloApplication.java. It should look like this:

say-hello/src/main/java/hello/SayHelloApplication.java

link:complete/say-hello/src/main/java/hello/SayHelloApplication.java[role=include]

The @RestController annotation gives the same effect as if we were using @Controller and @ResponseBody together. It marks SayHelloApplication as a controller class (which is what @Controller does) and ensures that return values from the class’s @RequestMapping methods will be automatically converted appropriately from their original types and written directly to the response body (which is what @ResponseBody does). We have one @RequestMapping method for /greeting and then another for the root path /. (We’ll want that second method when we get to working with Ribbon in just a bit.)

We’re going to run multiple instances of this application locally alongside a client service application, so create the directory src/main/resources, create the file application.yml within it, and then in that file, set a default value for server.port. (We’ll instruct the other instances of the application to run on other ports, as well, so that none of the Say Hello instances will conflict with the client when we get that running.) While we’re in this file, we’ll set the spring.application.name for our service too.

say-hello/src/main/resources/application.yml

link:complete/say-hello/src/main/resources/application.yml[role=include]

Access from a client service

The User application will be what our user sees. It will make a call to the Say Hello application to get a greeting and then send that to our user when the user visits the endpoint at /hi.

In the User application directory, under src/main/java/hello, add the file UserApplication.java:

user/src/main/java/hello/UserApplication.java

package hello;

import org.springframework.boot.SpringApplication;
import org.springframework.boot.autoconfigure.SpringBootApplication;
import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.context.annotation.Bean;
import org.springframework.web.bind.annotation.RequestMapping;
import org.springframework.web.bind.annotation.RequestParam;
import org.springframework.web.bind.annotation.RestController;
import org.springframework.web.client.RestTemplate;

@SpringBootApplication
@RestController
public class UserApplication {

  @Bean
  RestTemplate restTemplate(){
    return new RestTemplate();
  }

  @Autowired
  RestTemplate restTemplate;

  @RequestMapping("/hi")
  public String hi(@RequestParam(value="name", defaultValue="Artaban") String name) {
    String greeting = this.restTemplate.getForObject("http://localhost:8090/greeting", String.class);
    return String.format("%s, %s!", greeting, name);
  }

  public static void main(String[] args) {
    SpringApplication.run(UserApplication.class, args);
  }
}

To get a greeting from Say Hello, we’re using Spring’s RestTemplate template class. RestTemplate makes an HTTP GET request to the Say Hello service’s URL as we provide it and gives us the result as a String. (For more information on using Spring to consume a RESTful service, see the Consuming a RESTful Web Service guide.)

Add the spring.application.name and server.port properties to src/main/resources/application.properties or src/main/resources/application.yml:

user/src/main/resources/application.yml

spring:
  application:
    name: user

server:
  port: 8888

Load balance across server instances

Now we can access /hi on the User service and see a friendly greeting:

$ curl http://localhost:8888/hi
Greetings, Artaban!

$ curl http://localhost:8888/hi?name=Orontes
Salutations, Orontes!

To move beyond a single hard-coded server URL to a load-balanced solution, let’s set up Ribbon. In the application.yml file under user/src/main/resources/, add the following properties:

user/src/main/resources/application.yml

say-hello:
  ribbon:
    eureka:
      enabled: false
    listOfServers: localhost:8090,localhost:9092,localhost:9999
    ServerListRefreshInterval: 15000

This configures properties on a Ribbon client. Spring Cloud Netflix creates an ApplicationContext for each Ribbon client name in our application. This is used to give the client a set of beans for instances of Ribbon components, including:

  • an IClientConfig, which stores client configuration for a client or load balancer,

  • an ILoadBalancer, which represents a software load balancer,

  • a ServerList, which defines how to get a list of servers to choose from,

  • an IRule, which describes a load balancing strategy, and

  • an IPing, which says how periodic pings of a server are performed.

In our case above, the client is named say-hello. The properties we set are eureka.enabled (which we set to false), listOfServers, and ServerListRefreshInterval. Load balancers in Ribbon normally get their server lists from a Netflix Eureka service registry. (See the Service Registration and Discovery guide for information on using a Eureka service registry with Spring Cloud.) For our simple purposes here, we’re skipping Eureka, so we set the ribbon.eureka.enabled property to false and instead give Ribbon a static listOfServers. ServerListRefreshInterval is the interval, in milliseconds, between refreshes of Ribbon’s service list.

In our UserApplication class, switch the RestTemplate to use the Ribbon client to get the server address for Say Hello:

user/src/main/java/hello/UserApplication.java

link:complete/user/src/main/java/hello/UserApplication.java[role=include]

We’ve made a couple of other related changes to the UserApplication class. Our RestTemplate is now also marked as LoadBalanced; this tells Spring Cloud that we want to take advantage of its load balancing support (provided, in this case, by Ribbon). The class is annotated with @RibbonClient, which we give the name of our client (say-hello) and then another class, which contains extra configuration for that client.

We’ll need to create that class. Add a new file, SayHelloConfiguration.java, in the user/src/main/java/hello directory:

user/src/main/java/hello/SayHelloConfiguration.java

link:complete/user/src/main/java/hello/SayHelloConfiguration.java[role=include]

We can override any Ribbon-related bean that Spring Cloud Netflix gives us by creating our own bean with the same name. Here, we override the IPing and IRule used by the default load balancer. The default IPing is a NoOpPing (which doesn’t actually ping server instances, instead always reporting that they’re stable), and the default IRule is a ZoneAvoidanceRule (which avoids the Amazon EC2 zone that has the most malfunctioning servers, and might thus be a bit difficult to try out in our local environment).

Our IPing is a PingUrl, which will ping a URL to check the status of each server. Say Hello has, as you’ll recall, a method mapped to the / path; that means that Ribbon will get an HTTP 200 response when it pings a running Say Hello server. The IRule we set up, the AvailabilityFilteringRule, will use Ribbon’s built-in circuit breaker functionality to filter out any servers in an “open-circuit” state: if a ping fails to connect to a given server, or if it gets a read failure for the server, Ribbon will consider that server “dead” until it begins to respond normally.

Note

The @SpringBootApplication annotation on the UserApplication class is equivalent to (among others) the @Configuration annotation that marks a class as a source of bean definitions. This is why we don’t need to annotate the SayHelloConfiguration class with @Configuration: since it’s in the same package as UserApplication, it is already being scanned for bean methods.

This approach does mean that our Ribbon configuration will be part of the main application context and therefore shared by all Ribbon clients in the User application. In a normal application, you can avoid this by keeping Ribbon beans out of the main application context (e.g., in this example, you could put SayHelloConfiguration in a different package from UserApplication).

Trying it out

Run the Say Hello service, using either Gradle:

$ ./gradlew bootRun

or Maven:

$ mvn spring-boot:run

Run other instances on ports 9092 and 9999, again using either Gradle:

$ SERVER_PORT=9092 ./gradlew bootRun

or Maven:

$ SERVER_PORT=9999 mvn spring-boot:run

Then start up the User service. Access localhost:8888/hi and then watch the Say Hello service instances. You can see Ribbon’s pings arriving every 15 seconds:

2016-03-09 21:13:22.115  INFO 90046 --- [nio-8090-exec-1] hello.SayHelloApplication                : Access /
2016-03-09 21:13:22.629  INFO 90046 --- [nio-8090-exec-3] hello.SayHelloApplication                : Access /

And your requests to the User service should result in calls to Say Hello being spread across the running instances in round-robin form:

2016-03-09 21:15:28.915  INFO 90046 --- [nio-8090-exec-7] hello.SayHelloApplication                : Access /greeting

Now shut down a Say Hello server instance. Once Ribbon has pinged the down instance and considers it down, you should see requests begin to be balanced across the remaining instances.

Summary

Congratulations! You’ve just developed a Spring application that performs client-side load balancing for calls to another application.