Circuit Breaker, Fallback and Load Balancing with Apache Camel

Apache Camel has just released a new version of their framework – 2.19. In one of my previous articles on DZone I described details about microservices support which was released in the Camel 2.18 version. There are some new features in ServiceCall EIP component, which is responsible for microservice calls. You can see example source code which is based on the sample from my article on DZone. It is available on GitHub under new branch fallback.

In the code fragment below you can see DLS route’s configuration with support for Hystrix circuit breaker, Ribbon load balancer and Consul service discovery and registration. As a service discovery in the route definition you can also use some other solutions instead of Consul like etcd (etcServiceDiscovery) or Kubernetes (kubernetesServiceDiscovery).

from("direct:account")
	.to("bean:customerService?method=findById(${header.id})")
	.log("Msg: ${body}").enrich("direct:acc", new AggregationStrategyImpl());

from("direct:acc").setBody().constant(null)
	.hystrix()
		.hystrixConfiguration()
			.executionTimeoutInMilliseconds(2000)
		.end()
	.serviceCall()
		.name("account//account")
		.component("netty4-http")
		.ribbonLoadBalancer("ribbon-1")
		.consulServiceDiscovery("http://192.168.99.100:8500")
	.end()
	.unmarshal(format)
	.endHystrix()
	.onFallback()
	.to("bean:accountFallback?method=getAccounts");

We can easily configure all Hystrix’s parameters just by calling hystrixConfiguration method. In the sample above Hystrix waits max 2 seconds for the response from remote service. In case of timeout fallback @Bean is called. Fallback @Bean implementation is really simple – it return empty list.

@Service
public class AccountFallback {

	public List<Account> getAccounts() {
		return new ArrayList<>();
	}

}

Alternatively, configuration can be implemented using object delarations. Here is service call configuration with Ribbon and Consul. Additionally, we can provide some parameters to Ribbon like client read timeout or max retry attempts. Unfortunately it seems they doesn’t work in this version of Apache Camel 🙂 (you can try to test it by yourself). I hope this will be corrected soon.

ServiceCallConfigurationDefinition def = new ServiceCallConfigurationDefinition();

ConsulConfiguration config = new ConsulConfiguration();
config.setUrl("http://192.168.99.100:8500");
config.setComponent("netty4-http");
ConsulServiceDiscovery discovery = new ConsulServiceDiscovery(config);

RibbonConfiguration c = new RibbonConfiguration();
c.addProperty("MaxAutoRetries", "0");
c.addProperty("MaxAutoRetriesNextServer", "1");
c.addProperty("ReadTimeout", "1000");
c.setClientName("ribbon-1");
RibbonServiceLoadBalancer lb = new RibbonServiceLoadBalancer(c);
lb.setServiceDiscovery(discovery);

def.setComponent("netty4-http");
def.setLoadBalancer(lb);
def.setServiceDiscovery(discovery);
context.setServiceCallConfiguration(def);

I described similar case for Spring Cloud and Netflix OSS in one of my previous article. Just like in the example presented there, I also set here a delay inside account service, which depends on the port on which the microservice was started.

@Value("${port}")
private int port;

public List<Account> findByCustomerId(Integer customerId) {
	List<Account> l = new ArrayList<>();
	l.add(new Account(1, "1234567890", 4321, customerId));
	l.add(new Account(2, "1234567891", 12346, customerId));
	if (port%2 == 0) {
		try {
			Thread.sleep(5000);
		} catch (InterruptedException e) {
			e.printStackTrace();
		}
	}
	return l;
}

Results for Spring Cloud sample were much more satisfying. The introduced configuration parameters such as read timeout for Ribbon worked and in addition Hystrix was able to automatically redirect a much smaller number of requests to slow service – only 2% of the rest to the non-blocking thread instance for 5 seconds. This shows that Apache Camel still has a few things to improve if wants to compete in microservice’s support with Sprint Cloud framework.

Spring Cloud Microservices at Pivotal Platform

Imagine you have multiple microservices running on different machines as multiple instances. It seems natural to think about the tools that helps you in the process of monitoring and managing all of them. If we add that our microservices are created based on the Spring Cloud framework obviously seems we should look at the Pivotal platform. Here is figure with platform’s architecture download from the main Pivotal’s site.

PVDI-Microservices-Architecture

Although Pivotal Platform can run applications written in many languages it has the best support for Spring Cloud Services and Netflix OSS tools like you can see in the figure above. From the possibilities offered by Pivotal we can take advantage of three ways.

Pivotal Cloud Foundry – solution can be ran on public IaaS or private cloud like AWS, Google Cloud Platform, Microsoft Azure, VMware vSphere, OpenStack.

Pivotal Web Services – hosted cloud-native platform available at pivotal.io site.

PCF Dev – the instance which can be run locally as a a single virtual machine. It offers the opportunity to develop apps using an offline environment which basic services installed like Spring Cloud Services (SCS), MySQL, Redis databases and RabbitMQ broker. If you want to run it locally with SCS you need more than 6GB RAM free.

As a Spring Cloud Services there are available Circuit Breaker (Hystrix), Service Registry (Eureka) and standard Spring Configuration Server based on git configuration.

scs

That’s all I wanted to say about the theory. Let’s move on to practice. On the Pivotal website we have detailed materials on how to set it up, create and deploy a simple microservice based on Spring Cloud solutions. In this article I will try to present the essence collected from these descriptions based on one of my standard examples from the previous posts. As always sample source code is available on GitHub. If you are interested in detailed description of the sample application, microservices and Spring Cloud read my previous articles:

Part 1: Creating microservice using Spring Cloud, Eureka and Zuul

Part 3: Creating Microservices: Circuit Breaker, Fallback and Load Balancing with Spring Cloud

If you have a lot of free RAM you can install PCF Dev on your local workstation. You need to have Virtual Box installed. Then download and install Cloud Foundry Command Line Interface (CF CLI) and PCF Dev. All is described here. Finally you can run command below and take a small break for coffee. Virtual machine needs to downloaded and started.

cf dev start -s scs

For those who do not have RAM enough (like me) there is Pivotal Web Services platform. It is available here. Before use it you have to register on Pivotal site. The rest of the article is identical for both options.
In comparison to previous examples of Spring Cloud based microservices, we need to make some changes. There is one additional dependency inside every microservice’s pom.xml.

<properties>
	...
	<spring-cloud-services.version>1.4.1.RELEASE</spring-cloud-services.version>
	<spring-cloud.version>Dalston.RELEASE</spring-cloud.version>
</properties>

<dependencies>
	<dependency>
		<groupId>io.pivotal.spring.cloud</groupId>
		<artifactId>spring-cloud-services-starter-service-registry</artifactId>
	</dependency>
	...
</dependencies>

<dependencyManagement>
	<dependencies>
		<dependency>
			<groupId>org.springframework.cloud</groupId>
			<artifactId>spring-cloud-dependencies</artifactId>
			<version>${spring-cloud.version}</version>
			<type>pom</type>
			<scope>import</scope>
		</dependency>
		<dependency>
			<groupId>io.pivotal.spring.cloud</groupId>
			<artifactId>spring-cloud-services-dependencies</artifactId>
			<version>${spring-cloud-services.version}</version>
			<type>pom</type>
			<scope>import</scope>
		</dependency>
	</dependencies>
</dependencyManagement>

We also use Maven Cloud Foundry plugin cf-maven-plugin for application deployment on Pivotal platform. Here is sample for account-service. We run two instances of that microservice with max memory 512MB. Our application name is piomin-account-service.

<plugin>
	<groupId>org.cloudfoundry</groupId>
	<artifactId>cf-maven-plugin</artifactId>
	<version>1.1.3</version>
	<configuration>
		<target>http://api.run.pivotal.io</target>
		<org>piotrminkowski</org>
		<space>development</space>
		<appname>piomin-account-service</appname>
		<memory>512</memory>
		<instances>2</instances>
		<server>cloud-foundry-credentials</server>
	</configuration>
</plugin>

Don’t forget to add credentials configuration into Maven settings.xml file.

<server>
	<id>cloud-foundry-credentials</id>
	<username>piotr.minkowski@gmail.com</username>
	<password>***</password>
</server>

Now, when building sample application we to append cf:push command.

mvn clean install cf:push

Here is circuit breaker implementation inside customer-service.

@Service
public class AccountService {

	@Autowired
	private AccountClient client;

	@HystrixCommand(fallbackMethod = "getEmptyList")
	public List<Account> getAccounts(Integer customerId) {
		return client.getAccounts(customerId);
	}

	List<Account> getEmptyList(Integer customerId) {
		return new ArrayList<>();
	}

}

There is randomly generated delay on the account’s service side, so 25% of calls circuit breaker should be activated.

@RequestMapping("/accounts/customer/{customer}")
public List<Account> findByCustomer(@PathVariable("customer") Integer customerId) {
	logger.info(String.format("Account.findByCustomer(%s)", customerId));
	Random r = new Random();
	int rr = r.nextInt(4);
	if (rr == 1) {
		try {
			Thread.sleep(2000);
		} catch (InterruptedException e) {
			e.printStackTrace();
		}
	}
	return accounts.stream().filter(it -> it.getCustomerId().intValue() == customerId.intValue())
		.collect(Collectors.toList());
}

After successfully deploying application using Maven cf:push command we can go to Pivotal Web Services console available at https://console.run.pivotal.io/. Here are our two deployed services: two instances of piomin-account-service and one instance of piomin-customer-service.

pivotal-1

I have also activated Circuit Breaker and Service Registry from Marketplace.

pivotal-2

Every application need to be bound to service. To enable it select service, then expand Bound Apps overlap and select checkbox next to each service name.

pivotal-4

After this step applications needs to be restarted. It also can be be using web dashboard inside each service.

pivotal-5

Finally, all services are registered in Eureka and we can perform some tests using customer endpoint https://piomin-customer-service.cfapps.io/customers/{id}.

pivotal-4

Final words

With Pivotal solution we can easily deploy, scale and monitor our microservices. Deployment and scaling can be done using Maven plugin or via web dashboard. On Pivotal there are also available some services prepared especially for microservices needs like service registry, circuit breaker and configuration server. Pivotal is a competition for such solutions like Kubernetes which based on Docker containerization (more about this tools here). It is especially useful if you are creating a microservices based on Spring Boot and Spring Cloud frameworks.

Part 3: Creating Microservices: Circuit Breaker, Fallback and Load Balancing with Spring Cloud

Probably you read some articles about Hystrix and you know in what purpose it is used for. Today I would like to show you an example of exactly how to use it, which gives you the ability to combine with other tools from Netflix OSS stack like Feign and Ribbon. In this I assume that you have basic knowledge on topics such as microservices, load balancing, service discovery. If not I suggest you read some articles about it, for example my short introduction to microservices architecture available here: Part 1: Creating microservice using Spring Cloud, Eureka and Zuul. The code sample used in that article is also also used now. There is also sample source code available on GitHub. For the sample described now see hystrix branch, for basic sample master branch. 

Let’s look at some scenarios for using fallback and circuit breaker. We have Customer Service which calls API method from Account Service. There two running instances of Account Service. The requests to Account Service instances are load balanced by Ribbon client 50/50.

micro-details-1

Scenario 1

Hystrix is disabled for Feign client (1), auto retries mechanism is disabled for Ribbon client on local instance (2) and other instances (3). Ribbon read timeout is shorter than request max process time (4). This scenario also occurs with the default Spring Cloud configuration without Hystrix. When you call customer test method you sometimes receive full response and sometimes 500 HTTP error code (50/50).

ribbon:
  eureka:
    enabled: true
  MaxAutoRetries: 0 #(2)
  MaxAutoRetriesNextServer: 0 #(3)
  ReadTimeout: 1000 #(4)

feign:
  hystrix:
    enabled: false #(1)

Scenario 2

Hystrix is still disabled for Feign client (1), auto retries mechanism is disabled for Ribbon client on local instance (2) but enabled on other instances once (3). You always receive full response. If your request is received by instance with delayed response it is timed out after 1 second and then Ribbon calls another instance – in that case not delayed. You can always change MaxAutoRetries to positive value but gives us nothing in that sample.

ribbon:
  eureka:
    enabled: true
  MaxAutoRetries: 0 #(2)
  MaxAutoRetriesNextServer: 1 #(3)
  ReadTimeout: 1000 #(4)

feign:
  hystrix:
    enabled: false #(1)

Scenario 3

Here is not a very elegant solution to the problem. We set ReadTimeout on value bigger than delay inside API method (5000 ms).

ribbon:
  eureka:
    enabled: true
  MaxAutoRetries: 0
  MaxAutoRetriesNextServer: 0
  ReadTimeout: 10000

feign:
  hystrix:
    enabled: false

Generally configuration from Scenario 2 and 3 is right, you always get the full response. But in some cases you will wait more than 1 second (Scenario 2) or more than 5 seconds (Scenario 3) and delayed instance receives 50% requests from Ribbon client. But fortunately there is Hystrix – circuit breaker.

Scenario 4

Let’s enable Hystrix just by removing feign property. There is no auto retries for Ribbon client (1) and its read timeout (2) is bigger than Hystrix’s timeout (3). 1000ms is also default value for Hystrix timeoutInMilliseconds property. Hystrix circuit breaker and fallback will work for delayed instance of account service. For some first requests you receive fallback response from Hystrix. Then delayed instance will be cut off from requests, most of them will be directed to not delayed instance.

ribbon:
  eureka:
    enabled: true
  MaxAutoRetries: 0 #(1)
  MaxAutoRetriesNextServer: 0
  ReadTimeout: 2000 #(2)

hystrix:
  command:
    default:
      execution:
        isolation:
          thread:
            timeoutInMilliseconds: 1000 #(3)

Scenario 5

This scenario is a more advanced development of Scenario 4. Now Ribbon timeout (2) is lower than Hystrix timeout (3) and also auto retries mechanism is enabled (1) for local instance and for other instances (4). The result is same as for Scenario 2 and 3 – you receive full response, but Hystrix is enabled and it cuts off delayed instance from future requests.

ribbon:
  eureka:
    enabled: true
  MaxAutoRetries: 3 #(1)
  MaxAutoRetriesNextServer: 1 #(4)
  ReadTimeout: 1000 #(2)

hystrix:
  command:
    default:
      execution:
        isolation:
          thread:
            timeoutInMilliseconds: 10000 #(3)

I could imagine a few other scenarios. But the idea was just a show differences in circuit breaker and fallback when modifying configuration properties for Feign, Ribbon and Hystrix in application.yml.

Hystrix

Let’s take a closer look on standard Hystrix circuit breaker and  usage described in Scenario 4. To enable Hystrix in your Spring Boot application you have to following dependencies to pom.xml. Second step is to add annotation @EnableCircuitBreaker to main application class and also @EnableHystrixDashboard if you would like to have UI dashboard available.

<dependency>
	<groupId>org.springframework.cloud</groupId>
	<artifactId>spring-cloud-starter-hystrix</artifactId>
</dependency>
<dependency>
	<groupId>org.springframework.cloud</groupId>
	<artifactId>spring-cloud-starter-hystrix-dashboard</artifactId>
</dependency>

Hystrix fallback is set on Feign client inside customer service.

@FeignClient(value = "account-service", fallback = AccountFallback.class)
public interface AccountClient {

    @RequestMapping(method = RequestMethod.GET, value = "/accounts/customer/{customerId}")
    List<Account> getAccounts(@PathVariable("customerId") Integer customerId);

}

Fallback implementation is really simple. In this case I just return empty list instead of customer’s account list received from account service.

@Component
public class AccountFallback implements AccountClient {

	@Override
	public List<Account> getAccounts(Integer customerId) {
		List<Account> acc = new ArrayList<Account>();
		return acc;
	}

}

Now, we can perform some tests. Let’s start discovery service, two instances of account service on different ports (-DPORT VM argument during startup) and customer service. Endpoint for tests is /customers/{id}. There is also JUnit test class which sends multiple requests to this enpoint available in customer-service module pl.piomin.microservices.customer.ApiTest.

	@RequestMapping("/customers/{id}")
	public Customer findById(@PathVariable("id") Integer id) {
		logger.info(String.format("Customer.findById(%s)", id));
		Customer customer = customers.stream().filter(it -> it.getId().intValue()==id.intValue()).findFirst().get();
		List<Account> accounts =  accountClient.getAccounts(id);
		customer.setAccounts(accounts);
		return customer;
	}

I enabled Hystrix Dashboard on account-service main class. If you would like to access it call from your web browser http://localhost:2222/hystrix address and then type Hystrix’s stream address from customer-service http://localhost:3333/hystrix.stream. When I run test that sends 1000 requests to customer service about 20 (2%) of them were forwarder to delayed instance of account service, remaining to not delayed instance. Hystrix dashboard during that test is visible below. For more advanced Hystrix configuration refer to its documentation available here.

hystrix-1