[TOC]

hrpc(http-rpc),基于http的java rpc框架。

Usecase

• gracefully export servive to others
• eliminate template code for built rpc service

1.Guideline

1.1 Prerequisite

Maven dependency

        <dependency>
            <groupId>com.shareit.live</groupId>
            <artifactId>spring-boot-starter-hrpc</artifactId>
            <version>0.0.1-SNAPSHOT</version>
        </dependency>

1.2 Creating rpc service interface

@HrpcService(name = "example")
public interface ExampleService {
    String sayHi(String userName);
}

1.3 Implements hrpc service interface at server side

@Service
public class ExampleServiceImpl implements ExampleService {
    @Override
    public String sayHi(String userName) {
        return "Hi " + userName;
    }
}

enable hrpc server, by config application.yml

hrpc:
  enable: true

1.4 Config client yml & scan hrpc service at client side

application.yml

hrpc:
  remote-servers:
    - name: example
      servers: http://127.0.0.1:8080

scan hrpc service

@HrpcServiceScan(basePackages = {"com.shareit.live.example"})
@SpringBootApplication
public class WsClientDemoApplication {

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

}

1.5 Client invoke service

@Service
public class TestService {
    @Autowired
    private ExampleService exampleService;

    public void test() {
        String userName = "test_user";
        String greeting = exampleService.sayHi(userName);
        System.out.println(greeting);
    }
}

2.序列化框架对比

不同类型的请求参数，从client端经框架序列化后传输到server再由框架反序列；不同框架的结果对比

2.1 对比fastjson和protostuff

请求参数类型	fastjson	protostuff
java.lang.String	java.lang.String	java.lang.String
int	java.lang.Integer	java.lang.Integer
byte	java.lang.Byte	java.lang.Byte
byte[]	byte[]	byte[]
int[][]	com.alibaba.fastjson.JSONArray(error)	int[][]
com.shareit.live.proto.ReqMsg	com.alibaba.fastjson.JSONArray(error)	com.shareit.live.proto.ReqMsg
com.shareit.live.fecade.domain.UserModel[]	com.alibaba.fastjson.JSONArray(error)	com.shareit.live.fecade.domain.UserModel[]
int[]	com.alibaba.fastjson.JSONArray(error)	int[]
java.lang.Integer[]	com.alibaba.fastjson.JSONArray(error)	java.lang.Integer[]
java.util.List	com.alibaba.fastjson.JSONArray	java.util.ArrayList
java.util.Set	java.util.HashSet	java.util.HashSet

public class HrpcRequest {

    /**
     * implemented interface
     */
    private String clazz;

    /**
     * method name
     */
    private String method;

    /**
     * args of method
     */
    private Object[] args;

    /**
     * arg types
     */
    private Class<?>[] argTypes;

}
上面所有的请求参数都是赋值给Object[]后，再交给序列化框架处理；
fastjson在处理Object是需要使用autoType类型；
上面的测试都是开启了autoType的，所以fastjson能够处理java.util.Set；
不过fastjson将除byte[]以外的所有数组类型处理为jsonArray;
导致反射调用时抛出异常java.lang.IllegalArgumentException:
argument type mismatch] with root cause ；
java.util.List没有抛异常，完全是因为jsonArray implemts List<Object>的原因，
其他数组全都参数不匹配；
很明显,fastjson将所有数组都用jsonArray来处理，然后硬编码了byte[];


另外fastjson不能处理protobuf生成的java类: com.alibaba.fastjson.JSONException:
default constructor not found.
class com.google.protobuf.Descriptors$FieldDescriptor] with root cause;
fastjson反序列化依赖无参构造函数进行类的初始化，protobuf不走这一套；

2.2 排除Hessian的原因

Hessian 要求序列化的类必须实现 java.io.Serializable接口，这个要求对入参，返回值，嵌套类型的限制太大；

2.3 why protostuff

• 通过了所有的参数序列化反序列测试；
• 支持protobuf;

2.4 protostuff对model不一致的处理情况

• protostuff 当前版本支持forward-compatible,不支持backward-compatible ;
• pojo 的field数量可以不一致，多的fileds不处理，缺少的fields使用缺省值；
• pojo 的同名field的类型必须一致，不然无法进行反序列化
• pojo fields的顺序不一致会导致seq为负数，无法解析
• 不能处理repeated protobuf: 如果设置 enable morph capability globally,会导致和protobuf不兼容；protostuff/protostuff#240

3.Performance test

3.1 Test execution environment:

CPU: Intel(R) Core(TM) i7-8565U CPU @ 1.80GHz 1.99 GHz
MEM: 8.00GB
JDK Version: Java HotSpot(TM) 64-Bit Server VM "1.8.0_211"

JVM Args:
-Xms128m
-Xmx750m
-XX:ReservedCodeCacheSize=240m
-XX:+UseConcMarkSweepGC
-XX:SoftRefLRUPolicyMSPerMB=50
-ea
-XX:CICompilerCount=2
-Dsun.io.useCanonPrefixCache=false
-Djdk.http.auth.tunneling.disabledSchemes=""
-XX:+HeapDumpOnOutOfMemoryError
-XX:-OmitStackTraceInFastThrow
-Djdk.attach.allowAttachSelf=true
-Dkotlinx.coroutines.debug=off
-Djdk.module.illegalAccess.silent=true

3.2 Invocation performance test reuslt ####Java 8

Spring AOP uses either JDK dynamic proxies or CGLIB to create the proxy for a given target object. Spring proxy doc

-------------------------
| Invoke number: 100000  |
-------------------------

args	framework	time cost
void	hrpc-jdk-dynamic-proxy	1.79548 ms
void	hrpc-cglib-proxy	1.88901 ms
void	plain-http	0.21996 ms
pojo	hrpc-jdk-dynamic-proxy	1.84397 ms
pojo	hrpc-cglib-proxy	1.84491 ms
pojo	plain-http	1.68574 ms

4.Load Balance

hrpc support a default simple random load balance strategy; At the same time, you can implement custom load balance strategy follow steps below to implement custom load balance strategy:

1. config serverFetchBeanName with @HrpcServece, e.g. @HrpcService(name = "fecade", serverFetchBeanName = "serverFetch")
2. support a spring bean which implements interface ServerFetch and with beanName configed in step 1, e.g. 
@Service(value = "serverFetch")
public class HrpcServerFetch implements ServerFetch {
    @Override
    public String fetch(String name) {
        int random = RandomUtils.nextInt(0, 10);
        System.out.println("random=" + random);
        if (random > 5) {
            return "http://127.0.0.1:2048";
        } else {
            return "http://localhost:45677";
        }
    }
}

5.TODO LIST

• 当前hrpc的客户端默认重试配置为 maxAttempts=2, backoff=200ms，后续支持可配置重试参数