muse rpc
1.0.0
Introduction : A lightweight RPC framework based on Connect UDP, custom network protocol (simple request response protocol), and Reactor network model (one loop per thread + thread pool)!
C++ 17 standard memory pool千级访问concurrency in a single slave reactor and four worker threads!内网穿透in the case of non-public IP!留言message.Current usage :
# 需要提前安装zlib库
# 本人开发环境 GCC 11.3 CMake 3.25 clion ubuntu 22.04
git clone [email protected]:sorise/muse-rpc.git
cd muse-rpc
cmake -S . -B build
cmake --build
cd build
./muse #启动Architecture diagram :
The basic configuration content is as follows
int main () {
// 启动配置
// 4 设置 线程池最小线程数
// 4 设置 线程池最大线程数
// 4096 线程池任务缓存队列长度
// 3000ms; 动态线程空闲时间 3 秒
// 日志目录
// 是否将日志打印到控制台
muse::rpc::Disposition::Server_Configure ( 4 , 4 , 4096 , 3000ms, " /home/remix/log " , true );
//绑定方法的例子
Normal normal ( 10 , " remix " ); //用户自定义类
// 同步，意味着这个方法一次只能由一个线程执行，不能多个线程同时执行这个方法
muse_bind_sync ( " normal " , &Normal::addValue, & normal ); //绑定成员函数
muse_bind_async ( " test_fun1 " , test_fun1); // test_fun1、test_fun2 是函数指针
muse_bind_async ( " test_fun2 " , test_fun2);
// 开一个线程启动反应堆,等待请求
// 绑定端口 15000， 启动两个从反应堆，每个反应堆最多维持 1500虚链接
// ReactorRuntimeThread::Asynchronous 指定主反应堆新开一个线程运行，而不是阻塞当前线程
Reactor reactor ( 15000 , 2 , 1500 , ReactorRuntimeThread::Asynchronous);
try {
//开始运行
reactor. start ();
} catch ( const ReactorException &ex){
SPDLOG_ERROR ( " Main-Reactor start failed! " );
}
/*
* 当前线程的其他任务
* */
//程序结束
spdlog::default_logger ()-> flush (); //刷新日志
}After startup:
Use macros muse_bind_sync and muse_bind_async . The former calls SynchronousRegistry and the latter calls Registry. The prototype is as follows.
# include < iostream >
# include " rpc/rpc.hpp "
using namespace muse ::rpc ;
using namespace muse ::pool ;
using namespace std ::chrono_literals ;
//绑定方法的例子
Normal normal ( 10 , " remix " );
// 绑定类的成员函数、使用 只同步方法 绑定
muse_bind_sync ( " normal " , &Normal::addValue, & normal );
// 绑定函数指针
muse_bind_async ( " test_fun1 " , test_fun1);
// 绑定 lambda 表达式
muse_bind_async ( " lambda test " , []( int val)->int{
printf ( " why call me n " );
return 10 + val;
});Registration methods on the server side require the use of Registry and SynchronousRegistry objects.
SynchronousRegistry explanation: value is a field used to record how many clients have requested the addValue method. If 1000 clients have requested addValue and are registered using Registry, the value of value may not be 1000 because access to value is not thread-safe. If registered using SynchronousRegistry, it must be 1000.
class Counter {
public:
Counter ():value( 0 ){}
void addValue (){
this -> value ++;
}
private:
long value;
};Registration method : The definitions of the two macros are as follows
# define muse_bind_async (...)
Singleton<Registry>()-> Bind (__VA_ARGS__);
//同步方法，一次只能一个线程执行此方法
# define muse_bind_sync (...)
Singleton<SynchronousRegistry>()-> Bind (__VA_ARGS__);The client uses the Client object, which will return an Outcome<R> object, and the isOK method will indicate whether the return is successful! If false is returned, its member protocolReason will indicate whether there is a network abnormality, and the response member will indicate whether it is an rpc request and response error.
# include " rpc/rpc.hpp "
# include " rpc/client/client.hpp "
using namespace muse ::rpc ;
using namespace muse ::timer ;
using namespace std ::chrono_literals ;
int main{
// //启动客户端配置
muse::rpc::Disposition::Client_Configure ();
// MemoryPoolSingleton 返回一个 std::shared_ptr<std::pmr::synchronized_pool_resource>
//你可以自己定一个内存池
//传入 服务器地址和服务端端口号、一个C++ 17 标准内存池
Client remix ( " 127.0.0.1 " , 15000 , MemoryPoolSingleton ());
//调用远程方法
Outcome<std::vector< double >> result = remix. call <std::vector< double >>( " test_fun2 " ,scores);
std::cout << result. value . size () << std::endl;
//调用 无参无返回值方法
Outcome< void > result =remix. call < void >( " normal " );
if (result. isOK ()){
std::printf ( " success n " );
} else {
std::printf ( " failed n " );
}
//调用
auto ri = remix. call < int >( " test_fun1 " , 590 );
std::cout << ri. value << std::endl; // 600
};
Error handling: Under normal circumstances, you only need to pay attention to whether the isOk method is true. If you need to know the details of the error, you can use the following method. The two enumeration objects FailureReason and RpcFailureReason point out network errors and RPC request errors respectively.
auto resp = remix.call< int >( " test_fun1 " , 590 );
if (resp.isOK()){
//调用成功
std::cout << " request success n " << std::endl; // 600
std::cout << ri. value << std::endl; // 600
} else {
//调用失败
if (resp. protocolReason == FailureReason::OK){
//错误原因是RPC错误
std::printf ( " rpc error n " );
std::cout << resp. response . getReason () << std::endl;
//返回 int 值对应 枚举 RpcFailureReason
} else {
//错误原因是网络通信过程中的错误
std::printf ( " internet error n " );
std::cout << ( short )resp. protocolReason << std::endl; //错误原因
}
}
// resp.protocolReason() 返回 枚举FailureReason
enum class FailureReason : short {
OK, //没有失败
TheServerResourcesExhausted, //服务器资源耗尽，请勿链接
NetworkTimeout, //网络连接超时
TheRunningLogicOfTheServerIncorrect, //服务器运行逻辑错误，返回的报文并非所需
};
// resp.response.getReason() 返回值 是 int
enum class RpcFailureReason : int {
Success = 0 , // 成功
ParameterError = 1 , // 参数错误,
MethodNotExist = 2 , // 指定方法不存在
ClientInnerException = 3 , // 客户端内部异常，请求还没有到服务器
ServerInnerException = 4 , // 服务器内部异常，请求到服务器了，但是处理过程有异常
MethodExecutionError = 5 , // 方法执行错误
UnexpectedReturnValue = 6 , //返回值非预期
};Non-blocking requests mean that you only need to set up the request task and register a callback function to process the request results. The sending process is handled by the Transmitter object, so that the current thread will not be blocked due to network reasons and processing. Blocking will be handled based on callbacks and can be used to handle a large number of requests. Here we need a Transmitter object and the sending task is set through TransmitterEvent!
Note : The number of tasks sent by a single Transmitter at the same time should be less than 100. If the number of request tasks sent exceeds 100, the timeout needs to be increased. You can call the following interface to set the operation time.
void test_v (){
//启动客户端配置
muse::rpc::Disposition::Client_Configure ();
Transmitter transmitter ( 14500 );
// transmitter.set_request_timeout(1500); //设置请求阶段的等待超时时间
// transmitter.set_response_timeout(2000); //设置响应阶段的等待超时时间
//测试参数
std::vector< double > score = {
100.526 , 95.84 , 75.86 , 99.515 , 6315.484 , 944.5 , 98.2 , 99898.26 ,
9645.54 , 484.1456 , 8974.4654 , 4894.156 , 89 , 12 , 0.56 , 95.56 , 41
};
std::string name {
" asdasd54986198456h487s1as8d7as5d1w877y98j34512g98ad "
" sf3488as31c98aasdasd54986198sdasdasd456h487s1as8d7a "
" s5d1w877y98j34512g98ad "
};
for ( int i = 0 ; i < 1000 ; ++i) {
TransmitterEvent event ( " 127.0.0.1 " , 15000 ); //指定远程IP 、Port
event. call < int >( " read_str " , name,score); //指定方法
event. set_callBack ([](Outcome< int > t){ //设置回调
if (t. isOK ()){
printf ( " OK lambda %d n " , t. value );
} else {
printf ( " fail lambda n " );
}
});
transmitter. send ( std::move (event));
}
//异步启动发射器，将会新开一个线程持续发送
transmitter. start (TransmitterThreadType::Asynchronous);
//停止发射器，这是个阻塞方法，如果发送器还有任务没有处理完，将会等待
transmitter. stop ();
//如果想直接停止可以使用 transmitter.stop_immediately()方法
}Supports configuring the number of core threads, maximum number of threads, task cache queue length, and dynamic thread idle time!
ThreadPoolSetting::MinThreadCount = 4 ; //设置 核心线程数
ThreadPoolSetting::MaxThreadCount = 4 ; //设置 核心线程数
ThreadPoolSetting::TaskQueueLength = 4096 ; //设置 任务缓存队列长度
ThreadPoolSetting::DynamicThreadVacantMillisecond = 3000ms; //动态线程空闲时间Introduction : Simple Request Response Protocol (SR2P protocol) is a two-phase protocol specially customized for RPC. It is divided into two stages: request and response without establishing a link.
The protocol fields are as follows, the protocol header is 26 bytes, the field byte order is big endian, and the data part is little endian. Due to MTU limitations, the network standard MTU is 576, and the data part is up to 522 bytes. More Please see Protocol.pdf.
11110000 , one byte.The SR2P protocol will determine how many datagrams are generated each time based on the amount of data generated. The following takes 2 datagrams at a time as an example. The request flow chart under basically normal circumstances:
For details on the handling process of other situations, please see the Protocol.pdf document
