This tutorial provides basic introductions on how a C++ programmer can develop a service with tRPC-Cpp from scratch.
In Quick Start, you have successfully run HelloWorld demo. Next, we will lead you to implement a proxy service that forwarding requests to HelloWorld server. Through this example, you will learn how to:
- Define a service in a .proto file.
- Use trpc-cmdline tool to generate a project skeleton which contains server and client code.
- Use synchronous and asynchronous APIs to access services.
Below is the calling graph of the proxy service(we call it Forward
for short) with others:
Client <---> Forward <---> HelloWorld
HelloWorld
: Service in Quick Start.Forward
: Service that will receive the client's requests, and then forward it by sending RPC requests to theHelloWorld
service.Client
: Client that will send RPC requests toForward
service.
Before continue, you must make sure below environments are already satisfied:
- Follow Quick Start to install the development toolkits and compile
HelloWorld
successfully. - Install trpc-cmdline tools.
- Create a empty directory using
mkdir trpc-cpp-tutorial
anywhere you want, all the opeartions after will ongoing here.
First, you should give service the definition about it's service name, methods, request/response message type of each method. Luckily, you can use protobuf IDL to easily do the work.
Create a subdirectory using mkdir deps
under trpc-cpp-tutorial
and then create a file named message.proto
. Fill it with below request/response message type defition:
syntax = "proto3";
package trpc.demo.forward;
message ForwardRequest {
string msg = 1;
}
message ForwardReply {
string msg = 1;
}
Create a file forward.proto
under trpc-cpp-tutorial
directory and fill it with a service
named Forward
:
service Forward {
...
}
Then, you can define rpc
methods in service
. tRPC-Cpp allows you to define unary and streaming rpc methods. But only unary rpc method is showed here, for stream method, you can refer to trpc stream for details. Add method named Route
with request message type as ForwardRequest
and response message type as ForwardReply
. Pay attention that, the message types are defined in different proto file(deps/message.proto
) and should be imported using import
keyword. You will see the complete forward.proto
file as below:
syntax = "proto3";
package trpc.demo.forward;
import "deps/message.proto";
service ForwardService {
rpc Route (ForwardRequest) returns (ForwardReply) {}
}
For what the package
defintion means in both proto files, we disscuss at terminology. In conclusion, we suggest the three-part formula trpc.{app}.{server}
. Here, {app}
is demo, and {server}
is forward
.
Under trpc-cpp-tutorial
and execute tree .
, you will see below file structure:
.
├── deps
│ └── message.proto
└── forward.proto
If you have proto files with complex dependency, they must be organized at a directory before use trpc-cmdline tool(shown as below). For examples, if proto files have below dependency:
test3/test.proto ---> test2/test.proto --->
mytest/helloworld.proto
test/test.proto --->
You should organize them as below:
.
├── mytest
│ └── helloworld2.proto
├── test
│ └── test.proto
├── test2
│ └── test.proto
└── test3
└── test.proto
Now, in the trpc-cpp-tutorial
directory, you can simply execute following command:
trpc create -l cpp --protofile=forward.proto
# The complex example methioned above should use command: trpc create -l cpp --protofile=mytest/helloworld2.proto
Among them, trpc
is the trpc-cmdline tool you've installed at Environment requirements
step, the two parameters of this tool are -l
specifys that the tool should generate a tRPC-Cpp project(change to go will generate tRPC-Go project, etc) and --protofile
specifys the tool should generate based on which proto file.
Once done, a tRPC-Cpp project based on forward.proto
will be output under trpc-cpp-tutorial
. It's directory name is the same as proto file. Execute command tree forward
and you will see the following file structure:
forward
├── build.sh
├── clean.sh
├── client
│ ├── BUILD
│ ├── conf
│ │ ├── trpc_cpp_fiber.yaml
│ │ └── trpc_cpp_future.yaml
│ ├── fiber_client.cc
│ └── future_client.cc
├── proto
│ ├── BUILD
│ ├── deps
│ │ └── message.proto
│ ├── forward.proto
│ └── WORKSPACE
├── README.md
├── run_client.sh
├── run_server.sh
├── server
│ ├── BUILD
│ ├── conf
│ │ ├── trpc_cpp_fiber.yaml
│ │ └── trpc_cpp.yaml
│ ├── server.cc
│ ├── server.h
│ ├── service.cc
│ └── service.h
└── WORKSPACE
PS: the genearted project will use bazel to build. Genearting a project using cmake to build is not supported yet.
Direcotry server contains service implementation code, client contains testing code that will send RPC requests to service, and proto contains code that manages proto file with it's dependency.
Direcotry server contains 1 configuration direcotry, 4 souce code files, and 1 bazel BUILD file.
- Configuration directory: There are two yaml files describing framework configuration here. One is configure using fiber (M:N coroutine), which means that the current service will use the fiber runtime, and developers should use synchronous api in their code. The other is configure using thread runtime, which means that the current service will use the merged or separated runtime, and should use future-promise based asynchronous api. You must first choose a runtime environment before developing tRPC-Cpp services. For how to select among different runtimes(fiber or separted/merge), please refer to runtime for details.
- BUILD file: Bazel will use it to Manage the building process.
- Source code files: There are two types of them. One is service process-level source code(server.h and server.cc). They define a class named ForwardServer, which inherits the
TrpcApp
class of tRPC-Cpp. Developers shoud override its'Initialize
andDestroy
methods. So during startup or terminate of process, user-defined initialization operations will be executed. The other type is the service-level source code(service .h and service.cc). They implement the user-defined logic of how service methods shoud work.
The implementation steps of the Forward
service are:
- Choose a runtime environment(we also call it the thread model), and fill it to global->threadmodel part of tRPC-Cpp configuration yaml file.
- Define a process-level class (here ForwardServer) which must inherit the
TrpcApp
class of the tRPC-Cpp. Then override itsRegisterPlugins
,Initialize
andDestroy
methods as you need. - Define a service-level class (here ForwardServiceServiceImpl) which must inherit the auto-genearted service class defined by the proto file. Then override its' RPC methods, where the first parameter
ServerContext
of each methods is the context of the current RPC execution and the second/third parameter represents request/response instance. - If needed, registry user-defined plugin you want to use at
RegisterPlugins
mentioned in 2(eg: register the custom protocolCodec
). - Complete the process-level business-related initialization at
Initialize
mentioned in 2.- First initialize the business-related logical operations that depend on the framework, such as: pull a remote configuration, create and start a thread pool.
- After the business-related operations are completed, registry service. Here, the ForwardServiceServiceImpl instance is created, and then registered into ForwardServer by invoke
RegisterService
function. Note that there will be only one ForwardServiceServiceImpl instance in the process.
- Complete the process-level business destruction operation in the
Destroy
method mentioned in 2.- Generally, you should stop and destroy dynamic resources created in
Initialize
, for example: stop running the thread pool.
- Generally, you should stop and destroy dynamic resources created in
- Use process-level implementation class at program entrypoint
main
. Invoking its'Main
andWait
functions to start the program.
Directory client contains 1 configuration direcotry, 2 souce code files, and 1 bazel BUILD file.
- Configuration directory: There are two yaml files describing framework configuration here. One is configure using fiber (M:N coroutine), which means that the current service will use the fiber runtime, and developers should use synchronous api in their code. The other is configure using thread runtime, which means that the current service will use the merged or separated runtime, and should use future-promise based asynchronous api.
- BUILD file: Bazel will use it to Manage the building process.
- Source code files: Two are act as RPC clients. One will use synchronous rpc call to access Forward service (fiber_client.cc), The other will use future-promise based asynchronous rpc call(future_client.cc). choose one of them in your test.
The implemention of these two clients is similar. We divide it into following steps:
- Parse configuration yaml file and use it to initialize tRPC-Cpp runtime.
- Put
Run
function into the runtime environment where the business logic will be executed. As the implementation ofRun
:ServiceProxy
will be created first using configuration at yaml(client->service). This proxy will in charge of handing all the RPC invoke between client and conresponding backend service.- Then, use
ServiceProxy
to send RPC request. The difference between fiber_client and future_client is that one uses a synchronous call and the other uses an asynchronous call. In addition, you should always createone ClientContext for each RPC call
.
Directory proto contains a bazel WORKSPACE file, a bazel BUILD file, and 2 proto files.
- WORKSPACE file: Indicate the proto directory should be viewed as an independent bazel project. If you check WORKSPACE file in project root directory, you will find the proto project is imported via
local_repository
. - BUILD file: Use bazel rules to manage proto files which may contain complex dependency,it can be imported by
@proto//:xxx_proto
at BUILD files of server/client in forward project. Relatvie stub code files will be generated after build. Service should inherit the generated class to implment bussiness logic per RPC methods(eg. ForwardServiceServiceImpl). Client can use the generated proxy to do RPC call. - proto files: Contain 2 proto files:
forward.proto
depends ondeps/message.proto
. Their file structure attrpc-cpp-tutorial
will be keeped at proto directory.
Enter the forward directory,compile and run the Forward
service:
./run_server.sh
PS: It will take a minutes to do the progress for the first time.
Then complie and run fiber_client and future_client respectively:
./run_client.sh
If success, you will see the following results:
fiber_client output:
FLAGS_service_name: trpc.demo.forward.ForwardService
FLAGS_client_config: client/conf/trpc_cpp_fiber.yaml
get rsp success
future_client output:
FLAGS_service_name: trpc.demo.forward.ForwardService
FLAGS_client_config: client/conf/trpc_cpp_future.yaml
get rsp success
Now that you have a successfull running Forward
service. But how to let it access the backend service HelloWorld
we got in Quick Start? The Forward
program serves as both server and client. As for the future_client
and future_client
programs, they are only clients.
Add fiber runtime configuration. Open file ./server/conf/trpc_cpp_fiber.yaml
, you can check global->threadmodel part of it as below:
global:
threadmodel:
fiber: # Use fiber threadmodel
- instance_name: fiber_instance # ThreadModel instance unique identifier, currently does not support configuring multiple Fiber instances
concurrency_hint: 8 # Number of physical threads running Fiber Workers, it is recommended to configure it as the same number as machine available cores.
For the full fiber runtime configuration, please refer to Framework Configuration.
Here our backend service is the HelloWorld
service, so the proto IDL file we want to obtain is helloworld.proto. Thanks to bazel, instead of copying, we can easily get it via bazel remote target.
Add remote bazel target in deps of file ./server/BUILD
as follows:
cc_library(
name = "service",
srcs = ["service.cc"],
hdrs = ["service.h"],
deps = [
"@proto//:forward_proto",
"@trpc_cpp//examples/helloworld:helloworld_proto", # add this line
"@trpc_cpp//trpc/client:make_client_context",
"@trpc_cpp//trpc/client:trpc_client",
"@trpc_cpp//trpc/log:trpc_log",
],
)
Before creating the ServiceProxy of the backend service, you need to define the it's options. Here we demonstrate setting them through configuration. Besides, tRPC-Cpp also supports through only code, code + configuration, for details, refer to Client Development Guide.
Add below proxy options in configuration file ./server/conf/trpc_cpp_fiber.yaml
:
client:
service:
- name: trpc.test.helloworld.Greeter
target: 0.0.0.0:12345
protocol: trpc
timeout: 1000
network: tcp
conn_type: long
is_conn_complex: true
selector_name: direct
By the configuration, Forward
service will use target(ip:port) to access the HelloWorld
service.
After setting the options of the ServiceProxy, then create and use the ServiceProxy to access the helloworld service.
First, add a constructor to the ForwardServiceServiceImpl class in file ./server/service.h
, and define a GreeterServiceProxy type smart pointer member variable.
...
#include "examples/helloworld/helloworld.trpc.pb.h"
...
using GreeterProxyPtr = std::shared_ptr<::trpc::test::helloworld::GreeterServiceProxy>;
class ForwardServiceServiceImpl : public ::trpc::demo::forward::ForwardService {
public:
ForwardServiceServiceImpl();
...
private:
GreeterProxyPtr greeter_proxy_;
};
...
Then create GreeterServiceProxy at the ForwardServiceServiceImpl constructor in file ./server/service.cc
.
...
#include "trpc/client/make_client_context.h"
#include "trpc/client/trpc_client.h"
#include "trpc/log/trpc_log.h"
...
ForwardServiceServiceImpl::ForwardServiceServiceImpl() {
greeter_proxy_ =
::trpc::GetTrpcClient()->GetProxy<::trpc::test::helloworld::GreeterServiceProxy>("trpc.test.helloworld.Greeter");
}
...
Create a GreeterServiceProxy smart pointer using ::trpc::GetTrpcClient()->GetProxy
.
After creating GreeterServiceProxy, use it in the Route method implemention of ForwardServiceServiceImpl to call the HelloWorld
service.
...
::trpc::Status ForwardServiceServiceImpl::Route(::trpc::ServerContextPtr context,
const ::trpc::demo::forward::ForwardRequest* request,
::trpc::demo::forward::ForwardReply* reply) {
TRPC_FMT_INFO("request msg: {}, client ip: {}", request->msg(), context->GetIp());
auto client_context = ::trpc::MakeClientContext(context, greeter_proxy_);
::trpc::test::helloworld::HelloRequest hello_request;
hello_request.set_msg(request->msg());
::trpc::test::helloworld::HelloReply hello_reply;
// block current fiber, not block current fiber worker thread
::trpc::Status status = greeter_proxy_->SayHello(client_context, hello_request, &hello_reply);
TRPC_FMT_INFO("status: {}, hello_reply: {}", status.ToString(), hello_reply.msg());
reply->set_msg(hello_reply.msg());
return status;
}
...
Before testing, to avoid failure due to service listen on the same port, you need to confirm whether the ip/port used by the Forward
service and admin(configure using admin_ip
/admin_port
in yaml file) is different from the HelloWorld
service.
Then, compile and test the program.
First, Run HelloWorld
and execute the following commands in the trpc-cpp directory. For details, please refer to Quick Start.
./bazel-bin/examples/helloworld/helloworld_svr --config=./examples/helloworld/conf/trpc_cpp_fiber.yaml
Then open another terminal and execute ./run_server.sh
in the forward directory. If the execution is successful, you can see the following information.
Server InitializeRuntime use time:29(ms)
Then open another terminal and execute ./run_client.sh
in the forward directory. If the execution is successful, you can see the following information.
FLAGS_service_name: trpc.demo.forward.ForwardService
FLAGS_client_config: client/conf/trpc_cpp_fiber.yaml
get rsp success
FLAGS_service_name: trpc.demo.forward.ForwardService
FLAGS_client_config: client/conf/trpc_cpp_future.yaml
get rsp success
If you only use fiber for programming, you can skip the following sections.
Add future runtime configuration. Open file ./server/conf/trpc_cpp.yaml
, you can check global->threadmodel part of it as below:
global:
threadmodel:
default:
- instance_name: default_instance
io_handle_type: separate
io_thread_num: 2
handle_thread_num: 6
You can choose to merge or separate threadmodel runtime. Here we take the separate threadmodel runtime as an example.
For detailed separate or merge runtime configuration, please refer to Framework Configuration.
You can refer to the corresponding section in fiber.
You can refer to the corresponding section in fiber.
The future asynchronous calling code is as follows:
::trpc::Status ForwardServiceServiceImpl::Route(::trpc::ServerContextPtr context,
const ::trpc::demo::forward::ForwardRequest* request,
::trpc::demo::forward::ForwardReply* reply) {
TRPC_FMT_INFO("request msg: {}, client ip: {}", request->msg(), context->GetIp());
// use asynchronous response mode
context->SetResponse(false);
auto client_context = ::trpc::MakeClientContext(context, greeter_proxy_);
::trpc::test::helloworld::HelloRequest hello_request;
hello_request.set_msg(request->msg());
::trpc::test::helloworld::HelloReply route_reply;
greeter_proxy_->AsyncSayHello(client_context, hello_request)
.Then([context](::trpc::Future<::trpc::test::helloworld::HelloReply>&& fut) {
::trpc::Status status;
::trpc::demo::forward::ForwardReply forward_reply;
if (fut.IsReady()) {
std::string msg = fut.GetValue0().msg();
forward_reply.set_msg(msg);
TRPC_FMT_INFO("Invoke success, hello_reply:{}", msg);
} else {
auto exception = fut.GetException();
status.SetErrorMessage(exception.what());
status.SetFrameworkRetCode(exception.GetExceptionCode());
TRPC_FMT_ERROR("Invoke failed, reason:{}", exception.what());
forward_reply.set_msg(exception.what());
}
context->SendUnaryResponse(status, forward_reply);
return ::trpc::MakeReadyFuture<>();
});
return ::trpc::kSuccStatus;
}
Just follow the operations in the fiber-related sections above. The only difference is that you need to modify the run_server.sh
content in the forward directory.
Before modify:
...
bazel-bin/demo/forward/forward --config=demo/forward/conf/trpc_cpp_fiber.yaml
...
After modify:
bazel-bin/demo/forward/forward --config=demo/forward/conf/trpc_cpp.yaml
At this point, you have successfully used fiber and future in tRPC-Cpp to access backend service.
- Learn how tRPC-Cpp works through tRPC-Cpp Architecture Design.
- Read User Guide to use tRPC-Cpp more comprehensively.