rpc_core

a tiny C++14 rpc library, supports all platforms (macOS, Linux, Windows, iOS, Android, etc.) and most microchips ( Arduino, STM32, ESP32/ESP8266, etc.)

Recommend TCP-based implementation: asio_net

Introduction

Full-feature rpc frameworks (e.g. gRPC and bRPC) are very complex and not suitable for embedded systems.

This project offers a lightweight and user-friend rpc library that is better suited for one-to-one rpc calls. It supports all platforms and a wide range of microchips, including Arduino, STM32, ESP32/ESP8266, and more.

Note: This project only offers the protocol layer and API, it does not include the implementation of the transport layer. For TCP-based implementation: asio_net

Features

• Header-Only
• No-Schema
• Support performance-limited platforms including microchips
• Support any connection type (tcp socket, serial port, etc.)
• High Performance Serialization, support most STL containers and user type
• Serialization plugins implementations for flatbuffers and nlohmann::json
• Support co_await, depend on C++20 and asio, or custom implementation
• Support subscribe async callback, async coroutine, and custom scheduler
• RAII-based dispose for automatic cancel request
• Support timeout, retry, cancel api

TCP-based implementations

• C++

  • asio_net: based on asio
    Support macOS, Linux, Windows, iOS, Android, etc. and can be used on MCUs that support asio, such as ESP32.

• Rust

  • ./rust: based on tokio
    Support callback and async/await, details: README.md

Requirements

• C++14
• Provide your connection implementation: connection
  NOTICE: complete data packets are required for data transmission, such as websocket.
  If using tcp socket, serial port, etc., message pack and unpack need to be implemented. Or you can use stream_connection.
• Optional: C++20 (for coroutine api, co_await async_call)

Usage

• async callback:

// The Receiver
rpc->subscribe("cmd", [](const std::string& msg) -> std::string {
    assert(msg == "hello");
    return "world";
});

// The Sender
rpc->cmd("cmd")
    ->msg(std::string("hello"))
    ->rsp([](const std::string& rsp) {
      assert(rsp == "world");
    })
    ->call();

• async coroutine:

// The Receiver
rpc->subscribe("cmd", [&](request_response<std::string, std::string> rr) -> asio::awaitable<void> {
    assert(rr->req == "hello");
    asio::steady_timer timer(context);
    timer.expires_after(std::chrono::seconds(1));
    co_await timer.async_wait();
    rr->rsp("world");
}, scheduler_asio_coroutine);

// The Sender
// use C++20 co_await with asio, or you can use custom async implementation, and co_await it!
auto rsp = co_await rpc->cmd("cmd")->msg(std::string("hello"))->async_call<std::string>();
assert(rsp.data == "world");

Inspect the code for more details: rpc_s_coroutine.cpp and rpc_c_coroutine.cpp

• Addition:

1. msg and rsp support any serializable type, refer to Serialization.
2. Detailed usages and unittests can be found here: rpc_test.cpp
3. There is an example shows custom async impl: rpc_c_coroutine.hpp

Serialization

High-performance and memory-saving binary serialization.

Why design a new serialization

Fist of all, I want to keep rpc_core library standalone, without any dependencies, except for STL.

Moreover, these serialization libraries do not align with my design goals:

• protobuf library is too large for some platforms, and it's Varint, Zigzag, and GZIP will use a lot of cpu.
• msgpack has similarity reason to protobuf.
• flatbuffers serialized data is too large.

Of course, when communicating across languages, it is recommended to use the above serialization libraries!

Finally, it also provides a way to use thirdparty serialization libraries directly, refer to Serialization Plugins.

Usage

For example, user data:

struct Type {
  uint8_t id = 1;
  uint8_t age = 18;
  std::string name = "test";
};

json: {"id":1,"age":18,"name":"test"}

• [x] std::string
• [x] std::wstring
• [x] std::array
• [x] std::vector
• [x] std::list
• [x] std::forward_list
• [x] std::deque
• [x] std::pair
• [x] std::tuple
• [x] std::map
• [x] std::unordered_map
• [x] std::multimap
• [x] std::unordered_multimap
• [x] std::set
• [x] std::unordered_set
• [x] std::multiset
• [x] std::unordered_multiset
• [x] std::stack
• [x] std::queue
• [x] std::priority_queue
• [x] std::bitset
• [x] std::complex
• [x] std::chrono::duration
• [x] std::chrono::time_point
• [x] std::unique_ptr
• [x] std::shared_ptr
• [x] rpc_core::binary_wrap
• [x] custom struct/class

  #include "rpc_core/serialize.hpp"
  struct TestStruct {
  uint8_t a;
  std::string b;
  OtherType c
  // RPC_CORE_DEFINE_TYPE_INNER(a, b, c);
  };
  RPC_CORE_DEFINE_TYPE(TestStruct, a, b, c);

  choose RPC_CORE_DEFINE_TYPE or RPC_CORE_DEFINE_TYPE_INNER for private member variable.

Serialization Plugins

• flatbuffers.hpp
  Supports using types generated by flatbuffers directly as message
  (add the option --gen-object-api when using flatc)

• json_msg.hpp
  Supports using types supported by nlohmann/json directly as message
  (the to_json/from_json rules in nlohmann/json need to be satisfied, and use DEFINE_JSON_CLASS).

• json.hpp
  A flexible way to use nlohmann/json

License

This project is licensed under the MIT license.

Links

• Implementation based on asio: asio_net

• Implementation suitable for ESP8266: esp_rpc