Design and Implementation of a Custom RPC Framework

RPC (Remote Procedure Call) enables developers to invoke remote services as if they were local methods, abstracting network communication complexities.

The framework consists of three core components: client, server, and a registry (Zookeeper). During a call, the server registers its services, the client discovers them, encodes the request via a proxy, sends it over Netty, the server decodes, invokes the local service, encodes the response, and returns it to the client.

Key roles include service governance, load balancing, fault tolerance, serialization/deserialization, codec, thread pool, dynamic proxy, and optional features such as connection pool, logging, and security.

Load balancing strategies are implemented in rpc-core , supporting round‑robin (FullRoundBalance) and random (RandomBalance) with random as default, configurable via rpc.client.balance.

The custom message protocol follows a "length + content" format to solve TCP packet fragmentation and aggregation, containing fields such as magic number, version, serialization algorithm, message type, status, message ID, data length, and payload.

Serialization is handled in rpc-core using Hessian by default, with optional JSON support.

TCP packet framing issues are addressed by prefixing each message with its length, allowing the decoder to determine when a complete packet has been received.

Implementation uses Spring Boot auto‑configuration: the client starter registers

org.springframework.boot.autoconfigure.EnableAutoConfiguration=com.rrtv.rpc.client.config.RpcClientAutoConfiguration

, while the server starter registers analogous configuration. Services are annotated with @RpcService (provider) and injected with @RpcAutowired (consumer). Dynamic proxies and Netty handlers manage request/response processing.

Environment setup requires Windows, IntelliJ IDEA, Spring Boot 2.5.2, JDK 1.8, Netty 4.1.42.Final, Maven, and Zookeeper 3.7.0. After starting Zookeeper, run the provider and consumer applications and test via http://localhost:9090/hello/world?name=hello, which returns “您好：hello”.

The full source code is available at https://gitee.com/listen_w/rpc .