Design and Implementation of a Lightweight Distributed RPC Framework (buddha)

Introduction

Remote Procedure Call (RPC) is a communication protocol that allows a program on one computer to invoke a sub‑program on another computer without the programmer having to write explicit networking code. It aims to simplify building distributed applications while preserving the simplicity of local call semantics.

During a spare period before an internship, the author implemented a lightweight distributed RPC framework named buddha (https://github.com/tinylcy/buddha). Although the codebase is small, it includes all essential components, and this article explains its design, component breakdown, and considerations.

Serialization and Deserialization

Since data transmitted over the network must be converted to bytes, an RPC framework needs to serialize objects to byte arrays and deserialize them back. Java’s default serialization can become a performance bottleneck under high concurrency, so many efficient serialization libraries such as Kryo, fastjson, and Protobuf are used. Buddha currently supports Kryo and fastjson.

TCP Packet Framing (Sticky/Partial Packets)

TCP is a byte‑stream protocol and does not know the boundaries of higher‑level messages. Large messages are split (requiring the server to reassemble them), while small messages may be buffered and sent together (requiring the server to separate them). The common solution is to add length information or delimiters to each packet. Buddha adopts the length‑header approach.

BIO vs. NIO

Blocking I/O (BIO) follows a one‑connection‑one‑thread model, which leads to high thread‑creation cost, large memory consumption, and expensive context switches under high concurrency. Non‑blocking I/O (NIO) allows a single thread to manage many connections by registering interest events. Buddha uses Netty to build a clear NIO‑based networking layer.

Service Registration and Discovery

To ensure stability and reliability, RPC services are often deployed in clusters. A service registry stores the addresses of available providers; clients query the registry to obtain service lists and perform load balancing. Buddha implements this functionality with ZooKeeper, supporting dynamic updates and failure notifications.

Implementation

The complete source code of buddha is available on GitHub (https://github.com/tinylcy/buddha).

References

RPC concept model and implementation analysis (http://mindwind.me/blog/2016/05/22/RPC)

NettyRpc (https://github.com/luxiaoxun/NettyRpc)