Exploring PhxRPC: A Lightweight High-Performance RPC Framework for Backend Development

Overview

Uses Protobuf as IDL to describe RPC interfaces and data structures.

Automatically generates client and server code from Protobuf files.

Hybrid sync‑async model with independent IO threads managed by Epoll and a fixed‑size worker thread pool; IO and workers communicate via memory queues.

Built‑in overload protection that adapts dynamically without manual thresholds.

Simple client/server configuration loading.

Supports Google’s Backup Requests pattern via lambda‑based concurrent server access.

Limitations

Does not support multi‑process mode.

Performance

Benchmark using the Search RPC C/S sample under idle worker conditions.

Test Environment

CPU: 24 x Intel(R) Xeon(R) CPU E5-2620 v3 @ 2.40GHz
Memory: 32 GB
Network: 1 Gbps
Ping between client/server: 0.05 ms
Concurrent request threads: 1000
Business data size (excluding HTTP header): 20 B
Worker threads: 20

Short Connection QPS

System (1 thread): 4.1w
System (3 threads): 8.5w
System (8 threads): 9.0w
System (20 threads): 9.2w
Boost (1 thread): 4.5w
Boost (3 threads): 9.5w
Boost (8 threads): 9.5w
Boost (20 threads): 9.5w

Long Connection QPS

System (1 thread): 5.5w
System (3 threads): 16w
System (8 threads): 36w
System (20 threads): 50w
Boost (1 thread): 6.2w
Boost (3 threads): 17.5w
Boost (8 threads): 41w
Boost (20 threads): 50w

How to Use

Write proto file

Example proto file is shown below.

Generate code

(PhxRPC root)/codegen/phxrpc_pb2server -I (PhxRPC root) -I (Protobuf include) -f (proto path) -d (output dir)
# sample
../codegen/phxrpc_pb2server -I ../ -I ../third_party/protobuf/include -f search.proto -d .

After running the tool, execute make in the output directory to generate all RPC code.

Enable Boost optimization

Edit the generated Makefile to uncomment the Boost LDFLAGS line, then run make clean && make. Ensure the Boost plugin for PhxRPC is compiled first.

Add custom code

Server implementation (xxx_service_impl.cpp)

Client implementation (xxx_client.cpp)

Client concurrent call example

The following macros are provided by PhxRPC for coroutine handling: uthread_begin, uthread_end, uthread_s, uthread_t. They enable implementation of Google’s Backup Requests pattern, where two servers are called concurrently and the first response ends the operation. The macros allow writing asynchronous I/O in a synchronous style.

Server configuration (xxx_server.conf)