# Volga
Fast, simple, and high-performance web framework for Rust, built on top of
[Tokio](https://tokio.rs/) and [hyper](https://hyper.rs/).
Volga is designed to make building HTTP services straightforward and explicit,
while keeping performance predictable and overhead minimal.
[](https://crates.io/crates/volga)
[](https://crates.io/crates/volga)
[](https://github.com/RomanEmreis/volga/blob/main/LICENSE)
[](https://github.com/RomanEmreis/volga/actions/workflows/rust.yml)
[](https://github.com/RomanEmreis/volga/actions/workflows/release.yml)
> 💡 **Status**: Volga is currently in preview.
> The public API may change while core abstractions are being finalized.
## Why Volga?
Volga focuses on clarity and control without sacrificing performance.
It avoids hidden behavior and framework-driven magic.
Macros are used sparingly and primarily to reduce boilerplate. Handlers, middleware, and routing behave exactly as they look in code.
Volga is a good fit if you:
- Want simple and readable handler signatures
- Care about predictable performance and low overhead
- Need fine-grained control over the HTTP request/response lifecycle
- Work with streaming, WebSockets, or long-lived connections
- Prefer explicit APIs over code generation
## Features
- HTTP/1 and HTTP/2 support
- Explicit and robust routing
- Composable async middlewares
- Dependency Injection without derive macros
- Typed request extraction
- WebSockets and WebSocket-over-HTTP/2
- Streaming-friendly HTTP
- Full **Tokio** compatibility
- Runs on stable Rust **1.90+**
## Getting Started
### Dependencies
```toml
[dependencies]
volga = "0.8.5"
tokio = { version = "1", features = ["full"] }
```
### Simple request handler
```rust
use volga::{App, ok};
#[tokio::main]
async fn main() -> std::io::Result<()> {
let mut app = App::new();
app.map_get("/hello/{name}", async |name: String| {
ok!("Hello {name}!")
});
app.run().await
}
```
This example demonstrates:
* typed path parameter extraction
* async request handlers
* minimal setup with zero boilerplate
More advanced examples (middleware, DI, auth, rate limiting) can be found in the
[documentation](https://romanemreis.github.io/volga-docs/) and [here](https://github.com/RomanEmreis/volga/tree/main/examples).
## Performance
Tested on a single instance with 4 threads and 500 concurrent connections:
```
OS: Arch Linux
CPU: Intel i7-8665U (8) @ 4.800GHz
RAM: 31686MiB
```
### Results
```
Running 10s test @ http://127.0.0.1:7878/hello
4 threads and 500 connections
Thread Stats Avg Stdev Max +/- Stdev
Latency 1.39ms 1.05ms 18.45ms 81.47%
Req/Sec 89.69k 18.07k 126.91k 57.50%
3575551 requests in 10.07s, 395.55MB read
Requests/sec: 355053.82
Transfer/sec: 39.28MB
```
> ⚠️ Benchmark results are provided for reference only.
> Actual performance depends on workload, middleware, and handler logic.
## License
Volga is licensed under the MIT License. Contributions welcome!