salvo-concurrency-limiter（齐射并发限制器）

Salvo Concurrency Limiter

max_concurrency(n) from salvo::concurrency_limiter is a hoop that wraps a tokio::sync::Semaphore. It try_acquires on each request — on failure it short-circuits with 429 Too Many Requests. Unlike rate limiters it caps parallel requests, not request frequency.

Setup

Lives in salvo-extra; enable the feature:

[dependencies]
salvo = { version = "0.89.3", features = ["concurrency-limiter"] }

Import path: salvo::concurrency_limiter::{max_concurrency, MaxConcurrency}.

Basic usage

use salvo::prelude::*;
use salvo::concurrency_limiter::max_concurrency;

#[handler]
async fn upload(res: &mut Response) {
    tokio::time::sleep(std::time::Duration::from_secs(10)).await;
    res.render("Upload complete");
}

#[tokio::main]
async fn main() {
    let router = Router::new().push(
        Router::with_path("upload")
            .hoop(max_concurrency(1))   // at most 1 in flight
            .post(upload),
    );
    let acceptor = TcpListener::new("0.0.0.0:8080").bind().await;
    Server::new(acceptor).serve(router).await;
}

Each hoop(max_concurrency(n)) call creates its own semaphore scoped to that route subtree. Reuse across subtrees by hoisting the hoop on a shared parent.

Per-route limits

let router = Router::new()
    .push(Router::with_path("upload")
        .hoop(max_concurrency(2))
        .post(upload_handler))
    .push(Router::with_path("reports/generate")
        .hoop(max_concurrency(1))
        .post(generate_report))
    .push(Router::with_path("api/{**rest}")
        .hoop(max_concurrency(100))
        .goal(api_handler))
    .push(Router::with_path("health").get(health_check));   // no limit

Combining middleware

hoop order matters — outer hoops run first. Put rate-limit/timeout outside the concurrency cap so rejected requests don't hold permits:

Router::with_path("api/{**rest}")
    .hoop(rate_limiter)             // runs first
    .hoop(Timeout::new(Duration::from_secs(30)))
    .hoop(max_concurrency(50))      // permit acquired last, released on handler exit
    .goal(api_handler)

Customizing the 429 response

The middleware renders StatusError::too_many_requests().brief("Max concurrency reached."). Override in a catcher:

use salvo::catcher::Catcher;

#[handler]
async fn on_overload(res: &mut Response, ctrl: &mut FlowCtrl) {
    if res.status_code() == Some(StatusCode::TOO_MANY_REQUESTS) {
        res.render(Json(serde_json::json!({
            "error": "busy",
            "retry_after": 5,
        })));
        ctrl.skip_rest();
    }
}

let service = Service::new(router).catcher(Catcher::default().hoop(on_overload));

Sizing guidance

Gotchas

• Shares no state across processes — for multi-replica deployments, use a distributed limiter upstream.
• Each hoop(max_concurrency(..)) is a separate semaphore; instantiating it inside a loop or per-request builder creates independent limits and defeats the cap.
• Permits are released when the inner handler returns. Pair with Timeout so stuck handlers eventually free the slot.
• The middleware uses try_acquire: requests are rejected, not queued. If you want queueing, put a tower::limit layer in via tower-compat instead.

Related Skills

• salvo-rate-limiter: Request frequency (per-second/minute) caps
• salvo-timeout: Combine with request deadlines so stuck handlers don't block slots
• salvo-realtime: Cap simultaneous SSE/WebSocket connections