Mir Backend Rust Actix

/mir-backend-rust-actix · Make It Right (Actix-web)

Bottom tier of the chain: mir-backend (generic gates) → mir-backend-rust (Tokio runtime model) → this (Actix-web library mechanics). Run the gates first; load the Rust runtime tier for blocking/cancellation/Arc concerns; reach for this at Gate 5 (design mechanics), Gate 6 (implementation), and Gate 7 review. Runtime-level concerns (blocking the async runtime, cancellation safety, bounded channels, timeouts) live in mir-backend-rust — not here.

Stack assumed: Actix-web (4.x) · actix-rt (Tokio-backed single-threaded workers) · async Rust.

The Actix-web footguns AI walks into most

1. The multi-worker App Data trap — the defining Actix footgun

HttpServer::new accepts a closure (the App factory) and calls it once per worker thread (default: one worker per CPU core). Any state you construct inside the closure is created N times — yielding N completely independent copies. This is silent and compiles perfectly: there's no error, just N separate caches, N separate in-memory counters, or N separate connection pools that never coordinate.

// WRONG — a fresh Vec is created for each of the N workers
// Every worker has its own isolated list; requests are distributed
// across workers, so the list appears empty or inconsistent
HttpServer::new(|| {
    let state = web::Data::new(Mutex::new(Vec::::new())); // N copies!
    App::new()
        .app_data(state)
        .route("/items", web::get().to(list_items))
})

// RIGHT — construct state OUTSIDE the closure, clone the Arc in
let state = web::Data::new(Mutex::new(Vec::::new())); // one Arc

HttpServer::new(move || {
    App::new()
        .app_data(state.clone()) // clone the Arc, not the data
        .route("/items", web::get().to(list_items))
})
.workers(4)
.bind("0.0.0.0:8080")?
.run()
.await

The same rule applies to DB pools, Redis connections, caches, and any other shared resource. Build them once, outside HttpServer::new, wrap in web::Data::new (which is Arc), and .clone() the Data handle into each App factory invocation.

AI almost always builds state inside the closure because that's where all the App configuration lives — it looks natural. The bug only surfaces at runtime when you notice your in-memory state is silently partitioned.

2. web::Data is an Arc — clone the handle, not the data

web::Data is a newtype wrapper around Arc. Cloning a web::Data clones the Arc (cheap reference count bump) — the underlying T is shared. This is correct and expected:

// In the factory:
.app_data(pool.clone()) // clones the Arc wrapper, not the pool

// In a handler:
async fn get_user(pool: web::Data, path: web::Path) -> impl Responder {
    // pool.get_ref() gives &PgPool, or use Deref coercion
    let row = sqlx::query!("SELECT * FROM users WHERE id = $1", *path)
        .fetch_one(pool.get_ref())
        .await;
    ...
}

.data() is deprecated (removed in Actix-web 4). Use .app_data(web::Data::new(...)). The old .data() method exists in 3.x codebases — if you see it, update to .app_data.

3. Workers are single-threaded actix-rt — !Send types allowed, but blocking still starves the worker

Each Actix-web worker runs its own single-threaded actix-rt executor (a single-threaded Tokio runtime). Because the executor is single-threaded, handler futures do not need to be Send — you can use Rc, Cell, and other !Send types freely inside handlers. This is a deliberate design difference from Axum (which requires Send futures for multi-thread Tokio).

However: a single-threaded executor means blocking work blocks the entire worker thread, stalling all requests routed to that worker. The mir-backend-rust rule (don't block async tasks) applies here too, with one Actix-specific escape hatch:

• For blocking/CPU-heavy work inside an Actix handler, use web::block — it runs the closure on a separate blocking threadpool and returns a Future:

async fn compress_data(body: web::Bytes) -> actix_web::Result {
    let compressed = web::block(move || {
        // runs on blocking pool — won't stall the actix worker
        compress_sync(&body)
    })
    .await
    .map_err(|e| actix_web::error::ErrorInternalServerError(e))??;
    Ok(compressed)
}

Do not use tokio::task::spawn_blocking in Actix handlers — it works, but web::block is the idiomatic Actix-web API and integrates with the error model. std::thread::sleep and sync I/O inside handlers are still blocked.

4. Extractors and their configuration — JsonConfig limits and error shaping

Actix-web extractors (Json, Form, Query, Path) are configured via typed config objects registered with .app_data(). AI frequently forgets to set JsonConfig limits, accepting arbitrarily large bodies:

// Set body size limit and custom error handler for JSON extractor
let json_cfg = web::JsonConfig::default()
    .limit(1_048_576) // 1 MB max body
    .error_handler(|err, req| {
        let response = HttpResponse::BadRequest().json(json!({ "error": err.to_string() }));
        actix_web::error::InternalError::from_response(err, response).into()
    });

HttpServer::new(move || {
    App::new()
        .app_data(json_cfg.clone())
        .app_data(pool.clone())
        // ...
})

Similarly, QueryConfig and PathConfig accept .error_handler to return clean JSON errors instead of the default plain-text Actix error responses.

5. Error handling — implement ResponseError, not unwrap

Actix-web's error trait is ResponseError (not IntoResponse as in Axum). Implement it on your error type to produce consistent HTTP responses:

#[derive(Debug, thiserror::Error)]
enum AppError {
    #[error("not found: {0}")]
    NotFound(String),
    #[error("database error")]
    DbError(#[from] sqlx::Error),
    #[error("unauthorized")]
    Unauthorized,
}

impl ResponseError for AppError {
    fn status_code(&self) -> StatusCode {
        match self {
            AppError::NotFound(_)  => StatusCode::NOT_FOUND,
            AppError::DbError(_)   => StatusCode::INTERNAL_SERVER_ERROR,
            AppError::Unauthorized => StatusCode::UNAUTHORIZED,
        }
    }

    fn error_response(&self) -> HttpResponse {
        HttpResponse::build(self.status_code())
            .json(json!({ "error": self.to_string() }))
    }
}

async fn get_user(
    pool: web::Data,
    path: web::Path,
) -> Result, AppError> {
    let user = sqlx::query_as!(User, "SELECT * FROM users WHERE id = $1", *path as i64)
        .fetch_optional(pool.get_ref())
        .await
        .map_err(AppError::DbError)?
        .ok_or_else(|| AppError::NotFound(format!("user {} not found", *path)))?;
    Ok(web::Json(user))
}

Never .unwrap() inside a handler — panics in Actix-web handlers propagate as 500 responses (Actix catches panics per request by default) but suppress the actual error from structured logging and clients.

How this slots into the core pipeline

• Gate 5 (Design): state construction outside HttpServer::new as the mandatory pattern; define AppState-equivalent as web::Data-wrapped values built before the factory; declare the error type implementing ResponseError; note JsonConfig limits.
• Gate 6 (Implementation): shared state built outside closure; .app_data(data.clone()) inside factory; web::block for blocking/CPU work; ? propagation through AppError: ResponseError; JsonConfig with explicit size limit and error handler.
• Gate 7 (Review): verify no state construction inside HttpServer::new; no unwrap in handlers; JsonConfig limits present; no .data() (deprecated) usage; web::block used for any blocking work.

Edit boundary (what belongs here vs. above/below)

This module holds ONLY Actix-web library mechanics. Apply the 3-tier placement test before adding anything:

• True for Go/Python/Node too (idempotency, invariants, gates, observability)? → generic core (mir-backend).
• True for every async Rust backend on Tokio (blocking, guard-across-await, cancellation, Arc/'static, channels, timeouts)? → runtime tier (mir-backend-rust).
• A mechanical footgun of this library (multi-worker App Data trap, web::Data Arc semantics, web::block, JsonConfig limits, ResponseError)? → here.
• A different Rust framework (Axum, Warp) → its own mir-backend-rust- module. Never widen this one.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: anantbhandarkar
• Source: anantbhandarkar/make-it-right
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.