workers-rs

Ergonomic Rust bindings to Cloudflare Workers environment. Write your entire worker in Rust!

Read the Notes and FAQ

Example Usage

use worker::*;

#[event(fetch)]
pub async fn main(mut req: Request, env: Env, _ctx: worker::Context) -> Result<Response> {
    console_log!(
        "{} {}, located at: {:?}, within: {}",
        req.method().to_string(),
        req.path(),
        req.cf().unwrap().coordinates().unwrap_or_default(),
        req.cf().unwrap().region().unwrap_or("unknown region".into())
    );

    if !matches!(req.method(), Method::Post) {
        return Response::error("Method Not Allowed", 405);
    }

    if let Some(file) = req.form_data().await?.get("file") {
        return match file {
            FormEntry::File(buf) => {
                Response::ok(&format!("size = {}", buf.bytes().await?.len()))
            }
            _ => Response::error("`file` part of POST form must be a file", 400),
        };
    }

    Response::error("Bad Request", 400)
}

Getting Started

The project uses wrangler for running and publishing your Worker.

Use cargo generate to start from a template:

cargo generate cloudflare/workers-rs

There are several templates to choose from. During generation you will be prompted to enable panic=unwind and abort recovery (see Panic Recovery below). You should see a new project layout with a src/lib.rs. Start there! Use any local or remote crates and modules (as long as they compile to the wasm32-unknown-unknown target).

Once you’re ready to run your project, run your worker locally:

npx wrangler dev

Finally, go live:

# configure your routes, zones & more in your worker's `wrangler.toml` file
npx wrangler deploy

If you would like to have wrangler installed on your machine, see instructions in wrangler repository.

http Feature

worker 0.0.21 introduced an http feature flag which starts to replace custom types with widely used types from the http crate.

This makes it much easier to use crates which use these standard types such as axum and hyper.

This currently does a few things:

1. Introduce Body, which implements http_body::Body and is a simple wrapper around web_sys::ReadableStream.
2. The req argument when using the [event(fetch)] macro becomes http::Request<worker::Body>.
3. The expected return type for the fetch handler is http::Response<B> where B can be any http_body::Body<Data=Bytes>.
4. The argument for Fetcher::fetch_request is http::Request<worker::Body>.
5. The return type of Fetcher::fetch_request is Result<http::Response<worker::Body>>.

The end result is being able to use frameworks like axum directly (see example):

pub async fn root() -> &'static str {
    "Hello Axum!"
}

fn router() -> Router {
    Router::new().route("/", get(root))
}

#[event(fetch)]
async fn fetch(
    req: HttpRequest,
    _env: Env,
    _ctx: Context,
) -> Result<http::Response<axum::body::Body>> {
    Ok(router().call(req).await?)
}

We also implement try_from between worker::Request and http::Request<worker::Body>, and between worker::Response and http::Response<worker::Body>. This allows you to convert your code incrementally if it is tightly coupled to the original types.

Or use the Router:

Parameterize routes and access the parameter values from within a handler. Each handler function takes a Request, and a RouteContext. The RouteContext has shared data, route params, Env bindings, and more.

use serde::{Deserialize, Serialize};
use worker::*;

#[event(fetch)]
pub async fn main(req: Request, env: Env, _ctx: worker::Context) -> Result<Response> {

    // Create an instance of the Router, which can use parameters (/user/:name) or wildcard values
    // (/file/*pathname). Alternatively, use `Router::with_data(D)` and pass in arbitrary data for
    // routes to access and share using the `ctx.data()` method.
    let router = Router::new();

    // useful for JSON APIs
    #[derive(Deserialize, Serialize)]
    struct Account {
        id: u64,
        // ...
    }
    router
        .get_async("/account/:id", |_req, ctx| async move {
            if let Some(id) = ctx.param("id") {
                let accounts = ctx.kv("ACCOUNTS")?;
                return match accounts.get(id).json::<Account>().await? {
                    Some(account) => Response::from_json(&account),
                    None => Response::error("Not found", 404),
                };
            }

            Response::error("Bad Request", 400)
        })
        // handle files and fields from multipart/form-data requests
        .post_async("/upload", |mut req, _ctx| async move {
            let form = req.form_data().await?;
            if let Some(entry) = form.get("file") {
                match entry {
                    FormEntry::File(file) => {
                        let bytes = file.bytes().await?;
                    }
                    FormEntry::Field(_) => return Response::error("Bad Request", 400),
                }
                // ...

                if let Some(permissions) = form.get("permissions") {
                    // permissions == "a,b,c,d"
                }
                // or call `form.get_all("permissions")` if using multiple entries per field
            }

            Response::error("Bad Request", 400)
        })
        // read/write binary data
        .post_async("/echo-bytes", |mut req, _ctx| async move {
            let data = req.bytes().await?;
            if data.len() < 1024 {
                return Response::error("Bad Request", 400);
            }

            Response::from_bytes(data)
        })
        .run(req, env).await
}

Durable Object, KV, Secret, & Variable Bindings

All “bindings” to your script (Durable Object & KV Namespaces, Secrets, Variables and Version) are accessible from the env parameter provided to both the entrypoint (main in this example), and to the route handler callback (in the ctx argument), if you use the Router from the worker crate.

use worker::*;

#[event(fetch, respond_with_errors)]
pub async fn main(req: Request, env: Env, _ctx: worker::Context) -> Result<Response> {
    let router = Router::new();

    router
        .on_async("/durable", |_req, ctx| async move {
            let namespace = ctx.durable_object("CHATROOM")?;
            let stub = namespace.id_from_name("A")?.get_stub()?;
            // `fetch_with_str` requires a valid Url to make request to DO. But we can make one up!
            stub.fetch_with_str("http://fake_url.com/messages").await
        })
        .get("/secret", |_req, ctx| {
            Response::ok(ctx.secret("CF_API_TOKEN")?.to_string())
        })
        .get("/var", |_req, ctx| {
            Response::ok(ctx.var("BUILD_NUMBER")?.to_string())
        })
        .post_async("/kv", |_req, ctx| async move {
            let kv = ctx.kv("SOME_NAMESPACE")?;

            kv.put("key", "value")?.execute().await?;

            Response::empty()
        })
        .run(req, env).await
}

For more information about how to configure these bindings, see:

• https://developers.cloudflare.com/workers/cli-wrangler/configuration#keys
• https://developers.cloudflare.com/workers/learning/using-durable-objects#configuring-durable-object-bindings
• https://developers.cloudflare.com/workers/runtime-apis/bindings/version-metadata/

Durable Objects

Define a Durable Object in Rust

To define a Durable Object using the worker crate you need to implement the DurableObject trait on your own struct. Additionally, the #[durable_object] attribute macro must be applied to the struct definition.

use worker::{durable_object, DurableObject, State, Env, Result, Request, Response};

#[durable_object]
pub struct Chatroom {
    users: Vec<User>,
    messages: Vec<Message>,
    state: State,
    env: Env, // access `Env` across requests, use inside `fetch`
}

impl DurableObject for Chatroom {
    fn new(state: State, env: Env) -> Self {
        Self {
            users: vec![],
            messages: vec![],
            state: state,
            env,
        }
    }

    async fn fetch(&self, _req: Request) -> Result<Response> {
        // do some work when a worker makes a request to this DO
        Response::ok(&format!("{} active users.", self.users.len()))
    }
}

You’ll need to “migrate” your worker script when it’s published so that it is aware of this new Durable Object, and include a binding in your wrangler.toml.

• Include the Durable Object binding type in you wrangler.toml file:

# ...

[durable_objects]
bindings = [
  { name = "CHATROOM", class_name = "Chatroom" } # the `class_name` uses the Rust struct identifier name
]

[[migrations]]
tag = "v1" # Should be unique for each entry
new_classes = ["Chatroom"] # Array of new classes

SQLite Storage in Durable Objects

Durable Objects can use SQLite for persistent storage, providing a relational database interface. To enable SQLite storage, you need to use new_sqlite_classes in your migration and access the SQL storage through state.storage().sql().

use worker::{durable_object, DurableObject, State, Env, Result, Request, Response, SqlStorage};

#[durable_object]
pub struct SqlCounter {
    sql: SqlStorage,
}

impl DurableObject for SqlCounter {
    fn new(state: State, _env: Env) -> Self {
        let sql = state.storage().sql();
        // Create table if it does not exist
        sql.exec("CREATE TABLE IF NOT EXISTS counter(value INTEGER);", None)
            .expect("create table");
        Self { sql }
    }

    async fn fetch(&self, _req: Request) -> Result<Response> {
        #[derive(serde::Deserialize)]
        struct Row {
            value: i32,
        }

        // Read current value
        let rows: Vec<Row> = self
            .sql
            .exec("SELECT value FROM counter LIMIT 1;", None)?
            .to_array()?;
        let current = rows.get(0).map(|r| r.value).unwrap_or(0);
        let next = current + 1;

        // Update counter
        self.sql.exec("DELETE FROM counter;", None)?;
        self.sql
            .exec("INSERT INTO counter(value) VALUES (?);", vec![next.into()])?;

        Response::ok(format!("SQL counter is now {}", next))
    }
}

Configure your wrangler.toml to enable SQLite storage:

# ...

[durable_objects]
bindings = [
  { name = "SQL_COUNTER", class_name = "SqlCounter" }
]

[[migrations]]
tag = "v1" # Should be unique for each entry
new_sqlite_classes = ["SqlCounter"] # Use new_sqlite_classes for SQLite-enabled objects

• For more information about migrating your Durable Object as it changes, see the docs here: https://developers.cloudflare.com/workers/learning/using-durable-objects#durable-object-migrations-in-wranglertoml

Queues

Enabling queues

As queues are in beta you need to enable the queue feature flag.

Enable it by adding it to the worker dependency in your Cargo.toml:

worker = {version = "...", features = ["queue"]}

Example worker consuming and producing messages:

use worker::*;
use serde::{Deserialize, Serialize};
#[derive(Serialize, Debug, Clone, Deserialize)]
pub struct MyType {
    foo: String,
    bar: u32,
}

// Consume messages from a queue
#[event(queue)]
pub async fn main(message_batch: MessageBatch<MyType>, env: Env, _ctx: Context) -> Result<()> {
    // Get a queue with the binding 'my_queue'
    let my_queue = env.queue("my_queue")?;

    // Deserialize the message batch
    let messages = message_batch.messages()?;

    // Loop through the messages
    for message in messages {
        // Log the message and meta data
        console_log!(
            "Got message {:?}, with id {} and timestamp: {}",
            message.body(),
            message.id(),
            message.timestamp().to_string()
        );

        // Send the message body to the other queue
        my_queue.send(message.body()).await?;

        // Ack individual message
        message.ack();

        // Retry individual message
        message.retry();
    }

    // Retry all messages
    message_batch.retry_all();
    // Ack all messages
    message_batch.ack_all();
    Ok(())
}

You’ll need to ensure you have the correct bindings in your wrangler.toml:

# ...
[[queues.consumers]]
queue = "myqueueotherqueue"
max_batch_size = 10
max_batch_timeout = 30


[[queues.producers]]
queue = "myqueue"
binding = "my_queue"

RPC Support

workers-rs has experimental support for Workers RPC. For now, this relies on JavaScript bindings and may require some manual usage of wasm-bindgen.

Not all features of RPC are supported yet (or have not been tested), including:

• Function arguments and return values
• Class instances
• Stub forwarding

RPC Server

Writing an RPC server with workers-rs is relatively simple. Simply export methods using wasm-bindgen. These will be automatically detected by worker-build and made available to other Workers. See example.

RPC Client

Creating types and bindings for invoking another Worker’s RPC methods is a bit more involved. You will need to write more complex wasm-bindgen bindings and some boilerplate to make interacting with the RPC methods more idiomatic. See example.

With manually written bindings, it should be possible to support non-primitive argument and return types, using serde-wasm-bindgen.

Generating Client Bindings

There are many routes that can be taken to describe RPC interfaces. Under the hood, Workers RPC uses Cap’N Proto. A possible future direction is for Wasm guests to include Cap’N Proto serde support and speak directly to the RPC protocol, bypassing JavaScript. This would likely involve defining the RPC interface in Cap’N Proto schema and generating Rust code from that.

Another popular interface schema in the WebAssembly community is WIT. This is a lightweight format designed for the WebAssembly Component model. workers-rs includes an experimental code generator which allows you to describe your RPC interface using WIT and generate JavaScript bindings as shown in the rpc-client example. The easiest way to use this code generator is using a build script as shown in the example. This code generator is pre-alpha, with no support guarantee, and implemented only for primitive types at this time.

CPU Limits

Rust Workers have CPU limits assigned by the platform. To gracefully detect these limits we provide a worker::signals API.

This allows for graceful backoff by detecting when the Worker is near its CPU limit and will be terminated using:

use worker::signals;

pub fn do_work () {
    while !signals::is_near_cpu_limit() {
        // hot loop
    }
}

See Signal Example for a full end-to-end workflow.

Emscripten

The experimental worker-build --emscripten flag can be used to build for wasm32-unknown-emscripten, giving Workers a libc and epoll-backed sockets.

Just update the build command to add this flag, and the Emscripten toolchain will be installed automatically:

main = "build/index.js"

[build]
command = "cargo install -q worker-build && worker-build --emscripten --release"

The build links a bin target rather than a cdylib, so the package must have a bin target (e.g. src/main.rs with an empty fn main() {}).

See the emscripten, emscripten-tokio and emscripten-tcp examples, which build up from std::fs on the in-memory filesystem, through Tokio timers, tasks and channels, to raw TCP.

Panic Recovery with --panic-unwind

By default, Rust panics in Workers compile with panic=abort, which terminates the WebAssembly instance. The --panic-unwind flag for worker-build changes this behavior so that panics are caught and converted to JavaScript exceptions, allowing the Worker to continue serving requests after a panic.

When running worker-build directly:

worker-build --panic-unwind

Example wrangler.toml build command:

[build]
command = "cargo install worker-build && worker-build --release --panic-unwind"

This flag:

• Uses the nightly Rust toolchain (installed automatically if missing)
• Rebuilds std with -Zbuild-std=std,panic_unwind and -Cpanic=unwind (the rust-src component and wasm32-unknown-unknown target for nightly are installed automatically if missing)
• Enables wasm-bindgen’s panic catching support, which catches panics at FFI boundaries and converts them to JavaScript PanicError exceptions
• Registers schedule_reinit() via wasm-bindgen’s abort handling hooks to automatically recover from critical errors (e.g. unreachable, stack overflow, or out-of-memory). After a hard abort the WebAssembly instance is transparently reinitialized on the next request, with an internal instance ID bump so that Durable Object instances are recreated.

Without this flag, any panic will terminate the isolate. With it, individual requests that trigger a panic will fail with an error response while subsequent requests continue to work normally.

Unwind Safety

When building with panic=unwind, exported function arguments and closure captures must satisfy Rust’s UnwindSafe trait. The worker crate macros wrap handler callbacks with AssertUnwindSafe automatically, but if you pass closures to JavaScript via Closure::new or similar APIs you may need to wrap non-unwind-safe captures (e.g. &mut T, Cell<T>, RefCell<T>) in std::panic::AssertUnwindSafe:

use std::cell::Cell;
use std::panic::AssertUnwindSafe;
use wasm_bindgen::prelude::*;

let counter = Cell::new(0u32);
let counter_ref = AssertUnwindSafe(&counter);
let closure = Closure::new(move || {
    counter_ref.set(counter_ref.get() + 1);
});

Alternatively, Closure::own_aborting and other *_aborting variants skip the UnwindSafe requirement but will abort on panic instead of catching it. See the wasm-bindgen closures documentation for the full set of closure APIs and their unwind behavior.

Testing with Miniflare

In order to test your Rust worker locally, the best approach is to use Miniflare. However, because Miniflare is a Node package, you will need to write your end-to-end tests in JavaScript or TypeScript in your project. The official documentation for writing tests using Miniflare is available here. This documentation being focused on JavaScript / TypeScript codebase, you will need to configure as follows to make it work with your Rust-based, WASM-generated worker:

Step 1: Add Wrangler and Miniflare to your devDependencies

npm install --save-dev wrangler miniflare

Step 2: Build your worker before running the tests

Make sure that your worker is built before running your tests by calling the following in your build chain:

wrangler deploy --dry-run

By default, this should build your worker under the ./build/ directory at the root of your project.

Step 3: Configure your Miniflare instance in your JavaScript / TypeScript tests

To instantiate the Miniflare testing instance, configure an explicit module manifest containing the generated JavaScript and WebAssembly modules:

// test.mjs
import assert from "node:assert";
import { readFileSync } from "node:fs";
import { Miniflare } from "miniflare";

const mf = new Miniflare({
  workers: [{
    config: {
      name: "test",
      type: "worker",
      compatibilityDate: "2026-08-21",
      manifest: {
        mainModule: "build/index.js",
        modules: {
          "build/index.js": {
            type: "esm",
            contents: readFileSync("./build/index.js", "utf8"),
          },
          "build/index_bg.wasm": {
            type: "wasm",
            contents: new Uint8Array(readFileSync("./build/index_bg.wasm")),
          },
        },
      },
    },
  }],
});

const res = await mf.dispatchFetch("http://localhost");
assert(res.ok);
assert.strictEqual(await res.text(), "Hello, World!");

D1 Databases

Enabling D1 databases

As D1 databases are in alpha, you’ll need to enable the d1 feature on the worker crate.

worker = { version = "x.y.z", features = ["d1"] }

Example usage

use worker::*;

#[derive(Deserialize)]
struct Thing {
	thing_id: String,
	desc: String,
	num: u32,
}

#[event(fetch, respond_with_errors)]
pub async fn main(request: Request, env: Env, _ctx: Context) -> Result<Response> {
	Router::new()
		.get_async("/:id", |_, ctx| async move {
			let id = ctx.param("id").unwrap()?;
			let d1 = ctx.env.d1("things-db")?;
			let statement = d1.prepare("SELECT * FROM things WHERE thing_id = ?1");
			let query = statement.bind(&[id])?;
			let result = query.first::<Thing>(None).await?;
			match result {
				Some(thing) => Response::from_json(&thing),
				None => Response::error("Not found", 404),
			}
		})
		.run(request, env)
		.await
}

Notes and FAQ

It is exciting to see how much is possible with a framework like this, by expanding the options developers have when building on top of the Workers platform. However, there is still much to be done. Expect a few rough edges, some unimplemented APIs, and maybe a bug or two here and there. It’s worth calling out here that some things that may have worked in your Rust code might not work here - it’s all WebAssembly at the end of the day, and if your code or third-party libraries don’t target wasm32-unknown-unknown, they can’t be used on Workers. Additionally, you’ve got to leave your threaded async runtimes at home; meaning no Tokio or async_std support. However, async/await syntax is still available and supported out of the box when you use the worker crate.

We fully intend to support this crate and continue to build out its missing features, but your help and feedback is a must. We don’t like to build in a vacuum, and we’re in an incredibly fortunate position to have brilliant customers like you who can help steer us towards an even better product.

So give it a try, leave some feedback, and star the repo to encourage us to dedicate more time and resources to this kind of project.

If this is interesting to you and you want to help out, we’d be happy to get outside contributors started. We know there are improvements to be made such as implementing additional APIs, utility crates, and more. In fact, we’re always on the lookout for great engineers, and hiring for many open roles - please take a look.

FAQ

1. Can I deploy a Worker that uses tokio or async_std runtimes?

• Currently no. All crates in your Worker project must compile to wasm32-unknown-unknown target, which is more limited in some ways than targets for x86 and ARM64. However, you should still be able to use runtime-agnostic primitives from those crates such as those from tokio::sync.

2. The worker crate doesn’t have X! Why not?

• Most likely, it should, we just haven’t had the time to fully implement it or add a library to wrap the FFI. Please let us know you need a feature by opening an issue.

3. My bundle size exceeds Workers size limits, what do I do?

• We’re working on solutions here, but in the meantime you’ll need to minimize the number of crates your code depends on, or strip as much from the .wasm binary as possible. Here are some extra steps you can try: https://rustwasm.github.io/book/reference/code-size.html#optimizing-builds-for-code-size. Release builds keep Wasm function names for profiles and stack traces; to drop them, set the profile’s wasm-opt arguments without -g (see the worker-build README) rather than using strip = true.

Releasing

1. Trigger a workflow to create a release PR.
2. Review version changes and merge PR.
3. A draft GitHub release will be created. Author release notes and publish when ready.
4. Crates (worker-sys, worker-macros, worker) will be published automatically.

Contributing

Your feedback is welcome and appreciated! Please use the issue tracker to talk about potential implementations or make feature requests. If you’re interested in making a PR, we suggest opening up an issue to talk about the change you’d like to make as early as possible.

Project Contents

• worker: the user-facing crate, with Rust-familiar abstractions over the Rust<->JS/WebAssembly interop via wrappers and convenience library over the FFI bindings.
• worker-sys: Rust extern “C” definitions for FFI compatibility with the Workers JS Runtime.
• worker-macros: exports event and durable_object macros for wrapping Rust entry point in a fetch method of an ES Module, and code generation to create and interact with Durable Objects.
• worker-sandbox: a functioning Cloudflare Worker for testing features and ergonomics.
• worker-build: a cross-platform build command for workers-rs-based projects.