3 unstable releases
0.2.0 | Nov 20, 2023 |
---|---|
0.1.1 | Jul 18, 2023 |
0.1.0 | Jul 17, 2023 |
#1017 in Network programming
30 downloads per month
Used in tower-async-http
185KB
2.5K
SLoC
Tower Async
Tower Async is a library of modular and reusable components for building robust networking clients and servers.
Fork
Tower Async is a fork of https://github.com/tower-rs/tower
and makes use of async traits
to simplify things and make it more easier
to integrate async functions into middleware.
If you want to see a prime example of how much simpler using
tower_async
is versustower
, you can see an example here:A delay service using
tower
: https://github.com/plabayo/tower-async/blob/4ae0c4747fac6cc69b27c87a7ea5cacdd8bab3fb/tower-async-bridge/src/into_async/async_layer.rs#L91-L169.That same service can be written in
tower_async
as follows:#[derive(Debug)] struct DelayService<S> { inner: S, delay: std::time::Duration, } impl<S> DelayService<S> { fn new(inner: S, delay: std::time::Duration) -> Self { Self { inner, delay } } } impl<S, Request> tower_async_service::Service<Request> for DelayService<S> where S: tower_async_service::Service<Request>, { type Response = S::Response; type Error = S::Error; async fn call(&self, request: Request) -> Result<Self::Response, Self::Error> { tokio::time::sleep(self.delay).await; self.inner.call(request) } }
If you compare that with the linked
tower
version you can probably agree that things are a lot simpler if you do not have to hand write future state machines yourself, which is the reason why we have theasync
sugar in the first place.Of course I do acknowledge that if you make use of amazing utilities provided by crates such as https://docs.rs/futures-util-preview/latest/futures_util/future/index.html That you an write pretty much the same code without having to handwrite the future yourself.
This is however not always possible, and it does mean that you need to (1) know about this and (2) pull it as a dependency and all that it brings with it. While in reallity you really just want to be able write your middleware in an
async
manner.We fully acknowledge that
tower
had to use the approach it used as for many years this was simply the only sensible thing to do, unless you want to force your users to make use of the https://docs.rs/async-trait/latest/async_trait/, a choice not everyone is willing to make, and sometimes they might not even have the luxary to do so.
Come join us at discord on the #tower-async
public channel at Discord
or tag @glendc
at Tokio's Tower discord instead.
Where suitable we'll keep in sync (manually) with Tower and if the opportunity arises we'll contribute back "upstream" as well. Given however how big the diversange we aren't sure how likely that is.
This set of libraries is best suited in an ecosystem of its own,
that is to say, making use only of tower-async
libraries and dependents on it.
At the very least it is desired that tower-async
is the puppeteer with where needed
making use of tower
(classic) (middleware) layers.
For an example on how to operate purely within a tower-async
environment you can
explore the Rama codebase, a proxy framework,
written purely with a tower-async
mindset, and the main motivator to start this fork.
You can however also bridge tower
and tower-async
in any other way. Please consult
the "Bridging to Tokio's official Tower Ecosystem" chapter
for more information on how to do that.
Overview
Tower Async aims to make it as easy as possible to build robust networking clients and servers. It is protocol agnostic, but is designed around a request / response pattern. If your protocol is entirely stream based, Tower may not be a good fit.
Tower Async provides a simple core abstraction, the Service
trait, which
represents an asynchronous function taking a request and returning either a
response or an error. This abstraction can be used to model both clients and
servers.
Generic components, like timeouts, can be modeled as Service
s
that wrap some inner service and apply
additional behavior before or after the inner service is called. This allows
implementing these components in a protocol-agnostic, composable way. Typically,
such services are referred to as middleware.
An additional abstraction, the Layer
trait, is used to compose
middleware with Service
s. If a Service
can be thought of as an
asynchronous function from a request type to a response type, a Layer
is
a function taking a Service
of one type and returning a Service
of a
different type. The ServiceBuilder
type is used to add middleware to a
service by composing it with multiple Layer
s.
Difference with Tokio's official Tower Ecosystem?
- Make use of
Async Traits
(RFC-3185: Static async fn in traits) instead of requiring the user to manually implement Futures;- Which in fact forces users to Box Services that rely on futures which cannot be named,
e.g. those returned by
async functions
that the user might have to face by using common utility functions from the wider Tokio ecosystem;
- Which in fact forces users to Box Services that rely on futures which cannot be named,
e.g. those returned by
- Drop the notion of
poll_ready
(See the FAQ). - Use
&self
forService::call
instead of&mut self
:- this to simplify its usage;
- makes it clear that the user is responsible for proper state sharing;
- makes it more compatible with the ecosystem (e.g.
hyper
(v1) also takes services by&self
);
Bridging to Tokio's official Tower Ecosystem
You can make use of the tower-async-bridge
crate as found in this repo in the ./tower-async-bridge directory,
and published at crates.io under the same name.
At a high level it allows you to:
- Turn a
tower::Service
into atower_async::Service
(requires theinto_async
feature); - Turn a
tower_async::Service
into atower::Service
; - Use a
tower_async::Layer
within atower
environment (e.g.tower::ServiceBuilder
); - Use a
tower::Layer
within atower_async
environment (e.g.tower_async::ServiceBuilder
) (requires theinto_async
feature);
Please check the crate's unit tests and examples to see specifically how to use the crate in order to achieve this.
Furthermore we also urge you to only use this kind of approach for transition purposes and not as a permanent way of life.
Best in our opinion is to use one or the other and not to combine the two. But if you do absolutely must
use one combined with the other, tower-async-bridge
should allow you to do exactly that.
The Tower Async Ecosystem
Tower Async is made up of the following crates:
tower-async
(this crate)tower-async-bridge
tower-async-service
tower-async-layer
tower-async-test
tower-async-http
Since the Service
and Layer
traits are important integration points
for all libraries using Tower, they are kept as stable as possible, and
breaking changes are made rarely. Therefore, they are defined in separate
crates, tower-async-service
and tower-async-layer
. This crate contains
re-exports of those core traits, implementations of commonly-used
middleware, and utilities for working with Service
s and Layer
s.
tower-async-bridge
is there to bridge Tokio's official Tower ecosystem
with this (Aync Trait) version (Fork).
Testing Layer
s can be done with unit tests very easily suing tower-async-test
.
Finally in case you are using tower-async
for HTTP purposes (e.g. an HTTP web server),
then you might find it useful to also make use of tower-async-http
as it provides you
with builder extensions and middleware specifically tailored for http purposes.
Usage
Tower (Async) provides an abstraction layer, and generic implementations of various
middleware. This means that the tower-async
crate on its own does not provide
a working implementation of a network client or server. Instead, Tower's
Service
trait provides an integration point between
application code, libraries providing middleware implementations, and
libraries that implement servers and/or clients for various network
protocols.
Depending on your particular use case, you might use Tower in several ways:
-
Implementing application logic for a networked program. You might use the
Service
trait to model your application's behavior, and use the middleware provided by this crate and by other libraries to add functionality to clients and servers provided by one or more protocol implementations. -
Implementing middleware to add custom behavior to network clients and servers in a reusable manner. This might be general-purpose middleware (and if it is, please consider releasing your middleware as a library for other Tower users!) or application-specific behavior that needs to be shared between multiple clients or servers.
-
Implementing a network protocol. Libraries that implement network protocols (such as HTTP) can depend on
tower-async-service
to use theService
trait as an integration point between the protocol and user code. For example, a client for some protocol might implementService
, allowing users to add arbitrary Tower middleware to those clients. Similarly, a server might be created from a user-providedService
.Additionally, when a network protocol requires functionality already provided by existing Tower middleware, a protocol implementation might use Tower middleware internally, as well as an integration point.
Library Support
Following are some libraries that make use of Tower Async (instead of Tower)
and the Service
trait:
rama
: A proxy framework to anonymise your network traffic.
If you're the maintainer of a crate that supports Tower Async, we'd love to add your crate to this list! Please open a PR adding a brief description of your library!
Getting Started
The various middleware implementations provided by this crate are feature flagged, so that users can only compile the parts of Tower they need. By default, all the optional middleware are disabled.
To get started using all of Tower's optional middleware, add this to your
Cargo.toml
:
tower-async = { version = "0.2", features = ["full"] }
Alternatively, you can only enable some features. For example, to enable
only the timeout
middleware, write:
tower-async = { version = "0.2", features = ["timeout"] }
See here for a complete list of all middleware provided by Tower.
Browse the examples at tower-async-http/examples
to see some examples
on how to use tower-async
and its sibling crates. While these are focussed on http examples,
note that:
tower-async
can work for any request-response flow (akin totower
);- you can also use
tower-async
with http web services without making use of thetower-async-http
crate, it only is there to provide extra middleware for http-specific purposes, but this is all optional.
The documentation also contains some smaller examples and of course the codebase can be read as well, together with its unit tests.
Supported Rust Versions
Tower Async requires nightly Rust for the time being and has no backwards compatibility promises for the time being.
Once async traits
are stabilized we'll start supporting stable rust once again,
and we can start working towards backwards compatibility.
Read https://blog.rust-lang.org/inside-rust/2023/05/03/stabilizing-async-fn-in-trait.html for more information on this roadmap by the Rust Language Core Team.
Sponsorship
Regular and onetime sponsors alike help us to pay the development and service costs done in function of all Plabayo's Free and Open Source work.
We're also a monthly sponsor of Tokio ourselves, to give back to all the great work done and continued effort being put in by them.
You can find more about Plabayo Sponsorship at https://github.com/sponsors/plabayo.
One time sponsorships (the so called "buy me a coffee", but then via GitHub Sponsors payments), are welcome as much as regular sponsors. Not everybody have the financial means to sponsor, so feel free to contribute in any other way that you can think of.
FAQ
Where is the
poll_ready
method from Tower's Service?
This has been removed for the purpose of simplification and because the authors of this fork consider it a problem out of scope:
- most Tower services / layers do not ever need the
poll_ready
method, and simply call the inner service for that; - for some backpressure purposes you do want to know the request to know how to handle it, so
poll_ready
wouldn't work for these;
poll_ready
was also used for load balancing services but this is considered out of scope:
- load balancing incoming network streams is in our humble opinion more something to be handled by your network infrastructure surrounding your service (using a... load balancer);
- and again... if you do want to load balance within a service it might be because you
actually require context from the request to know what to do, in which case
poll_ready
wouldn't work for you;
Where you do still want to apply some kind of rate limiting, back pressure or load balancing
within a Tower (Async) Service you are to do it within the call
function instead.
This fork is however still in its early days, so feel free to start a discussion if you feel different about this topic. The authors of this library are always open for feedback but retain the reservation to deny any request they wish.
Where is my favourite Tower Utility?
As all the tower code has to be manually ported, there might be some features missing. The tower ecosystem also continues to thrive and live happy, so there might still be new features added there as well. Feel free to chat with us or open a ticket on GitHub in case you wish to add/port such feature(s).
Note that some features are not supported on purpose:
- all the 'ready' related functionality was removed on purpose as we believe it to be out of scope
- as such also all utilities that rely on this or build on top of this aren't supported
See the previous FAQ point to get our point of view related to load balancing and the like.
License
This project is licensed under the MIT license.
Big thanks and credits go towards the original Tower authors which licensed their code under the same License type.
Contribution
Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in Tower Async by you, shall be licensed as MIT, without any additional terms or conditions.
Dependencies
~0–5.5MB
~25K SLoC