[go: nahoru, domu]

😱 Status quo stories: Barbara wishes for an easy runtime switch

🚧 Warning: Draft status 🚧

This is a draft "status quo" story submitted as part of the brainstorming period. It is derived from real-life experiences of actual Rust users and is meant to reflect some of the challenges that Async Rust programmers face today.

The story

Barbara has been working on an async codebase for the past 5 years. It is extremely mature and quite large (in the millions of lines of code). They've been relying on tokio as their async runtime and the codebase makes heavy use of its rich API. It has served them well over the years and they're very happy with it.

Barbara knows about async-std but has never used it. She has wondered for a while how her application would work and perform if she had used async-std instead. She decides to test it out by porting her projects from tokio to async-std.

To their disappointment, they discover many areas, where their choice of runtime permeates the code base:

  • tokio provides variants of helpers macros and types, like tokio::select! and tokio::Mutex. These helpers can be used without the rest of tokio, and there are also alternatives from the futures crate and elsewhere (albeit with subtle differences).
  • tokio uses a custom version of AsyncRead and AsyncWrite traits which differ from the ones used by other parts of the ecosystem.
  • The tokio API is needed to create core runtime operations like timers (tokio::time::sleep) and to launch tasks; there doesn't seem to be a standard way to abstract over those kinds of things in a runtime-independent way.
  • Some of their dependencies (e.g hyper and reqwest) are tied to tokio. In some cases, there are configuration options or ways to use those dependencies that don't depend on tokio, but there is no standard mechanism for that.

These things aren't specific to tokio. There just doesn't seem to be a lot of consensus in the ecosystem on how to write "runtime-independent" code and in some cases there aren't any great options available (e.g., spawning tasks).

They investigate the possibility of providing some sort of compatibility layer between tokio and their new runtime of choice but this turns out to not seem like the right way to go as this compatibility layer would require too much overhead.

Realizing that the task of porting the entire code base to async-std, will take a lot of effort and time, Barbara decides to give up. She is very disappointed.

🤔 Frequently Asked Questions

What are the morals of the story?

  • Using a certain executor often means using a certain run-time ecosystem. This often locks the user into that ecosystem.
  • Tying yourself to a certain executor means that you are tied to the priorities of that executor. You may be happy with the run-time ecosystem, but have special needs that the default executor does not provide. If the executor doesn't have an extensibility model, you're stuck. Note: It is perfectly reasonable for a general purpose executor to not be able or willing to cater for specialized needs.
  • All of this is made worse by that fact that run-time agnostic libraries are difficult and sometimes even impossible to write.

What are the sources for this story?

This story is more of a thought experiment than a recounting of a true story. We just asked logically what would happen if a team working on code base where it was assumed they could use a specific runtime decides to use a different runtime.

Why did you choose Barbara to tell this story?

The story assumes a Rust programmer that has worked for several years on a large and complex Rust codebase, so Barbara is the natural choice here.

How would this story have played out differently for the other characters?

It wouldn't. If this story happens them, they're on the same level of Rust expertise as Barbara is.