[weiming]

Going down the same rabbit hole earlier this week, found this to be a good explanation:

All of the data needed by a task is contained within its future. That means we can neatly sidestep problems of dynamic stack growth and stack swapping, giving us truly lightweight tasks without any runtime system implications. ... Perhaps surprisingly, the future within a task compiles down to a state machine, so that every time the task wakes up to continue polling, it continues execution from the current state—working just like hand-rolled code. [1]

[1] https://aturon.github.io/blog/2016/09/07/futures-design/

[davidw]

> Perhaps surprisingly, the future within a task compiles down to a state machine, so that every time the task wakes up to continue polling, it continues execution from the current state

How are those tasks implemented, and what's scheduling them?

[weiming]

You need a third-party library to provide an executor. Rust does not come with one to keep the runtime size small. The community seems to have centralized on https://tokio.rs/ (under the hood it uses epoll/whatever OS-specific functionality to schedule M:N)

See: https://news.ycombinator.com/item?id=20722297

[davidw]

How's that play out in terms of, say, performing some long-running calculations, rather than something that performs IO?

[weiming]

(Of course, this is Tokio-specific)

Cooperative scheduling is used to schedule tasks on executors. A single executor is expected to manage many tasks across a small set of threads. There will be a far greater number of tasks than threads. There also is no pre-emption. This means that when a task is scheduled to execute, it blocks the current thread until the poll function returns.

Because of this, it is important for implementations of poll to only execute for very short periods of time. For I/O bound applications, this usually happens automatically. However, if a task must run a longer computation, it should defer work to a blocking pool [1] or break up the computation into smaller chunks and yield back to the executor after each chunk. [2]

"poll" here refers to the callback function in the future that actually does the work.

[1] https://docs.rs/tokio-threadpool/0.1.15/tokio_threadpool/fn....

[2] https://tokio.rs/docs/internals/runtime-model/

[oconnor663]

When you want to do long-running blocking work (either computation, or synchronous IO from some library that doesn't use futures), you usually want to farm that work out to a thread pool. I think Tokio provides convenience functions for doing that.

It's also possible to write a long-running computation as a future, which can yield after some smaller amount of work to let some IO get done. But I'm not sure if that's the recommended approach.

[ori_b]

You're essentially implementing statically scheduled cooperative userspace threads on top of a single thread with async/await.

So, for computation, this is a bad solution.

[tmandry]

This is a great quote, and one that I missed while first writing the post! I've added it now.