Re: aio_read2() and aio_write2()

On Sun, Jan 14, 2024 at 10:30 AM Vinícius dos Santos Oliveira
<vini.ipsmaker@gmail.com> wrote:
>
> Em dom., 14 de jan. de 2024 às 14:13, Alan Somers
> <asomers@freebsd.org> escreveu:
> > The problem is that this flag would be almost impossible to use
> > correctly for the intended use cases of POSIX AIO.  Your application
> > is actually pretty unusual in that it only has one operation in-flight
> > at a time.  I think it would be better to use the lseek solution
> > rather than add a footgun to POSIX AIO.
>
> There are two things here: Boost.Asio, and applications built on top
> of Boost.Asio. I can't speak for other applications built on top of
> Boost.Asio. With that in mind, let me proceed.
>
> There is nothing unusual about my application. I just built an
> execution context that implements the green threading model. That's
> nothing unusual about that. NodeJS, Python's asyncio, Rust's asyncio,
> Golang, luvit, and so on and so on. All these frameworks just mimic
> interfaces from the blocking world (e.g. two threads doing a blocking
> read() on two different files become two fibers doing a read() within
> the same thread). If anything, the whole world is moving to this model
> after NodeJS proved its usefulness. So what's unusual about that?

I think you're using the term "green threading" too broadly.  Golang
uses green threads, but Rust does not.  The difference is that in Rust
with async/await, the task-switching boundaries are explicit (the
.await syntax).  So Rust uses explicit concurrency, not green
threading.  I can't speak to the other languages you mention.

>
> To clarify: I don't have a single in-flight operation at a time. The
> same thread dispatches IO requests for different streams (sockets and
> files). AIO is not only useful for batching operations (on the same
> file). AIO is also useful for a batch of IO operations on different
> files. File IO is always “ready” and the readiness model doesn't work
> for files. On Linux, I can just use io_uring (proactor/completion
> model) for everything (as in it won't prevent me from skipping an
> explicit offset). Same with Windows (IOCP). What is special about
> FreeBSD here?
>
> POSIX AIO by itself is useless to me. It's only useful to me with BSD
> extensions (SIGEV_KEVENT for kqueue integration). I don't see a reason
> why it can't have another small extension that is pretty much
> non-invasive.

You propose an extension that would essentially create asynchronous
(and racy) versions of read, write, readv, and writev .  But what
about copy_file_range and fspacectl?  Or for that matter all the
dozens of control-path syscalls like open, stat, chmod, and truncate?
This flag that you propose is not a panacea that will eliminate all
blocking file operations.  There will still be a need for things that
block.  Rust (via the Tokio library) still uses a thread pool for such
things.  It even uses the thread pool for the equivalent of read() and
write() (but not pread and pwrite).

My point is that if you want fully asynchronous file I/O that never
blocks you can't achieve that by adding one additional flag to POSIX
AIO.  That would be a involved project and it needs a serious design
effort.  Even then, I suspect that the best solution would be to
completely eschew the seek pointer.  Instead, all operations would
either specify the offset (as with pwrite, pread) or operate only at
EoF as if O_APPEND were used.

-Alan