High-precision timers

[polachok]

Version

0.1.16

Platform

Linux e8ab1d1f59b1 4.9.125-linuxkit #1 SMP Fri Sep 7 08:20:28 UTC 2018 x86_64 x86_64 x86_64 GNU/Linux

Subcrates

tokio-timer

Description

tokio-timer supports delays with 1ms resolution, while my use case requires higher resolution timers.
On Linux, timerfd provides access to sub-millisecond timers as file descriptors. I implemented a simple periodic timer in tokio-timerfd, which works fine, but I would also like to have APIs like DelayQueue. I'm asking for advice on how to best proceed with this. I'd like to avoid duplicating tokio-timer code in my crate, but it seems like there's no way to plug in my own timer implementation.

[carllerche]

Good question. Do you have any thoughts on how to proceed?

[polachok]

I played with one approach in this PR: https://github.com/polachok/tokio/pull/1/files.
It adds TimerImpl trait which can be implemented by timers and injected into runtime.

[polachok]

@carllerche any comments?

[tobz]

@polachok Is there anyway you could submit your PR on this repo?

It'll give others a better chance to review and contribute.

[polachok]

@tobz it's more a proof-of-concept than a proper PR, but if you think it's approriate, I will

[DoumanAsh]

I've been playing around with generic timer implementation here, if you're interested

But it targets std's futures, so there is only romio integration for timerfd and kqueue.
Not sure how easy it would be integrate with tokio as it needs Waker

[carllerche]

It would probably be fine to provide a highres-timer option in Tokio... or update tokio_timer::Delay to have an option to pick high res.

[carllerche]

That said, I'm going to close due to inactivity.

Any proposal that would be considered for inclusion in Tokio would need to:

• Support linux, *bsd, and windows platforms.
• Include an API proposal
• Include an argument for why it is useful to have a timer with greater than a 1ms resolution in an asynchronous context.

I looked quickly at windows and I do not see an obvious option there. However, if someone wants to take a stab at this, I would suggest opening a new issue w/ the proposal and cross reference this issue.